JP2000001114A - Air conditioning device for vehicle - Google Patents

Abstract

PROBLEM TO BE SOLVED: To permit control of vehicle interior temperature at high accuracy by using outside temperature and amount of solar radiation as environmental factors, the statuses of the air conditioning device being a blowoff mode, blowoff temperature, and blowoff amount, and fixing the temperature data of blowers other than the blower set in the blowoff mode to predetermined values. SOLUTION: A control unit 1 has a temperature estimation value 1b composed of a neutral network which uses a measured temperature T1 from a room temperature sensor 6 and an environmental factor and statuses of an air conditioning device from a control logic 1a as input signals and which outputs an estimation value TN of a vehicle interior temperature to the control logic 1a. The input signals to the neutral network include, measured temperature T1 from the room temperature sensor 6, outside temperature TA from an outside temperature sensor, amount of solar radiation from a solar radiation sensor 8, a duty ratio of a blower 2b, and a blowoff mode that has been set. A foot blowoff temperature TF and a vent blowoff temperature TB for blowers other than the set blower are fixed at predetermined values.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an air conditioner, and more particularly to an air conditioner for a vehicle for appropriately controlling the temperature in a passenger compartment to a target temperature.

[0002]

BACKGROUND ART FIG. 8 is a diagram showing a control flow of a conventional vehicle air conditioner, a conventional air conditioner for a vehicle, with respect to the target temperature T _Z that is set in the temperature setting unit 4, the vehicle interior the located control logic 1p was the feedback value of the measured temperature T _i measured by the inside air temperature sensor 6 as the vehicle interior temperature detection means, so as to control the temperature and air volume, etc. of the conditioned air to be blown into the passenger compartment . However, the inside air temperature sensor 6 is normally therefore are installed below the front panel, due to the effect of such ambient temperature and solar radiation, different from the temperature (vehicle interior temperature) T ₀ of the occupant's seating position near I have.
Therefore, the feedback value is set to the actual vehicle interior temperature T _0.
, A vehicle is provided with an outside air temperature sensor and a solar radiation sensor, and the measured temperature T _{i is} determined based on the output of each sensor.
Or corrected, and further, based on the blowing mode and the like of the conditioned air was correcting the measured temperature T _i.

[0003]

[SUMMARY OF THE INVENTION However, in order to correct the difference between the actual and the inside temperature T ₀ and the measured temperature T _i measured by the inside air temperature sensor, since parameters used by the control logic is large, the matching work It takes time. JP-A-6-195323, as shown in FIG. 9, the target temperature _{T Z,} (measured temperature of the inside air temperature sensor) inside temperature _{T i,} the outside air temperature _{T A,} the input signal and a solar radiation amount _{T S} A technology for controlling the amount of air-conditioned air to be blown using the neural network type additional learning device described above is disclosed. In the neural network, the final blowing rate is controlled by learning and calculating respective arithmetic expressions such as a target blowing temperature, a blowing mode state, and a blower air volume. In order to learn the input / output relationship of the neural network, Since a large number of teacher signals are used, there is a problem that the learning does not easily converge and the learning time becomes long.

Therefore, the present applicant has a temperature estimator configured by a neural network that outputs the vehicle interior temperature, which is a main element of the control logic of the air conditioner, as an output value. A technique has been proposed in which the control logic is easily adjusted by controlling the temperature and the temperature in the vehicle compartment is controlled with high precision (Japanese Patent Application No. 10-0079).
No. 30). This is as an input signal of the neural network, and the measured temperature T _i measured by the inside air temperature sensor, outside temperature, and environmental factors consisting of solar radiation, blowing mode, mix door opening degree, the air-conditioning equipment consisting of a blower motor duty ratio state estimating the temperature T _N of the passenger compartment with the door, it is to control the temperature in the passenger compartment based on the estimated value T _N. However, among the input signals, the mixed door opening, which is an input signal corresponding to the blowout temperature, is easily affected by the blowout mode and the blower motor duty ratio, and therefore, the accuracy of estimating the vehicle interior temperature cannot be said to be sufficient.

SUMMARY OF THE INVENTION The present invention has been made in view of the conventional problems, and appropriately selects an input signal of a neural network when estimating a vehicle interior temperature which is a main element of control logic of an air conditioner. It is another object of the present invention to provide a vehicle air conditioner capable of controlling a vehicle interior temperature with high accuracy.

[0006]

According to a first aspect of the present invention, there is provided an air conditioner for a vehicle, wherein a temperature detected by a vehicle interior temperature detecting means, an environmental factor of the vehicle, and a state of an air conditioner are input signals.
When constructing a neural network that uses the estimated value of the temperature in the vehicle cabin as an output value, the environmental factors are the outside air temperature and the amount of solar radiation, and the state of the air conditioner is the blowing mode, the blowing temperature,
In addition to the blowout air volume, the temperature data of the outlets other than the outlets set in the blowout mode is fixed at a predetermined value.

According to a second aspect of the present invention, there is provided a vehicle air conditioner wherein the estimated value is used as a feedback value to control the temperature in the passenger compartment.

In the vehicle air conditioner according to the present invention, the teacher signal used at the time of learning the neural network is a four-point average temperature at positions corresponding to the head and the foot of the driver's seat and the passenger's seat. Things.

According to a fourth aspect of the present invention, the input state of the teacher signal used in learning of the neural network is normalized from 0.02 to 0.98.

According to a fifth aspect of the present invention, the sigmoid function used for the neural network is constituted by a linear function having coefficients different depending on an input range.
And the absolute value of the error between the output value of the linear function and the output value of the sigmoid function in each input range is 3%.
Within the range, the input range and the coefficient of the linear function are approximated by a set function.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. Embodiment 1 FIG. FIG. 1 is a diagram showing a configuration of a vehicle air conditioner according to Embodiment 1 of the present invention, in which 1 is a control device, 2 is an air duct, 3 is a setting panel for setting the temperature and air volume of conditioned air, and 4 is A temperature setter for setting a target temperature TZ in the vehicle compartment input from the setting panel 3 and outputting the target temperature _TZ to the control device 1;
5 is a plurality of driving devices for adjusting the opening of the mix door 2c in the air duct 2, 6 is an inside air temperature sensor as room temperature detecting means installed below the front panel, and 7 is installed near the bumper of the vehicle. The outside air temperature sensor 8 is a solar radiation sensor installed above the front panel. The air duct 2 includes an inside / outside air switching door 2a that adjusts a ratio of inside air and outside air introduced into the air duct, a blower 2b that blows air sucked from the inside / outside air switching door 2a, and an evaporator 2c that cools the blowing air. A heater 2d for warming the blast air, a mix door 2e for controlling the temperature of the blast air by controlling the amount of air passing through the heater 2d out of the blast air cooled by the evaporator 2c based on the degree of opening and closing thereof, and An outlet switching door 2h for distributing air to the upper outlet 2f and the lower outlet 2g. The upper outlet 2f is connected to the upper outlet 2f.
An upper outlet temperature sensor 2m for detecting the temperature of the lower outlet 2g is provided, and a lower outlet temperature sensor 2n for detecting the temperature of the lower outlet 2g is provided for the lower outlet 2g. The driving device 5 includes a plurality of actuators for opening and closing the inside / outside air switching door 2a, the mixing door 2e, and the outlet switching door 2h, and a blower driving circuit for driving the blower 2b. The control device 1 outputs a control signal for controlling the opening degree of the mix door 2e or the like to each drive device 5 in accordance with an environmental factor such as an outside air temperature or a state (control factor) of the air conditioner such as a conditioned air blowing mode. Control logic 1a
When the environmental factors and the information of the state of air conditioning equipment from the measured temperature T _i and the control logic 1a from the inside air temperature sensor 6 as an input signal, outputs the estimated value T _N of the temperature of the vehicle interior to be output to the control logic 1a And a temperature estimator 1b composed of a neural network for setting a value. The control factors include, for example, outputs from the outside air temperature sensor 7 and the solar radiation sensor 8, the degree of opening and closing of the inside / outside air switching door 2a, the mix door 2e, and the outlet switching door 2h, and the drive voltage of the blower 2b.

FIG. 2 is a diagram showing a configuration (temperature model) of a neural network in the temperature estimator 1b. The neural network is a three-layer hierarchical network including an input layer, a hidden layer, and an output layer. The input signal is a measured temperature T _i (° C.) from the inside air temperature sensor (Inc.s) 6,
Outside temperature T _A (° C.) from outside temperature sensor 7, solar radiation sensor 8
(Kcal / m ² hour), blower 2
b, the blower duty ratio (%) corresponding to the drive voltage, and the blowing mode (1, 2, 3,
4, 5), the foot outlet temperature _TF (° C.) from the lower outlet temperature sensor 2n, and the vent outlet temperature T _B (° C.) from the upper outlet temperature sensor 2m.
This is an estimated value T _N (° C.) of the temperature in the vehicle interior. However, the foot air temperature _{T F} (° C.) and the vent outlet air temperature _{T B}
(° C.) means that the temperature data of the outlets other than the outlets set in the above-mentioned outlet mode is a predetermined value (for example, 25 ° C.).
° C). In this neural network, a sigmoid function as shown in the following equation (1) is used. _{y j = 1 / (1 +} exp (- | x j |) ‥‥‥ (1) where, _{x j} is the weight to each input signal _{v i} to the layer i w
_ij multiplied by the bias b _j (x _j
= In _{Σv i · w ij -b j)} , the _{y j} is the output signal from layer i (input signal to the layer j). As shown in FIG. 3, this sigmoid function outputs 0 to 1 for input variables of -∞ to + ∞. Temperature estimator. 1b, as a teacher signal in the neural network training, using a 4-point average temperature of the position corresponding to each of the head and foot of the driver's seat and the passenger seat, the arithmetic expression for the estimated value T _N The weight w _ij and the bias b _j for each input signal are obtained by learning, and at the time of control, an estimated value _TN of the vehicle interior temperature is output to the control logic 1a for each of the input signals.

[0013] As the input state to the neural network, each input signal represents the minimum value to the maximum value of the measured data as 0.
Is normalized to 1 so that the weights w _ij for the types of the input signals can be evaluated in the same manner. on the other hand,
The teacher signal normalizes the minimum value to the maximum value of the measurement data from 0.02 to 0.98 in consideration of the fact that 0 and 1 are saturation output values as output characteristics of the sigmoid function. That is, in learning a neural network using the sigmoid function as an input function of a neuron, in consideration of learning efficiency and safety, a range of 0.
The convergence speed is faster and more effective when normalized from 02 to 0.98.

[0014] Figure 4 is the temperature estimator 1b composed of a neural network as described above, was thoroughly learned by the learning conditions below, is a graph estimating the vehicle interior temperature T ₀ by entering the outside air temperature data. In addition, the vehicle interior temperature T
₀ is the average value of the measured temperatures measured at the positions corresponding to the head and feet of the driver's seat and the passenger's seat, respectively. Learning conditions Outside temperature １０〜-10 to 35 (° C.) Solar radiation ‥‥ 0 to 660 (Kcal / m ² hour) Vehicle speed ‥‥ Idle to 40 (equivalent to Km / h) Output value of temperature estimator 1b indicated by a solid line ( Looking at the change in the estimated value T _N ), the maximum error is 1.9 ° C., the average of the absolute value of the error is 0.5 ° C., and the mixed door opening is used as an input signal (maximum error; 9 ° C., the average of the absolute value of the error; 0.83 ° C.), and has good followability to the change in the vehicle interior temperature T ₀ indicated by a circle in FIG. This is because not only the factor used as the input signal is properly selected, but also the temperature data of the outlets other than the outlets set in the outlet mode is set to a predetermined value (for example, 25
° C), unnecessary temperature fluctuations are eliminated, and the estimated value _TN of the temperature in the vehicle compartment can be obtained more accurately. The broken line in FIG.
s (internal temperature sensor) 6.

Next, a method of controlling the temperature in the cabin of the vehicle air conditioner will be described with reference to a control flow of FIG. The control logic 1a extracts a control factor serving as an input signal of the temperature estimator 1b from each input control factor including an environmental factor such as an outside air temperature and a state of an air conditioner such as a conditioned air blowing mode, and estimates the temperature. Output to the container 1b. The temperature estimator 1b includes the extracted control factor, Inc. s and a measuring temperature T _i from (inside air temperature sensor) 6 as an input signal,
The estimated value T of the cabin temperature by the neural network
_N is obtained and output to the control logic 1a. Control logic 1a shows the above from the control factors and the estimated value T _N, as vehicle interior temperature T ₀ becomes the target temperature T _Z that is set in the temperature setting unit 4, a blower duty ratio (%), etc. air duct 2
The feedback control of the temperature in the vehicle interior is performed on the control elements according to the preset control logic 1a. That is, the temperature in the vehicle interior is measured by the inside air temperature sensor 6 and then input to the temperature estimator 1b and fed back to the control logic 1a as the estimated value _TN . Thus, the control logic 1a, not the measured temperature T _i from the inside air temperature sensor 6, the temperature estimator 1 is very close to the vehicle interior temperature T ₀
since the control with the feedback value estimates T _N of cabin temperature from b, it is possible to make the temperature of the passenger compartment accurately and quickly the target temperature T _Z.

Embodiment 2 In the first embodiment, the sigmoid function shown in equation (1) is used for the neural network. By approximating this sigmoid function with a plurality of straight lines, the number of required memories of the temperature estimator 1b can be reduced, Since the calculation time can be greatly reduced, the temperature in the vehicle compartment can be more quickly brought to the target temperature Tz. That is, when the sigmoid function shown in the equation (1) is programmed by, for example, an 8-bit built-in microcomputer, an exponential function and a floating-point arithmetic library are required to maintain accuracy. growing. On the other hand, since the linear function does not include division, there is no need for a floating-point operation, so that the operation accuracy can be maintained even when an integer operation is used, and the ROM capacity and the calculation time can be reduced. Therefore, when the neural network learns, the sigmoid function of the above equation (1) is used. When the learned network is programmed in an actual embedded microcomputer, the sigmoid function of the above equation (1) is used as shown in FIG. Is reduced by, for example, 17 straight lines (linear functions) having an error within ± 0.005 as a substitute for the sigmoid function of the above equation (1), thereby reducing the ROM capacity and calculation time. It is possible to maintain the calculation accuracy.
FIG. 7 is a diagram showing an error between the sigmoid function of the equation (1) and a function obtained by approximating the sigmoid function with the linear function. It can be seen that the magnitude of the error is within ± 0.005 in the entire input range. .

In the above example, the sigmoid function is approximated to 17 straight lines such that the error is within ± 0.005. However, if the error is within 0.03 (± 3)%, there is no practical problem. . In addition, the approximated function does not necessarily need to be a broken line, and in some cases, it may be better to make the function discontinuous in order to reduce the number of straight lines (the number of divisions of the input range) and increase the calculation speed. In the first and second embodiments, the case where the logarithmic sigmoid function of the equation (1) is used has been described. However, other types of sigmoids such as the tanh-type sigmoid function shown in the following equation (2) are used. Of course, functions may be used. y _j = (tanh (x _j ) +1) / 2 ‥‥‥ (2)

[0018]

As described above, in the vehicle air conditioner according to the first aspect, the input signal of the neural network receives the detected temperature of the vehicle interior temperature detecting means, the outside temperature and the amount of solar radiation as environmental factors. In addition to the air condition, the air outlet mode, the air outlet temperature, and the air flow rate, and the temperature data of the air outlets other than the air outlets set in the air outlet mode are fixed to predetermined values, so that unnecessary temperature The fluctuation is eliminated, and the estimated value of the temperature in the vehicle compartment can be accurately obtained. Therefore, the temperature in the vehicle compartment can be quickly set to the target temperature.

The vehicle air conditioner according to the present invention adjusts the temperature and air volume of the conditioned air to be blown into the vehicle cabin by using the estimated value close to the actual room temperature as a feedback value to reduce the temperature in the vehicle cabin. Since the temperature is controlled, accurate and quick temperature control can be performed.

In the vehicle air conditioner according to the third aspect, the teacher signal used for learning the neural network is an average temperature at a position corresponding to the head and foot of the driver's seat and the passenger's seat. The difference between the estimated value of the vehicle interior temperature and the actual vehicle interior temperature can be made extremely small.

According to the vehicle air conditioner of the present invention, the input state of the teacher signal used for learning the neural network is normalized from 0.02 to 0.98, so that the learning efficiency of the learning of the neural network is improved. And safety can be improved.

According to a fifth aspect of the present invention, the sigmoid function used at the time of learning of the neural network has an absolute value of an error between an output value in each input range of the linear function and an output value of the sigmoid function of three. %, The number of required memories can be reduced and the calculation time can be greatly reduced by approximating the input range and the function of the coefficient of the linear function. Target temperature.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a configuration of a vehicle air conditioner according to a first embodiment of the present invention.

FIG. 2 is a diagram illustrating a configuration of a neural network of a temperature estimator according to the first embodiment of the present invention.

FIG. 3 is a diagram showing a sigmoid function used in a neural network.

FIG. 4 is a diagram illustrating a relationship between an estimated value of a vehicle interior temperature and an actually measured value by a temperature estimator according to the first embodiment of the present invention.

FIG. 5 is a diagram showing a control flow of the vehicle air conditioner according to the first embodiment of the present invention.

FIG. 6 is a diagram illustrating an approximation method of a sigmoid function according to the second embodiment of the present invention.

FIG. 7 is a diagram showing an error between a sigmoid function and a linear function.

FIG. 8 is a diagram showing a control flow of a conventional vehicle air conditioner.

FIG. 9 is a diagram showing a configuration of a conventional neural network of a vehicle air conditioner.

[Explanation of symbols]

Reference Signs List 1 control device, 1a control logic, 1b temperature estimator (by NN) 2 air duct, 3 setting panel, 4 temperature setter, 5
Drive device, 6 Inside temperature sensor, 7 Outside temperature sensor, 8
Solar radiation sensor.

Claims (5)

[Claims] 1. A vehicle interior temperature detecting means for detecting a temperature in a vehicle interior, and an estimated value of a temperature in the vehicle interior using the detected temperature of the vehicle interior temperature detecting means, an environmental factor of the vehicle and a state of an air conditioner as input signals. A temperature estimator configured as a neural network having an output value of: a vehicle air conditioner that controls the temperature in the vehicle cabin based on the estimated value, wherein the environmental factors are the outside air temperature and the amount of solar radiation, The air conditioner for a vehicle is characterized in that the states (1), (2), and (3) are a blowing mode, a blowing temperature, and a blowing air volume, and temperature data of the blowing ports other than the blowing ports set in the blowing mode are fixed to predetermined values. 2. The vehicle air conditioner according to claim 1, wherein the temperature in the vehicle interior is controlled using the estimated value as a feedback value. 3. A four-point average temperature at a position corresponding to a head and a foot of a driver's seat and a front passenger's seat, respectively, as a teacher signal used in learning of the neural network. Vehicle air conditioner. 4. The air conditioner for a vehicle according to claim 1, wherein the input state of the teacher signal used in learning of the neural network is normalized from 0.02 to 0.98. 5. A sigmoid function used for a neural network is constituted by a linear function having coefficients different depending on an input range, and an error between an output value in each input range of the linear function and an output value of the sigmoid function. The vehicle air conditioner according to claim 1, wherein the input range and the coefficient of the linear function are approximated by a function in which an absolute value is within 3%.