JP2010019106A - Control device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a control device with improved robustness against noises, etc., of a dead time estimating means, further, a sequential identification means of a plant model parameter. <P>SOLUTION: The device sequentially identifies the parameter of the plant model for expressing a control object by a formula discretized at a predetermined sampling period dt, estimates surplus dead time included in the plant model by using the identified parameter, moving-forward/putting-off updates a value of the number of waste-sampling times based on the estimated surplus dead time, and estimates the dead time based on the updated number of the waste-sampling times. A calculation period t20 of an identification calculation part M30 for performing the identification (meaning the sampling period dt) is made to be longer than a calculation period t20 by a sub F/B control part M20 (a control input calculation means) for calculating a value used for calculation of control input to the control object. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a control device that uses a plant model that expresses a control target of an internal combustion engine or the like by a mathematical formula, and automatically adapts the dynamic characteristics of the control target by sequentially identifying parameters of the plant model. is there.

  Conventionally, there has been proposed a control device that sequentially identifies the dynamic characteristics of a control target. This control device has a plant model that expresses the object to be controlled by a mathematical formula, and the plant model so that the error between the plant model output when the actual control input is input to the plant model and the output of the actual control object approaches zero. In the case of a system in which the controlled object includes a dead time, the plant model used for identification needs to consider the dead time of the actual controlled object. Therefore, in Patent Documents 1 and 2, the dead time can be estimated by the following identification method.

  That is, for a plant model that represents a control target by a mathematical expression discretized at a predetermined sampling period (dt), the quotient obtained by dividing the dead time (L) of the plant model by the sampling period (d), the remainder Is the extra time (L1). Then, by using the sequentially identified parameters, the dead time included in the plant model can be estimated, and the dead time is estimated as illustrated in the time chart of FIG. 8 described below. To do.

  As shown in FIG. 8A, it is assumed that an error as shown in the figure occurs between the nominal dead time and the actual dead time for a certain control input. A portion exceeding the sampling timing (d · dt) is a surplus time L1, and the nominal dead time is a nominal value of a preset dead time.

  In this case, the identification is sequentially performed, so that the estimated value L1_hat of the remaining time (hereinafter, x_hat represents the identification value or estimated value of x) changes as shown in FIG. To increase. Then, when the estimated value L1_hat of the dead time remains for a predetermined time or more in the vicinity of dt, as shown in FIG. 8C, the estimated value d_hat of the number of dead samplings is incremented by one (d_hat ← d + 1). Further, the estimated value L1_hat of the surplus time is updated to a value near zero. Thereafter, by performing identification sequentially, as shown in FIG. 8D, the estimated value L1_hat of the dead time increases, and the estimated value of the dead time converges to the actual dead time accordingly.

  When the number of dead sampling times d_hat and the dead time L1_hat updated in this way are used, the following equation is obtained: L (dead time) = d (dead sampling number) × dt (sampling period) + L1 (dead time) Based on this, the dead time L can be estimated.

FIG. 8 illustrates the case where the estimated value L1_hat of the extra time increases, and the situation where the estimated value d_hat of the number of dead samplings is incremented by one has been explained. When the estimated time value L1_hat decreases and stays in the vicinity of zero for a predetermined time or more, the estimated value d_hat of the number of dead samplings is lowered by one and the estimated value L1_hat of the remaining time is updated to a value near dt. Is done.
JP 2006-118429 A JP 2007-285167 A

  However, if L1_hat vibrates more than the width of zero to dt in a short time due to noise or the like included in the estimated value L1_hat of the dead time, the dead sampling time d and the dead time L1 are erroneously updated, resulting in a dead time. There is a risk of deteriorating estimation accuracy.

  The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a control device that improves robustness against noise and the like of dead time estimation means and plant model parameter sequential identification means. There is to do.

  Hereinafter, means for solving the above-described problems and the operation and effects thereof will be described.

In invention of Claim 1,
A control input calculation means for calculating a control input to the control target or a value used for calculation of the control input;
An identification means for sequentially identifying the plant model parameters for the plant model that represents the control target by a mathematical expression discretized at a predetermined sampling period (dt);
The identification parameter identified by the identification means when the quotient obtained by dividing the dead time (L) of the control object by the sampling period is modeled as a dead sampling number (d) and the remainder is a dead time (L1). A dead time estimation means for estimating the dead time included in the plant model using
When the estimated dead time estimated by the dead time estimation means has changed to the vicinity of the sampling cycle, the estimated value of the number of dead samplings is updated to be incremented by one. A waste sampling number updating means for updating the estimated value of the number of waste samplings to lower by one;
A dead time estimation means for estimating the dead time based on the estimated number of dead sampling times updated by the dead sampling number update means;
With
-"The sampling cycle is longer than the calculation cycle by the control input calculation means".

  According to this, the parameters of the discretized plant model are sequentially identified (identification means), the extra time included in the plant model is estimated using the identified parameters (excess time estimation means), When the estimated dead time changes to the vicinity of the sampling period, the value of the number of times of dead sampling is updated to be incremented by one, and when it has changed to near zero, it is updated to be decremented by one (dead sampling number update means). The dead time is estimated based on the updated number of dead samplings (dead time estimation means). Therefore, it is possible to perform highly accurate control by sequentially estimating the change in the dead time.

  In addition, by “making the sampling period longer than the calculation period of the control input calculation means”, the width of zero to dt becomes sufficiently larger than the amplitude of the dead time that vibrates due to noise or the like. Therefore, it is possible to suppress erroneous updating of the number of dead samplings. Therefore, it is possible to improve the robustness against noise and the like of the dead time estimation unit and the plant model parameter sequential identification unit.

  In the invention of claim 2, since the sampling period is longer than the fluctuation range due to noise included in the estimated residual time estimated by the residual time estimation means, Similarly to the first aspect of the invention, the width of zero to dt is sufficiently larger than the amplitude of the estimated dead time that vibrates due to noise or the like. Therefore, it is possible to suppress erroneous updating of the estimated value of the number of dead samplings. Therefore, it is possible to improve the robustness against noise and the like of the dead time estimation unit and the plant model parameter sequential identification unit.

  According to a third aspect of the present invention, the identification unit includes a filter unit for suppressing a high-frequency vibration component included in the estimated residual time estimated by the residual time estimation unit, and the sampling cycle includes the sampling period, It is characterized in that it is longer than the fluctuation range due to noise included in the estimated remaining time after being filtered by the filter means.

  According to this, out of the noise included in the estimation dead time, the high-frequency vibration component that protrudes and fluctuates can be suitably removed by the filter means, and for the fluctuation range due to noise that cannot be removed by the filter means, By making the sampling period longer than the fluctuation range, it is possible to suppress erroneous updating of the estimated value of the number of dead samplings.

  Here, when the inventions according to claims 1 and 2 are implemented, the sampling period in the identification means and the calculation period by the control input calculation means are different, but if the estimated dead time is only used for updating the plant model. However, when it is also used for calculation by the control input calculation means, due to the difference in both periods, the calculation accuracy by the control input calculation means decreases, the identification accuracy by the identification means decreases, the dead time estimation accuracy decreases, etc. There are concerns about defects.

  With respect to such a concern, the invention according to claim 4 further comprises a continuation means for converting the discretized identification parameter into a continuous model parameter, and the dead time estimation means is made continuous by the continuation means. The dead time is estimated using the continuous model parameter, and the estimated dead time estimated by the dead time estimation means is converted into an estimated value of the number of dead samplings discretized at the sampling period in the identification means. 1 discretization means, and the continuous model parameter made continuous by the continuation means and the estimated dead time estimated by the dead time estimation means are discretized at the calculation cycle in the control input calculation means Second discretization means for converting into model parameters and the number of waste sampling times for control, the first discrete The estimation value of the number of waste samplings discretized by the means is used for sequential identification in the identification unit, and the control model parameters discretized by the second discretization unit and the number of waste samplings for control are It is used for the calculation by the control input calculation means.

  According to this, the discrete plant model is once continuous to estimate the dead time, and then the estimated dead time is discretized at each calculation cycle in order to be used by the identification means and the control input calculation means. Therefore, it is possible to easily eliminate the above-described concerns about the reduction in accuracy due to the difference between the two calculation cycles.

Further, as described in claim 5, the control input calculating means is suitable as a main feedback control means or a sub feedback control means described below. That is,
The control objects include an exhaust gas purifying catalyst disposed in an exhaust passage of an internal combustion engine, an upstream exhaust gas sensor for detecting a specific exhaust gas concentration upstream of the catalyst in the exhaust passage, and the catalyst in the exhaust passage. A downstream exhaust gas sensor for detecting a specific exhaust gas concentration downstream of the air-fuel ratio control system,
Main feedback control means for calculating the supply air-fuel ratio as the control input by feeding back the air-fuel ratio detected by the upstream exhaust gas sensor so as to match the upstream target air-fuel ratio of the catalyst, and the downstream exhaust gas sensor Sub-feedback control means that feeds back the detected air-fuel ratio so as to coincide with the downstream target air-fuel ratio of the catalyst, and calculates the upstream target air-fuel ratio as a value used for calculation of the control input. The input calculation means is the main feedback control means or the sub feedback control means.

  Hereinafter, an embodiment in which the present invention is applied to an air-fuel ratio control system for an internal combustion engine will be described with reference to the drawings.

  First, a schematic configuration of the entire engine control system will be described with reference to FIG. An air cleaner 13 is provided at the most upstream portion of the intake pipe 12 of the engine 11 that is an internal combustion engine, and an air flow meter 14 that detects the intake air amount is provided downstream of the air cleaner 13. A throttle valve 16 whose opening is adjusted by a motor 15 and a throttle opening sensor 17 for detecting the opening (throttle opening) of the throttle valve 16 are provided on the downstream side of the air flow meter 14.

  Further, a surge tank 18 is provided on the downstream side of the throttle valve 16, and an intake pipe pressure sensor 19 for detecting the intake pipe pressure is provided in the surge tank 18. The surge tank 18 is provided with an intake manifold 20 for introducing air into each cylinder of the engine 11, and a fuel injection valve 21 for injecting fuel is attached in the vicinity of the intake port of the intake manifold 20 of each cylinder. Yes. An ignition plug 22 is attached to the cylinder head of the engine 11 for each cylinder, and the air-fuel mixture in each cylinder is ignited by the spark discharge of each ignition plug 22.

  The cylinder block of the engine 11 includes a coolant temperature sensor 26 that detects the coolant temperature, and a crank angle sensor 28 that outputs a crank angle signal (pulse signal) every time the crankshaft 27 of the engine 11 rotates a predetermined crank angle. It is attached. Based on the crank angle signal of the crank angle sensor 28, the crank angle and the engine speed are detected.

On the other hand, the exhaust pipe 23 of the engine 11 is provided with two catalysts 25 and 30 for purifying exhaust gas in series. Each of the catalysts 25 and 30 is constituted by, for example, a three-way catalyst, a NOx occlusion type three-way catalyst, or the like. Etc.), an upstream exhaust gas sensor 31 and a downstream exhaust gas sensor 32 are provided. In the present embodiment, an air-fuel ratio sensor is used as the upstream side exhaust gas sensor 31, and an oxygen sensor (O ₂ sensor) is used as the downstream side exhaust gas sensor 32, but it goes without saying that the present invention is not limited to this configuration.

  Outputs of these various sensors are input to an engine control circuit (hereinafter referred to as “ECU”) 29. The ECU 29 is mainly composed of a microcomputer, and executes various engine control programs stored in a built-in ROM (storage medium) to thereby determine the fuel injection amount of the fuel injection valve 21 according to the engine operating state. The ignition timing of the spark plug 22 is controlled.

  Further, as shown in FIG. 2, the ECU 29 feedback-corrects the supply air-fuel ratio (fuel injection amount) so that the air-fuel ratio detected by the upstream exhaust gas sensor 31 matches the target air-fuel ratio upstream of the upstream catalyst 25. The detection voltage (detected air-fuel ratio) of the main F / B controller M11 that performs main feedback control (in the following description, “feedback” is expressed as “F / B”) and the downstream exhaust gas sensor 32 is set to the target voltage (upstream side). It functions as a sub F / B controller M21 that performs sub F / B control for F / B correcting the target air-fuel ratio upstream of the upstream catalyst 25 so as to coincide with the target air-fuel ratio downstream of the catalyst 25).

  The main F / B controller M11 is designed based on a plant model (first delay system + dead time model) that has been modeled and adapted in advance, and the main F / B controller M11 uses an upstream side exhaust gas sensor. The air-fuel ratio correction coefficient is calculated so as to reduce the deviation between the detected air-fuel ratio of 31 and the target air-fuel ratio upstream of the upstream catalyst 25, and optimal air-fuel ratio F / B control is realized.

  On the other hand, the sub F / B controller M21 sets the upstream side of the upstream catalyst 25 so that the detected voltage (detected air / fuel ratio) of the downstream side exhaust gas sensor 32 matches the target voltage (target air / fuel ratio downstream of the upstream side catalyst 25). It is designed on the basis of a plant model (secondary delay system + dead time model) that is modeled and adapted in advance so as to perform sub F / B control for F / B correction of the target air-fuel ratio.

  However, a sub F / B control error occurs due to individual differences or deterioration of the actual control target (upstream catalyst 25, downstream exhaust gas sensor 32) of the sub F / B controller M21. Therefore, in this embodiment, a control method called adaptive control is used, and the F / B gain in the sub F / B controller M21 is automatically adapted to the current dynamic characteristics of the controlled object (plant), so that the control system The performance is always kept in the best condition. That is, using a plant model in which a controlled object is represented by a mathematical model, the plant model output and the controlled object output (downstream) when the controlled object input (inflow excess / deficient oxygen amount of the upstream catalyst 25) is input to the plant model. The variable parameter of the plant model is sequentially identified so that the error e with respect to the detection voltage of the side exhaust gas sensor 32 approaches zero (this function corresponds to the identification means in the claims).

  Here, the plant model is a discrete plant model expressed in discrete time, approximated by a second-order lag system having a dead time, and as shown in FIG. 3, the second-order lag system is composed of two first-order lag systems. After being divided into discretizations, they are combined and approximated.

  The discretization of the plant model (upstream catalyst 25 + downstream exhaust gas sensor 32 model) is performed offline as follows. In this embodiment, the plant model is approximated by a second-order lag system having a dead time.

  The continuous plant model is expressed by the following equation with the output Y (s) and the input U (s).

ここで、ω、ζ、Ｋ、Ｌは連続モデルパラメータで、ωは固有角振動数、ζは減衰係数、Ｋは定常ゲイン、Ｌはむだ時間である。

Here, ω, ζ, K, and L are continuous model parameters, ω is a natural angular frequency, ζ is a damping coefficient, K is a steady gain, and L is a dead time.

  The dead time L is expressed by the following equation.

L = d × dt + L1
Here, d is the number of dead samplings, dt is the sampling period, and L1 is the dead time. That is, the quotient d obtained by dividing the dead time L by the discretization sampling period dt is the dead sampling count, and the remainder L1 is the dead time. When the dead time L cannot be divided by the sampling period dt, the dead time L1 becomes a positive value of 0 to dt by the dead time update process described later.

  The transfer function G (s) of the continuous plant model expressed by the above [Equation 1] is expressed by the following equation.

この連続プラントモデルを離散化（拡張ｚ変換）する際に、部分分数展開を利用して、次のように１次遅れ系の和に分解する。ここで、Z［＊］は＊の拡張ｚ変換を表す。拡張ｚ変換は、例えば「ディジタル制御システム−解析と設計（日刊工業新聞社）」において公知である。

When this continuous plant model is discretized (extended z-transform), partial fraction expansion is used to decompose it into a sum of first-order lag systems as follows. Here, Z [*] represents an extended z transformation of *. The extended z transformation is known, for example, in "Digital Control System-Analysis and Design (Nikkan Kogyo Shimbun)".

従って、離散プラントモデルは、次式で表される。

Therefore, the discrete plant model is expressed by the following equation.

ここで、ｐ1，ｐ2，q1，q2，q3は離散モデルパラメータであり，q1，q2，q3の中に余むだ時間Ｌ１が含まれる。

Here, p1, p2, q1, q2, and q3 are discrete model parameters, and the remaining time L1 is included in q1, q2, and q3.

  The discretization equations described above are summarized as shown in FIG.

  When the discrete model parameters are made continuous, continuous model parameters ω, ζ, K, and L1 are derived as shown in FIG. Zre and Zim in FIG. 5 are obtained by the calculation formula shown in FIG.

  For each of the above formulas, an approximate calculation may be used in order to actually allow calculation on board. For example, a trigonometric function, an exponential function, and a logarithmic function are Taylor-expanded up to the second order, and the range of large error is assumed to be a table.

As shown in FIG. 4, when discretized by the extended z-transform, information of the dead time L1 remaining in the discrete model parameters q1, q2, and q3 (this corresponds to “part of dead time information”) is included. This is a feature of the extended z-transform and does not appear in discretization by a general z-transform. The identification system identifies the discrete model parameter θ (θ = [p1, p2, q1, q2, q3] ^T : superscript T represents transposition), and the identification value θ_hat (x_hat is the identification value of x or The estimated values ω_hat, ζ_hat, K_hat, and L1_hat of the continuous model parameters can be calculated using the continuation formula of FIG. This function corresponds to the dead time estimation means in the claims. By using the estimated value L1_hat of the dead time, it is possible to estimate the dead time L (the number of times of dead sampling d). This function corresponds to dead time estimation means in the claims.

  Hereinafter, an identification process for identifying discrete model parameters will be described. In this identification process, the discrete model parameters are sequentially estimated by the least square method so that the identification error e (deviation between the actual output of the controlled object and the discrete plant model output) is zero. Hereinafter, it demonstrates in detail based on FIG.

The discrete plant model is a mathematical model represented by the above [Equation 4], and the discrete model parameter θ (θ = [p1, p2, q1, q2, q3] ^T ) is sequentially estimated by an adaptive mechanism.

  As shown in FIG. 2, the adaptive mechanism M32 estimates the discrete model parameter θ so that the identification error e between the predicted output of the discrete plant model and the output of the controlled object (detected voltage of the downstream side exhaust gas sensor 32) approaches zero. To do. The identification error e is an error between the model output y_hat when the actual control input u is input to the discrete plant model and the actual output y of the controlled object. The control input u and the actual output y (the detected voltage of the downstream exhaust gas sensor 32) to be input to the discrete plant model include unnecessary DC components and noise components. Therefore, the direct current component is removed from the control input u by an HPF (high pass filter) as a low frequency component removing means.

  In addition, in the actual output y to be controlled, the noise component is removed from the output y for F / B control by an LPF (low pass filter) as a high frequency component removing means, and the output y for parameter identification is HPF and LPF. The DC component and the noise component are removed by combining (bandpass filter). At this time, different LPFs can be set for the output y for the air-fuel ratio F / B and the output y for identification.

  The identification error e is subjected to filter processing (LPF) and dead zone processing as necessary. This filtering process suppresses the vibration of the identification parameter, and the dead band process suppresses overidentification. Note that this filter processing and dead zone processing may be omitted.

  Here, when the above [Equation 4] is re-expressed as an autoregressive model having time delay in input, it is expressed as follows.

上式を線形パラメトリックな自己回帰モデルとして表現しなおすと、次のように表される。

Re-expressing the above equation as a linear parametric autoregressive model, it is expressed as follows.

ここで、θ＿hat(k-1)＝［ｐ1＿hat(k-1)，ｐ2＿hat(k-1)，ｑ１＿hat(k-1)，ｑ２＿hat(k-1)，ｑ３＿hat(k-1)］^Ｔ，ζ(k-1)＝［ｙ(k-1)，ｙ(k-2)，ｕ(k-d)，ｕ(k-d-1)，ｕ(k-d-2)］^Ｔである。同定誤差ｅは次式で求められる。

Here, θ_hat (k-1) = [p1_hat (k-1), p2_hat (k-1), q1_hat (k-1), q2_hat (k-1), q3_hat (k-1)] ^T , ζ ( k-1) = [y (k-1), y (k-2), u (kd), u (kd-1), u (kd-2)] ^T. The identification error e is obtained by the following equation.

ここで，同定誤差eに対してノイズ除去や不感帯処理を施すことにより，同定パラメータの振動や過同定を避ける構成としてもよい．
次に、適応機構Ｍ３２において、同定誤差ｅをゼロにするように離散モデルパラメータの推定値θ＿hatを算出するパラメータ調整則は、本実施形態では重み付き最小二乗法の原理に基づいて導出される。次式に示す同定誤差ｅの２乗和を評価関数として考える。

Here, it is possible to avoid the identification parameter oscillation and over-identification by applying noise removal and deadband processing to the identification error e.
Next, in this embodiment, the parameter adjustment rule for calculating the estimated value θ_hat of the discrete model parameter so that the identification error e is zero is derived based on the principle of the weighted least square method. The sum of squares of the identification error e shown in the following equation is considered as an evaluation function.

ここで、λは忘却係数とも呼ばれる重み係数である。上記評価関数が最小となるようなパラメータ調整則は次式のように与えられる。

Here, λ is a weighting factor called a forgetting factor. A parameter adjustment rule that minimizes the evaluation function is given by the following equation.

上式において、同定誤差ｅに乗算する行列ゲインΓ及びスカラゲインγは次式のようになる。

In the above equation, the matrix gain Γ and the scalar gain γ multiplied by the identification error e are as follows.

そして、上式で求められた離散モデルパラメータの推定値が離散プラントモデルに反映される。

And the estimated value of the discrete model parameter calculated | required by the said Formula is reflected in a discrete plant model.

  In addition to the above, the parameter adjustment law is, for example, a fixed gain law, a gradual decrease gain law, a fixed forgetting gain law, a variable gain law, a fixed trace gain law, an upper limit trace gain law, a lower limit trace gain law, a limit trace gain law, etc. May be used.

  The discrete model parameter identified by the adaptive mechanism M32 is determined by the learning determination process whether or not the identification error e is within a predetermined range. If the identification error e is within the predetermined range, learning of the dead time L is permitted. Is done.

  The identified discrete model parameters p 1, p 2, q 1, q 2, q 3 are made continuous by the continuation formula shown in FIG. 5 and converted into continuous model parameters ω, ζ, K, L 1. And the dead time L is learned from the dead time L1 by a method described later. The learned value is stored in a rewritable nonvolatile memory such as a backup RAM of the ECU 29 so that the learned value can be stored even when the power is off.

  To the sub F / B controller M21, the learning values of the continuous model parameters ω, ζ, K, and L1 are discretized and input. The sub F / B controller M21 calculates an input (input oxygen amount) to be controlled using the discrete model parameter calculated from the learning value. At this time, the learning value (identification value) may be reflected sequentially or after the learning value has converged. Further, the discrete model parameter calculated in advance or the control gain itself may be switched depending on the range of learning values.

  Next, the dead time update process will be described. A continuous system having a dead time L is expressed by the following equation. However, for the sake of simplicity, a first-order lag system will be described as an example, but the same concept applies to higher-order systems.

そして、上式を拡張ｚ変換により離散化することで、次式が得られる。

Then, the following equation is obtained by discretizing the above equation by the extended z-transform.

但し、上記［数１１］式、［数１２］式において、ｙ（ｋ）＝ｘ（ｋ）である。

However, in the above [Expression 11] and [Expression 12], y (k) = x (k).

  Here, FIG. 7 is a time chart showing a change in the control input u (control input u shifted in the positive direction by the dead time) that affects the output y (k + 1) when kdt ≦ t <(k + 1) dt. is there. In FIG. 7, when the dead time L is not divisible by the sampling period dt, the input u (τ−L) changes only once in the period τ = kdt to (k + 1) dt, and d times before And the input u d + 1 times before. In the above [Expression 12], the second term on the right side represents the influence of the input u before d times, and the third term represents the influence of the input u before d + 1 times.

  In this case, the actual dead time of the control target is unknown, but the discrete model parameters discretized by the extended z transform include information on the extra time L1, and the dead time is obtained by using this extra time L1. L can be estimated. In the present embodiment, the dead time of the discrete plant model is made to approach the actual dead time of the controlled object based on the change of the estimated dead time L1_hat calculated by continuation of the estimated value of the discrete model parameter θ_hat. The time (dead sampling number d) is updated, and the updated dead time (dead sampling number d) is reflected in the calculation of the plant model output in the discrete plant model.

  More specifically, it is determined whether or not the estimated value L1_hat of the surplus time is near the upper limit value or the lower limit value defined by the sampling period dt. Here, the upper limit value of the estimated value L1_hat is dt, and the lower limit value is 0. By a small positive constant ε2, the vicinity of the upper limit value is “range of dt−ε2 to dt”, and the vicinity of the lower limit value is “range of 0 to 0 + ε2. ". In this case, when the estimated value L1_hat of the dead time changes to the vicinity of the upper limit value (dt−ε2 to dt) and the state continues for a predetermined time, the dead sampling number d is incremented by 1 and the dead time is increased. The estimated value L1_hat is set to a predetermined value near the lower limit value (however, the lower limit value = 0 or more). Further, when the estimated value L1_hat of the dead time changes to the vicinity of the lower limit value (0 to 0 + ε2) and the state continues for a predetermined time, the dead sampling number d is decremented by one and the estimated value L1_hat of the dead time is decreased. Is a predetermined value near the upper limit value (however, the upper limit value is equal to or less than dt).

  The outline will be described with reference to the time chart of FIG. As shown in FIG. 8A, a case is assumed where an error as shown in the figure occurs between the nominal dead time (nominal value of the dead time) for a certain input and the actual dead time. The nominal dead time is calculated using a preset nominal model scheduler. The portion exceeding the sampling timing (d · dt) is the extra time L1.

  In this case, as the identification is performed sequentially, as shown in FIG. 8B, the estimated value L1_hat of the surplus time changes and increases to the vicinity of dt (the range of dt−ε2 to dt described above). Then, when the estimated value L1_hat of the dead time remains for a predetermined time or more in the vicinity of dt, as shown in FIG. 8C, the estimated value d_hat of the number of dead samplings is incremented by one (d_hat ← d + 1). Further, the estimated value L1_hat of the surplus time is updated to a value near zero.

  Thereafter, by performing identification sequentially, as shown in FIG. 8D, the estimated value L1_hat of the dead time increases, and the estimated value of the dead time converges to the actual dead time accordingly. In this case, even if the error between the nominal dead time and the actual dead time is more than twice the sampling period, the estimated value of the dead time converges to the actual dead time by repeating the above update. Further, even when there is a fluctuation in the actual dead time due to deterioration after the convergence to the actual dead time, the estimated value of the dead time can always be converged to the actual dead time by executing the above update.

  FIG. 8 illustrates the case where the estimated value L1_hat of the extra time increases, and the situation where the estimated value d_hat of the number of dead samplings is incremented by one has been explained. When the estimated time value L1_hat decreases, the estimated value d_hat of the number of dead samplings is reduced by one, and the estimated value L1_hat of the remaining time is updated to a value near dt.

The adaptive control processing and the like described above are executed by the ECU 29 according to the programs shown in FIGS. Hereinafter, the processing contents of each program of FIGS. 9 to 15 will be described.
[Fuel injection control program]
The fuel injection control program in FIG. 9 is executed at a predetermined cycle (for example, 30 ° C. A cycle) during engine operation. When this program is started, first, in step 101, for example, a basic injection amount TP is calculated based on operating state parameters such as engine speed and load at that time using a basic injection amount map or the like.

  Thereafter, the routine proceeds to step 102 where it is determined whether or not the air-fuel ratio F / B control execution condition is satisfied. Here, the air-fuel ratio F / B control execution condition is determined by the following conditions (1) to (4), for example.

(1) The upstream side exhaust gas sensor 31 and the downstream side exhaust gas sensor 32 are activated.
(2) The air-fuel ratio control system operates normally (failure judgment is not performed)
(3) The engine coolant temperature is higher than a predetermined temperature (for example, 70 ° C. or higher), that is, after the engine 11 is warmed up.
(4) The engine operating state is in the F / B execution range. If any of these conditions (1) to (4) does not satisfy any of the conditions, the air-fuel ratio F / B control execution condition is not satisfied. In step 103, the air-fuel ratio correction coefficient FAF is set to “1.0”. In this case, the air-fuel ratio F / B control is not performed.

  On the other hand, if all of the above conditions (1) to (4) are satisfied, the air-fuel ratio F / B control execution condition is satisfied, the process proceeds to step 104, and a target air-fuel ratio calculation program of FIG. After calculating the upstream target air-fuel ratio λTG of the upstream catalyst 25, the routine proceeds to step 105, where the air-fuel ratio correction is performed so as to reduce the deviation between the air-fuel ratio detected by the upstream-side exhaust gas sensor 31 and the target air-fuel ratio λTG. The coefficient FAF is calculated.

  As described above, after calculating the air-fuel ratio correction coefficient FAF in step 103 or 105, the process proceeds to step 106, and the air-fuel ratio correction coefficient FAF and other various correction coefficients (for example, the cooling water temperature correction coefficient, the learning correction coefficient, The required fuel injection amount TAU is calculated by correcting the basic injection amount TP using a correction coefficient during deceleration.

TAU = TP × FAF × various correction factors [target air-fuel ratio calculation program]
The target air-fuel ratio calculation program in FIG. 10 is a subroutine executed in step 104 of the fuel injection control program in FIG. When this program is started, first, in step 201, an identification execution condition determination program shown in FIG. 11 described later is executed to determine whether the identification execution condition is successful, and the identification execution flag is set / reset according to the determination result. (ON / OFF).

  Thereafter, the process proceeds to step 202, where it is determined whether or not the identification execution flag is ON based on the execution result of the identification execution condition determination program in FIG. 11. If the identification execution flag is ON, the identification execution condition is satisfied. The process proceeds to step 203 where the target voltage of the downstream side exhaust gas sensor 32 is set. At this time, for example, a voltage obtained by adding a predetermined amplitude to the original target voltage may be set as the target voltage.

  Thereafter, the process proceeds to step 204, where an identification processing program shown in FIG. 13 to be described later is executed, and the plant model output when the control target input (inflow excess / deficient oxygen amount of the upstream catalyst 25) is input to the discrete plant model. The variable parameter of the discrete plant model is identified so that the error e with respect to the output to be controlled (detection voltage of the downstream side exhaust gas sensor 32) approaches zero. Thereafter, the process proceeds to step 205.

  On the other hand, if the identification execution flag is OFF, it is determined that the identification execution condition is not satisfied, and the process proceeds to step 205 without performing the identification processing (steps 203 and 204).

  In step 205, sub F / B control input calculation processing is executed. In this sub F / B control input calculation process, a model-based control law (for example, modern control or robust control) is used to bring the detection voltage of the downstream side exhaust gas sensor 32 close to the target voltage based on the parameters learned by the identification process. An input oxygen amount (inflow excess / deficiency oxygen amount of the upstream catalyst 25) to be a sub F / B control input is calculated. At this time, the sub F / B control input may be calculated using classical control such as PI control or PID control.

  In this embodiment, the input oxygen amount is regarded as the target oxygen amount. However, the input oxygen amount may be equal to the detected oxygen amount, or the input oxygen amount may be calculated from the air-fuel ratio correction coefficient.

  Thereafter, the routine proceeds to step 206, where the target air-fuel ratio λTG is calculated from the target oxygen amount (input oxygen amount) as follows. First, the target excess fuel ratio φTG at the time of this calculation is calculated by the following equation.

φTG = O (k) / (K ・ W ・ dt)
O (k): Target oxygen amount this time
K: Standard air oxygen mass ratio
W: Inflow exhaust gas flow rate [g / s] of the upstream catalyst 25
dt: Calculation cycle Here, W may use the engine intake air amount instead of the inflow exhaust gas flow rate of the upstream catalyst 25. Further, the calculation cycle dt may be time synchronization or engine rotation speed synchronization. In the case of engine speed synchronization, the calculation cycle dt is a function of the engine speed Ne. Here, φTG represents the target fuel excess amount per calculation cycle, but the target fuel excess amount per unit time may be used. In that case, φTG is as follows.

φTG = O (k) / (K ・ W)
The reciprocal of this target fuel excess ratio φTG is the target air-fuel ratio λTG.

λTG = 1 / φTG
Thereafter, the process proceeds to step 207, and the final target air-fuel ratio λTG is obtained by adding (or multiplying) the base target air-fuel ratio to the target air-fuel ratio λTG. At this time, the base target air-fuel ratio is calculated by, for example, a map using engine operating conditions as parameters.
[Identification execution condition judgment program]
The identification execution condition determination program in FIG. 11 is a subroutine executed in step 201 of the target air-fuel ratio calculation program in FIG. When this program is started, first, at step 301, it is determined whether or not catalyst deterioration is detected. If catalyst deterioration is detected, the routine proceeds to step 305 and the identification execution flag is set to ON.

  On the other hand, if it is determined in step 301 that the catalyst deterioration is not detected, the process proceeds to step 302, where an operation state determination program shown in FIG. 12 described later is executed to determine whether the current engine operation state is steady or transient. . Thereafter, the process proceeds to step 303, where it is determined whether or not the current engine operation state is transient based on the execution result of the operation state determination program of FIG. Is reset to OFF, and if it is steady, the process proceeds to step 305 and the identification execution flag is set to ON.

It should be noted that the identification execution condition may be only when the catalyst deterioration is detected, or conversely, the identification execution condition may be excluded from the identification execution condition. Further, the identification execution flag may be set to ON in the entire operation region where the air-fuel ratio F / B control is executed.
[Operation status judgment program]
The operation state determination program in FIG. 12 is a subroutine executed in step 302 of the identification execution condition determination program in FIG. 11 and plays a role as operation state determination means. When this program is started, first, in step 301, it is determined whether or not the engine speed change amount (absolute value of the change rate) is equal to or less than a predetermined value. In step 404, it is determined that the current engine operating state is transient.

  On the other hand, if it is determined in step 401 that the engine speed change amount is equal to or less than the predetermined value, the process proceeds to step 402 to determine whether or not the load change amount (absolute value of the load change rate) is equal to or less than the predetermined value. If the load change amount is not less than or equal to the predetermined value, the process proceeds to step 404 to determine that the current engine operating state is transient.

  If both of the above steps 401 and 402 are determined as “Yes”, that is, if the engine speed change amount is not more than a predetermined value and the load change amount is not more than a predetermined value, the process proceeds to step 403 and the current engine It is determined that the operating state is steady.

In addition, when it is not determined that the engine is stationary for a predetermined period of time or longer during engine operation, a stationary determination value for the engine speed change amount and the load change amount (predetermined values in steps 401 and 402) is used to ensure the identification execution frequency. (Value) may be changed to a larger value.
[Identification processing program]
The identification processing program of FIG. 13 is a subroutine executed in step 204 of the target air-fuel ratio calculation program of FIG. 10, and plays a role as identification means in the claims. When this program is started, first, in step 501, the input oxygen amount (inflow excess / deficient oxygen amount of the upstream catalyst 25) is calculated by either of the following methods (1) and (2).

  (1) The input oxygen amount is calculated by correcting the target oxygen amount according to the change in the base target air-fuel ratio.

  (2) The inflow gas air-fuel ratio of the upstream catalyst 25 is set to a value based on the detected air-fuel ratio of the upstream exhaust gas sensor 31 or the fuel injection correction coefficient, and the input oxygen amount is calculated by the following equation.

Input oxygen amount = K · W · (theoretical equivalent ratio−catalyst inflow gas equivalent ratio) · dt
K: Standard air oxygen mass ratio
W: catalyst inflow exhaust gas flow rate [g / s] (may be engine intake air amount)
dt: Calculation cycle Here, W may use the engine intake air amount instead of the inflow exhaust gas flow rate of the upstream catalyst 25. Further, the calculation cycle dt may be time synchronization or engine rotation speed synchronization. In the case of engine speed synchronization, the calculation cycle dt is a function of the engine speed Ne. Here, the input oxygen amount represents the input oxygen amount per calculation cycle, but the input oxygen amount per unit time may be used. In that case, the input oxygen amount is calculated by the following equation.

Input oxygen amount = K · W · (theoretical equivalent ratio-catalyst inflow gas equivalent ratio)
After calculating the input oxygen amount, the process proceeds to step 502, and the output of the discrete plant model is calculated. At this time, the same predetermined signal processing is applied to the output and detection output of the discrete plant model. This predetermined signal processing is, for example, one of LPF, HPF, and BPF, but may be other processing.

  Thereafter, the process proceeds to step 503, where the identification error e is calculated from the actual output of the controlled object and the output of the discrete plant model. In the next step 504, parameter adaptation processing is executed according to the parameter adjustment rule described above. Thereby, the estimated value θ_hat of the discrete model parameter is calculated so that the error between the actual output of the controlled object and the output of the discrete plant model approaches zero.

  Thereafter, the process proceeds to step 505, where it is determined whether or not the identification error e is within a predetermined range by learning determination processing. If the identification error e is within the predetermined range, learning of the dead time L is permitted. Thereafter, the process proceeds to Step 506, where the calculated continuous model parameter estimated value θ_hat is subjected to a continuation process using the continuation formula of FIG. 5 to obtain a continuous model parameter (estimated value L1_hat of the remaining time). Is calculated. The learning determination process in step 505 may be performed after the continuous process in step 506.

  Then, in the next step 507 (update means), a dead time update processing program shown in FIGS. 14 and 15 described later is executed to learn the dead time L1. Thereafter, the process proceeds to step 508, where the continuous model parameters are discretized using the discretization formula of FIG. 4 to update the parameters of the discrete plant model.

Note that the algorithm for parameter adaptation processing is not limited to this embodiment, and other algorithms may be used.
[Dead time update program]
The dead time update processing program of FIG. 14 and FIG. 15 is a subroutine that is repeatedly executed at the calculation cycle dt (for example, every 180 ° C. A or every predetermined time) of the program in step 507 of the identification processing program of FIG. It plays a role as dead time estimation means and learning means. The minute positive constants ε1, ε2, and δ used in this program are set to have a relationship of ε1 <ε2 <δ.

  When this program is started, first, at step 601, it is determined whether or not the estimated value L1_hat of the remaining time calculated at step 506 in FIG. 13 is smaller than a predetermined determination value “0 + ε 2”. That is, it is determined whether or not the estimated value L1_hat of the surplus time is in the vicinity of the lower limit value (0 to 0 + ε2) of the surplus time. If L1_hat ≧ 0 + ε2, the process proceeds to step 603 and the counter cnt_mins is cleared to zero.

  On the other hand, if it is determined in step 601 that L1_hat <0 + ε2 (that is, if L1_hat is near the lower limit value), the process proceeds to step 602, and the counter cnt_mins is incremented by one. Thereafter, the process proceeds to step 604, where the value of the counter cnt_mins is equal to or greater than the predetermined value N1 and the estimated value L1_hat of the remaining time is smaller than a minute negative constant (−δ), or the value of the counter cnt_mins is the predetermined value N2. It is determined whether any of the above is true. Here, N1> N2. Note that the determination condition of step 604 is that the value of the counter cnt_mins is equal to or greater than the predetermined value N1 and the estimated value L1_hat of the remaining time is smaller than a minute negative constant (−δ), or It is also possible that only the value of cnt_mins is equal to or greater than a predetermined value.

  If “No” is determined in Step 604, the process proceeds to Step 606, and the dead time L1_adp is set to “0”.

  On the other hand, if “Yes” is determined in Step 604, the process proceeds to Step 605, and the dead time L1_adp, the dead sampling frequency change amount Δd_adp, and the counter cnt_mins are updated.

L1_adp = dt−δ
Δd_adp = −1
cnt_mins = 0
After executing the processing in step 603, 605, or 606 as described above, the process proceeds to step 607 in FIG. 15, and whether or not the estimated time L1_hat of the remaining time is equal to or greater than a predetermined determination value “dt−ε2”. Determine. That is, it is determined whether or not the estimated value L1_hat of the surplus time is near the upper limit value of the surplus time (dt−ε2 to dt).

  If L1_hat <dt−ε2, the process proceeds to step 608, and the counter cnt_pulus is cleared to zero.

  On the other hand, if it is determined in step 607 that L1_hat ≧ dt−ε2 (that is, if L1_hat is in the vicinity of the upper limit value), the process proceeds to step 609 and the counter cnt_puls is incremented by one.

  Thereafter, the process proceeds to step 610, where the value of the counter cnt_puls is equal to or larger than the predetermined value N1 and the estimated value L1_hat of the remaining time is smaller than a value (dt + δ) larger than the upper limit by a small positive constant δ, or the counter cnt_puls It is determined whether or not any of the values is equal to or greater than a predetermined value N2. Note that the determination condition in step 610 is that the value of the counter cnt_puls is equal to or greater than the predetermined value N1 and the estimated value L1_hat of the remaining time is smaller than (dt + δ), or the value of the counter cnt_puls is equal to or greater than the predetermined value N2. It is also possible to do something only.

  If “No” is determined in step 610, the process proceeds to step 611, and the remaining time L1_adp is updated as “dt−ε1”.

  On the other hand, if “Yes” is determined in step 610, the process proceeds to step 612, where the dead time L1_adp, the dead sampling frequency change amount Δd_adp, and the counter cnt_puls are updated.

L1_adp = δ
Δd_adp = + 1
cnt_puls = 0
After executing the processing of step 608 or 611 or 612 as described above, the process proceeds to step 613 of FIG. 14 to determine whether or not the learning flag is ON, meaning that the learning value is updated. If it is ON, the process proceeds to step 614 to clear the accumulated value Σ (Δd_adp) of the amount of change in the number of dead samplings up to the previous time, and if the learning flag is OFF, the accumulated value Σ (Δd_adp) is not cleared.

  Thereafter, the process proceeds to step 615, where the accumulated value Σ (Δd_adp) is added to the dead sampling number learning value d_lrn to update the dead sampling number d_adp.

d_adp = d_lrn + Σ (Δd_adp)
Incidentally, noise is included in the value input to the plant model at the time of identification and the output value from the plant model corresponding to the input, and this noise causes a decrease in the accuracy of identification. More specifically, if L1_hat vibrates more than zero to dt in a short time due to noise or the like included in the estimated value L1_hat of the extra time as the output value, the number of dead samplings d and the extra time L1 May be erroneously updated, leading to deterioration of dead time estimation accuracy.

  FIGS. 16 (a) and 16 (b) are diagrams showing changes in the estimated value L1_hat of the surplus time with the passage of time. FIG. 16 (a) shows that the vibration of L1_hat is larger than zero to dt as described above. The estimated value L1_hat fluctuates so as to frequently cross zero or dt (dt in the example of FIG. 16A) that becomes the threshold for updating the remaining time L1. Therefore, in this case, the sampling number d and the extra time L1 are erroneously updated due to the influence of noise. On the other hand, FIG. 16B shows a case where the sampling period dt is set longer than that in FIG. 16A, and in this case, it is possible to suppress the update from being erroneously performed due to the influence of noise. In this embodiment, the length of the sampling period dt is set to be sufficiently long in order to suppress erroneous updating in this way.

  Hereinafter, the length of the sampling period dt will be described more specifically with reference to FIG.

  A calculation cycle t10 in the main F / B controller M10 mainly composed of the main F / B controller M11, and a sub F / B controller M20 composed mainly of the sub F / B controller M21 (control input calculating means) ) And the calculation cycle t30 in the identification calculation unit M30 mainly composed of the plant model M31 and the adaptive mechanism M32 are set to different calculation cycles.

  The plant model M31 represents an object to be controlled by a mathematical expression discretized at a predetermined sampling period dt, and the sampling period dt is a calculation period t30 in the identification calculation unit M30. Then, by setting the calculation cycle t30 in the identification calculation unit M30 to be longer than the calculation cycle t20 in the sub F / B control unit M20, the sampling cycle dt to the extent that the above-described frequent update can be suppressed. The length is sufficiently long. The relationship between the calculation cycle t20 in the sub F / B control unit M20 and the calculation cycle t10 in the main F / B control unit M10 may be t20> t10, or t20 = t10.

  Here, the discretized plant model is obtained by discretizing a continuous plant model representing a control target (the catalyst 25 and the downstream side exhaust gas sensor 32) at a predetermined sampling period dt. Therefore, if this sampling period dt is excessively long, the accuracy of the output with respect to the input of the discretized plant model decreases. Therefore, in this embodiment, the sampling period dt is restricted from becoming excessively long so that sufficient output accuracy can be ensured for the discretized plant model. Specifically, the length of the sampling period dt is limited to be shorter than 1/10 of the time constant of the control target.

  Of the discrete model parameters θ estimated by the adaptive mechanism M32, the residual time estimation value L1_hat is output to the time delay learning unit M40 that updates and learns the time delay after being continuous by the continuation means M33. The time delay learning unit M40 calculates a time delay L based on the continuous time delay estimation value L1, updates the calculated time delay L, and learns the learning value (time delay L) from the identification calculation unit M30 and Output to each of the sub F / B control units M20.

  The identification calculation unit M30 includes discretization means M34 (first discretization means) for discretizing the dead time L from the dead time learning unit M40, and waste sampling obtained by conversion by the discretization means M34. Using the number of times d (dead time L), the model parameters of the plant model M31 are sequentially identified.

  Further, among the discrete model parameters θ estimated by the adaptive mechanism M32, parameters other than the dead time are also made continuous by the continuation means M33 and learned by a learning unit (not shown), and the learning value (continuous The conversion parameters ω, ζ, K) are output to the sub F / B control unit M20.

  The sub F / B control unit M20 includes discretization means M22 (second discretization means) for discretizing the dead time L and other parameters ω, ζ, and K from the dead time learning unit M40. The number of times of dead sampling d (dead time L) and various parameters ω, ζ, K obtained by the conversion in the converting means M22 are used for sub F / B control.

  According to the embodiment described in detail above, the following effects can be obtained.

  (1) By setting the calculation cycle t30 in the identification calculation unit M30 to be longer than the calculation cycle t20 in the sub F / B control unit M20, the sampling cycle dt that determines the threshold for updating the dead time L1 is set. It is big enough. Therefore, as shown in FIG. 16B, it is possible to suppress the fluctuating estimated value L1_hat from crossing the remaining time L1 update threshold value (zero or dt) by mistake, and thus it is possible to suppress the deterioration of the dead time estimation accuracy. Furthermore, the processing load on the ECU 29 can be reduced by setting the calculation cycle t30 in the identification calculation unit M30 having a large calculation amount to be long.

  (2) Since the length of the sampling period dt is limited so as to be shorter than one-tenth of the time constant of the controlled object, it is possible to sufficiently secure the output accuracy with respect to the input of the discretized plant model.

  (3) As described above, the control input u input to the discrete plant model and the actual output y (the detection voltage of the downstream side exhaust gas sensor 32) to be controlled include unnecessary DC components and noise components. . On the other hand, since the present embodiment includes various filters (for example, at least one of HPF, LPF, and BPF), the fluctuation of the remaining time estimate value L1_hat (see FIG. 16) includes noise extending in a bowl shape. Can be suppressed. Alternatively, the variation can be suppressed.

  (4) Here, the calculation cycle t20 in the sub F / B control unit M20 and the calculation cycle t30 in the identification calculation unit M30 are set to different values, but if both the cycles t20 and t30 are different in this way, There is a concern that the accuracy of feedback control by the sub F / B controller M21, the identification accuracy in the adaptive mechanism 32, etc., and the estimation accuracy of the dead time L are reduced. On the other hand, in this embodiment, the learning value (dead time L) learned by the dead time learning unit M40 based on the estimated dead time L1_hat is used to identify the discrete plant model M31 and the sub F / B controller M21. Since it is appropriately used for feedback control, the above-described concerns about the reduction in accuracy can be easily solved.

  (5) In the present embodiment, discretization is performed by extended z transformation so that the information of the extra time L1 is included in the discrete model parameter, and the extra time L1 calculated from the discrete model parameter is used as the basis. In addition, the dead time of the discrete plant model is estimated to be close to the actual dead time of the controlled object, and further, the estimated dead time is made to follow the change of the actual dead time by increasing or decreasing the number of dead sampling times d of the discretization. As a result, not only the response delay but also the dead time can be sequentially estimated, the dynamic characteristics of the controlled object can be sufficiently grasped regardless of the variation and aging of the controlled object, and high control accuracy can be maintained. . In addition, the dead time information can be used in various ways.

  (6) In this embodiment, the discrete model parameter sequentially estimated based on the identification error is converted into the continuous model parameter, and the dead time is estimated from the continuous model parameter. Therefore, the calculation is performed from the discrete model parameter. Compared to the case, the dead time estimation can be easily calculated.

  (7) Since the configuration is such that the estimated dead time is learned, even when there is a deviation in actual dead time due to individual differences in control objects, changes with time, etc., it is possible to continuously perform control following that. At this time, since the learning value is updated when the error between the plant model output and the actual output of the controlled object is within a predetermined range, the learning process is performed on the condition that proper identification is performed. There is also an advantage that the reliability of the learning value is improved.

(Other embodiments)
The above embodiment may be modified as follows. Further, the present invention is not limited to the description of the above embodiment, and the characteristic configurations of the embodiments described below may be arbitrarily combined.

  In the above embodiment, when setting the calculation cycle t30 (that is, the sampling cycle dt) in the identification calculation unit M30 to sufficiently increase the sampling cycle dt that determines the threshold for updating the surplus time L1, the sub-F / The calculation period t20 in the B control unit M20 is set as a reference so that t30> t20. On the other hand, a fluctuation range ΔN (see FIG. 16B) due to noise included in the remaining time estimate value L1_hat is set by measurement or the like, and the calculation cycle t30 ( The sampling period dt) may be set.

  In the above embodiment, in order to make the sampling cycle dt sufficiently long, the calculation cycle t30 in the identification calculation unit M30 is set longer than the calculation cycle t20 in the sub F / B control unit M20. In implementation, the calculation cycle t30 may be set longer than the calculation cycle t10 in the main F / B control unit M10. That is, the “control input calculation means” described in the claims is not limited to the sub F / B control unit M20, and may be the main F / B control unit M10, for example.

  In the above embodiment, the estimated value θ_hat of the discrete model parameter is converted into a continuous model parameter, and the dead time L (the number of times of dead sampling d) is estimated using the estimated value L1_hat of the dead time included in the continuous model parameter. Instead of this, it is also possible to estimate the dead time L (dead sampling number d) using the estimated values of the discrete parameters including the dead time L1 without continuation of the discrete parameters. .

  In the above embodiment, the control target (the upstream catalyst 25 and the downstream exhaust gas sensor 32) is approximated by a second-order lag system including a dead time, but the present invention is a third order including a dead time. You may make it approximate with the above high-order delay systems. In this case, the third-order or higher-order delay system may be divided into three or more first-order delay systems, discretized, and then combined and approximated.

The application range of the present invention is not limited to the sub F / B system (air-fuel ratio control system) including the upstream catalyst 25 and the downstream exhaust gas sensor 32, but the main F / B including the engine 11 and the upstream exhaust gas sensor 31. The present invention may be applied to a system (internal combustion engine control system).
In the above embodiment, the identification unit does not use the nominal model parameter, but may be configured to identify a deviation from the nominal model parameter (nominal value of the model parameter) set in accordance with the engine operating condition or the like. . In this case, it is preferable to provide a filter means for suppressing a high-frequency vibration component of a signal representing engine operating conditions and the like. The nominal model parameter is calculated using a preset nominal model scheduler.

  In the above embodiment, the two catalysts 25 and 30 are provided in the exhaust pipe 23, but the present invention can be applied to a system in which only one catalyst is provided. In addition, the present invention is not limited to the sub F / B system and the main F / B system, and may be applied to other control objects.

It is a lineblock diagram showing an outline of an engine control system in one embodiment of the present invention. It is a block diagram which shows the outline | summary of the main F / B control function and sub F / B control function which are implement | achieved by ECU. It is a figure explaining the method of producing | generating a discrete plant model by dividing | segmenting the continuous plant model approximated with a secondary delay + dead time into a primary delay system, and discretizing it. It is a figure explaining the method of discretizing a continuous model parameter and calculating a discrete model parameter. It is a figure explaining the method of calculating a continuous model parameter by making a discrete model parameter (identification value) continuous. It is a figure explaining the method of calculating Zre and Zim in FIG. It is a time chart showing change of control input u. It is a time chart which shows the outline | summary of a dead time update process concretely. It is a flowchart which shows the flow of a process of a fuel-injection control program. It is a flowchart which shows the flow of a process of a target air fuel ratio calculation program. It is a flowchart which shows the flow of a process of an identification execution condition determination program. It is a flowchart which shows the flow of a process of a driving | running state determination program. It is a flowchart which shows the flow of a process of an identification process program. It is a flowchart which shows the flow of a process of a dead time update process program (the 1). It is a flowchart which shows the flow of a process of a dead time update process program (the 2). It is a figure showing the state from which surplus dead time estimated value L1_hat changes, vibrating with noise, (a) is the state by conventional control, (b) is the state by control concerning one Embodiment of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 11 ... Engine (internal combustion engine), 23 ... Exhaust pipe, 25 ... Upstream catalyst, 29 ... ECU (Identification means, Dead time estimation means, Dead time estimation means, Dead sampling frequency update means), 30 ... Downstream side Catalyst, 31 ... upstream exhaust gas sensor, 32 ... downstream exhaust gas sensor, M10 ... main F / B controller, M11 ... main F / B controller, M20 ... sub F / B controller (control input calculation means), M21 ... Sub F / B controller, M30 ... identification calculation unit, M32 ... adaptive mechanism, dt ... dead sampling cycle, t30 ... calculation cycle of identification calculation unit (= dt), t20 ... calculation cycle of sub F / B control unit.

Claims (5)

Control input calculating means for calculating a control input to the control target or a value used for calculation of the control input;
An identification means for sequentially identifying the plant model parameters for a plant model representing the object to be controlled by a mathematical expression discretized at a predetermined sampling period (dt);
When the quotient obtained by dividing the dead time (L) of the control target by the sampling period is modeled as a dead sampling number (d) and the remainder is a dead time (L1), the identification parameter identified by the identification unit is A surplus time estimation means for estimating the surplus time included in the plant model using,
When the estimated dead time estimated by the dead time estimation means changes to the vicinity of the sampling period, the estimated value of the number of dead samplings is updated to be incremented by one. A waste sampling number update means for updating the estimated value of the number of waste sampling times to lower by one;
Dead time estimation means for estimating the dead time based on the estimated number of dead sampling times updated by the dead sampling number update means;
With
The control apparatus characterized in that the sampling cycle is longer than the calculation cycle by the control input calculation means. Control input calculating means for calculating a control input to the control target or a value used for calculation of the control input;
An identification means for sequentially identifying the plant model parameters for a plant model that represents the object to be controlled by a mathematical expression discretized at a predetermined sampling period (dt);
When the quotient obtained by dividing the dead time (L) of the control target by the sampling period is modeled as a dead sampling number (d) and the remainder is a dead time (L1), the identification parameter identified by the identification unit is A surplus time estimation means for estimating the surplus time included in the plant model using,
When the estimated residual time estimated by the residual time estimation means has changed to the vicinity of the sampling period, the estimated value of the number of dead samplings is updated to be incremented by one. A waste sampling number update means for updating the estimated value of the number of waste sampling times to lower by one;
Dead time estimation means for estimating the dead time based on the estimated number of dead sampling times updated by the dead sampling number update means;
With
The control apparatus according to claim 1, wherein the sampling period is longer than a fluctuation range due to noise included in the estimated surplus time estimated by the surplus time estimation means. The identification means includes a filter means for suppressing a high-frequency vibration component included in the estimated residual time estimated by the residual time estimation means,
3. The control apparatus according to claim 2, wherein the sampling period is longer than a fluctuation range due to noise included in an estimated remaining time after being filtered by the filter unit. Comprising continuous means for converting the discretized identification parameters into continuous model parameters;
The dead time estimation means estimates the dead time using the continuous model parameter continuous by the continuous means,
First discretization means for converting the estimated dead time estimated by the dead time estimation means into an estimated value of the number of dead samplings discretized at the sampling period in the identification means;
The model parameter for control and the waste for control obtained by discretizing the continuous model parameter continuous by the continuous means and the estimated dead time estimated by the dead time estimation means in the calculation cycle in the control input calculation means A second discretization means for converting the number of sampling times;
With
The estimated value of the number of dead samplings discretized by the first discretization means is used for sequential identification in the identification means, and the control model parameter discretized by the second discretization means and the control The control apparatus according to claim 1, wherein the number of dead samplings is used for the calculation by the control input calculation unit. The control objects include an exhaust gas purifying catalyst disposed in an exhaust passage of an internal combustion engine, an upstream exhaust gas sensor for detecting a specific exhaust gas concentration upstream of the catalyst in the exhaust passage, and the catalyst in the exhaust passage. A downstream exhaust gas sensor for detecting a specific exhaust gas concentration downstream of the air-fuel ratio control system,
A main feedback control means that feeds back the air-fuel ratio detected by the upstream exhaust gas sensor so as to coincide with the upstream target air-fuel ratio of the catalyst, and calculates a supply air-fuel ratio as the control input;
Sub-feedback control means that feeds back the air-fuel ratio detected by the downstream exhaust gas sensor so as to coincide with the downstream target air-fuel ratio of the catalyst, and calculates the upstream target air-fuel ratio as a value used for calculating the control input When,
With
The control apparatus according to claim 1, wherein the control input calculation unit is the main feedback control unit or the sub feedback control unit.

Images