WO2022000010A1 - Model-predictive control of a power converter - Google Patents

Abstract

In order for a thermal load of a capacitor of a DC intermediate circuit of a current converter to be reduced, a model-predictive control of the current converter (1) is proposed, including a cost function that contains a cost term Jd(xp), said cost term minimizing an AC component of an electric intermediate circuit current (id) flowing across the at least one capacitor (Cdc) of the DC intermediate circuit (12).

Description

Model predictive control of a converter

The present invention relates to a method for regulating a converter, preferably a multi-stage converter, which supplies an electrical load with electrical energy, a switching stage having a number of switching branches being connected to a DC voltage intermediate circuit with at least one capacitor, a plurality of semiconductor switches in each switching branch are connected in series and each switching strand is connected in parallel to the DC voltage intermediate circuit, whereby an output voltage of the converter is generated by switching the semiconductor switch of the switching stage of the converter, and the switching positions of the semiconductor switches of the switching stage to be set as manipulated variables of the control of the converter in each time step of the control can be determined with a model predictive regulation by regulating a cost function as a function of a time curve of at least one in each time step of the regulation the controlled variable of the converter and / or the electrical load, which is dependent on the manipulated variables, is _{optimized over a predetermined prediction horizon N p} > 1 in order to determine the optimal manipulated variables that are set on the semiconductor switches of the switching stage. The invention also relates to a converter with model predictive control.

An inverter (DC / AC converter) is a power converter that converts direct voltage (DC) from a direct voltage source into single or multi-phase alternating voltage (AC). A DC / DC converter is a power converter that converts a direct voltage into another direct voltage. The DC voltage to be converted in the converter is usually provided by a DC voltage intermediate circuit which is connected to a DC voltage source. The DC voltage source can be an energy store, but also the output of an AC / DC or DC / DC converter. In such a case, the converter can also be implemented as a rectifier for converting an alternating voltage into a direct voltage. Such converters use a switching stage with a number of switching strings, for example in the form of half bridges, with semiconductor switches, such as transistors, for conversion. The semiconductor switches of the switching stage are controlled in such a way that the desired output voltage of the converter is generated. Inverters are used, for example, to operate an electric motor, such as an asynchronous machine, in that the inverter generates the electric drive voltages for the drive windings of the electric motor. Rectifiers or DC voltage converters are used, for example, to generate a required DC voltage output voltage, for example in a battery tester to generate an electrical load for a battery, or a battery emulator to simulate a battery. A high switching frequency of the semiconductor switches in the switching stage of the inverter is advantageous because it can increase the efficiency of the converter. Fast switching operations of the switching elements can reduce switching losses and enable high switching frequencies. In particular, semiconductor switches with a large band gap, for example based on gallium nitride (GaN) or silicon carbide (SiC), allow high switching frequencies. On the other hand, however, the shorter switching times increase the loads on the passive electrical switching elements of the converter, such as switching elements in filters, due to the higher steepness of voltage and current edges that occur over time in the voltage and current. The electromagnetic emissions of the converter are also increased due to the higher switching frequencies and shorter switching times. These disadvantages can be reduced by using special topologies of the inverter, such as multi-stage converters (so-called multi-level converters) or converters with several interleaved strings per phase (so-called interleaved converter).

Multi-stage converters use topologies that enable more than two voltage levels at the output of each half-bridge of the switching stage. Converters with several nested strings per phase use several half bridges per phase of the AC voltage or the DC voltage at the output, which are connected to one another. For example, a three-phase converter with six half bridges (two per phase) can be implemented. The semiconductor switches of the interconnected half bridges are switched in a time-staggered manner ("nested" or "interleaved"). Such topologies are well known.

With these topologies, however, the load on the capacitors of the DC voltage intermediate circuit cannot be reduced, especially at high switching frequencies. The switching of the semiconductor switching elements in the switching stage is expressed in an alternating component (ripple current) in the current of the DC voltage intermediate circuit. This alternating component loads the capacitors of the DC voltage intermediate circuit, which is particularly noticeable in the heating of the capacitors. This thermal load limits the performance limit (in terms of the available electrical power) and reduces the service life of the capacitors or makes correspondingly large capacitors necessary.

The semiconductor switches of the switching stage of the converter can be switched at least once per time step of the regulation of the converter (typically in the 10ps to 100ps range). This results in a limited number of possible switching states of the semiconductor switches of the converter. This limited number of possible switching states can be used for control, for example in a model predictive control with a finite control quantity (Finite Control Set Model Predictive Control (FCS-MPC)). A model predictive control (MPC) generally calculates the future optimal temporal progression (usually in discrete time form) of a manipulated variable (also as a vector of several manipulated variables) with which the controlled system (e.g. an electric motor or an electrical load) is calculated in each time step of the control. is regulated, over a predetermined prediction horizon (which corresponds to a certain period of time). In the case of a converter, the manipulated variables are, for example, the switching positions of the semiconductor switches of the switching stage of the converter. The next future manipulated variable is used to regulate the controlled system and the others are discarded (principle of the receding horizon). This is repeated in every time step of the control. The longer the prediction horizon, i.e. the further into the future, the higher the achievable control quality, but the higher the computational effort, which is why the possible prediction horizon is often limited. In an MPC, a cost function is used as a function of the manipulated variable over the prediction horizon, which is optimized (usually minimized) according to the manipulated variables. The cost function can contain various cost terms that evaluate certain control objectives. Typical control objectives are a minimal deviation of an actual variable from a target variable (for example a torque of an electric motor or an output voltage), a limitation of the number of switching operations of the semiconductor switching elements or a limitation of the output currents or the deviations of the individual output currents from one another (current balancing). By optimizing the cost function, one obtains the optimal temporal course of the manipulated variable over the prediction horizon. An FCS-MPC of a DC / DC converter is described, for example, in WO 2015/154918 A1. With such an FCS-MPC, however, the thermal load on a capacitor of a DC voltage intermediate circuit cannot be reduced.

It is an object of the present invention to reduce the thermal load on a capacitor of a DC voltage intermediate circuit of a converter with a model predictive control.

This object is achieved according to the invention in that the cost function contains a cost term J _d (x) which minimizes an alternating component of an electrical intermediate circuit current flowing via the at least one capacitor of the DC voltage intermediate circuit. The optimal manipulated variables are therefore the manipulated variables that minimize the value of the cost function. By minimizing the AC component of the intermediate circuit current, the load on the intermediate circuit capacitor from high-frequency recharging processes can be reduced. In this way, the power capacity of the converter can be increased with the same intermediate circuit capacitors, because the heating of the intermediate circuit capacitor is reduced. Or it can DC link capacitors with smaller component values are used, which helps to reduce costs and size.

The cost function is advantageously optimized in such a way that the associated time profiles of the controlled variables are determined with a prediction model for the various possible time profiles of the manipulated variables over the prediction horizon and the value of the cost function is determined for each of these time profiles of the controlled variable and that from the time Course of the manipulated variable that is assigned to the temporal course of the controlled variable, which minimizes the value of the cost function, the optimal manipulated variables are determined. This is particularly advantageous with an FCS-MPC.

A cost term that minimizes the alternating component is particularly advantageous in connection with a multilevel converter. Due to the more than two voltage levels at the output of a switching section of the switching stage of the converter, there are more degrees of freedom for the regulation, with which the goal of minimizing the AC component can be achieved without impairing the load regulation (i.e. regulation of the output voltage or the output current). The additional degrees of freedom result from the fact that there are several switching combinations that lead to the same voltage level (which is required for load control) at the output of a switching section. In this way that switching combination can be selected which is advantageous in terms of minimizing the alternating component. This selection is made by the model predictive control.

Since the alternating component of the intermediate circuit current is not a periodic variable, but rather resembles a noise, the alternating component is advantageously evaluated with a statistical measure for the intermediate circuit current, for example as a variance or a standard deviation of the intermediate circuit current. This can be done easily and with little computing effort.

It is advantageous if further cost terms are taken into account in the cost function, with which further control objectives can be pursued. In this case, the cost function contains, in particular, a further cost term which evaluates a deviation of at least one setpoint variable for a controlled variable from the actual value of the associated controlled variable. In this way, the usual control objective that at least one specific control variable of the converter or the associated electrical load follows a predetermined setpoint can be achieved with the model predictive control.

In an implementation of the converter with nested switching branches, for example by means of a nested choke or by simply connecting the switching branches, a further cost term is preferably used in the cost function, which is the differential current as the difference in the electrical currents at the outputs of the Switching branches limited. In this way, on the one hand, current balancing can be implemented in order to keep the difference in the string currents in a desired difference band or to minimize it. If a nested choke is used, this can also prevent the nested choke from becoming saturated.

The differential current can be measured particularly easily with a measuring arrangement in the form of a bridge circuit, in which a first bridge branch with a series connection of two resistors is connected between the outputs of the chokes of the switching sections and a second bridge branch with a series connection of two measuring resistors between the outputs of the chokes of the switching lines is connected, a midpoint between the two resistors of the first bridge branch is connected to a shunt branch of the bridge circuit with a common output terminal between the two measuring resistors of the second bridge branch, and the electrical voltage applied to the two measuring resistors of the first bridge branch, which is proportional to the differential current is measured. In addition, in such a measuring arrangement, a further measuring sensor can be connected in the shunt arm of the bridge circuit, which measures an electrical voltage that is proportional to the total current as the sum of the two phase currents. This means that the total current can be measured just as easily at the same time.

The present invention is explained in more detail below with reference to FIGS. 1 to 7, which show exemplary, schematic and non-limiting advantageous embodiments of the invention. It shows

Fig. 1 the basic structure of a converter,

FIG. 2 shows an example of a single-phase converter with nested switching branches, FIG. 3 shows a nested choke,

4 shows a multi-phase multi-stage converter,

5 shows an MPC of a converter that supplies an electrical load,

6 shows an equivalent circuit diagram of a nested choke and

7 shows a measuring arrangement for measuring the differential current and the

Total current of the phase currents of a converter with nested

Switching lines.

A converter 1, as shown schematically in FIG. 1, generally consists of a switching stage 11 with a number n of switching lines SLn, each switching line SLn having a plurality m of semiconductor switches HSnm. A switching section SLn is usually designed in the form of a half bridge. At least one switching section SLn is provided for each phase of the output voltage v _a. The existing switching lines SLn are connected in parallel to a DC voltage intermediate circuit 12, which consists of at least one _{Intermediate circuit} capacitor C dc exists. The DC voltage intermediate circuit 12 is applied in parallel to the DC voltage input voltage V _dC . Often im

_{DC voltage intermediate circuit} 12 arranged a plurality of serially connected intermediate circuit capacitors C dc, so that the input voltage V _{dC is divided} accordingly on the intermediate _circuit capacitors C dc. _{The string output voltage v sn} , which is usually smoothed in a choke L, is tapped between two semiconductor switches HSnm of a switching string SLn (high-side and low-side switch). The string output voltage v _sn can also be an output voltage v _{a of} the converter 1 at the same time. On the input side (before the DC voltage intermediate circuit 12) and / or on the output side (after the switching stage 11), filter circuits (not shown) consisting of electrical components such as resistors, inductors, capacitors in any connection can be provided.

To regulate the string output voltage v _sn (and thus the output voltage v _a ), the semiconductor switches HSnm are regulated by a control unit 13 (only indicated in FIG HSnm can be set so that the output voltage v _a assumes a desired value, usually given as a time curve. The control unit 13 is therefore given a setpoint value SW of the regulation. The setpoint value SW can also depend on the electrical load 2. For example, in the case of an electrical machine, it is customary to specify a torque or a speed of the machine as a setpoint value. However, a time profile of the output voltage v _a , for example a sinusoidal voltage profile with a specific amplitude and phase (in the case of a polyphase output voltage v _a ) or a specific DC voltage, can also be specified.

FIG. 2 shows a topology of the converter 1 with a plurality of interleaved switching strings SL1, SL2 per phase of the output voltage v _a . The outputs of the switching lines SL1, SL2 routed via chokes L are connected to one another and form the voltage output of one phase of the converter 1. It can also be provided that the outputs are first connected and then routed via a choke L. In the case of several phases of the output voltage v _a , correspondingly more switching branches SLn would be provided, which can also be connected to one another to form nested branches.

2 der verschachtelten Drossel 14) in einer Topologie wie in Fig.2 dargestellt. Man erhält damit bei Verwendung einer verschachtelten Drossel 14 mit einer verschachtelten Strangtopologie einen Mehrstufenstromrichter. Pro Phase der Ausgangsspannung v_a können auch zumindest zwei Schaltstränge SLn über eine verschachtelte Drossel 14 gemäß Fig.3 miteinander verschachtelt werden, womit sich so auch mehrphasige Stromrichter 1 realisieren lassen.The chokes L of interleaved switching branches SLn can be designed as nested chokes 14, as shown in FIG. The individual chokes L are wound with opposing windings on a common core. This allows output voltages v _a to be realized with more than two voltage levels, V for example de (if leakage inductances and losses are neglected

2 of the nested choke 14) in a topology as shown in FIG. When using a nested choke 14 with a nested string topology, a multi-stage converter is thus obtained. For each phase of the output voltage v _a , at least two switching branches SLn can also be interleaved with one another via a nested choke 14 according to FIG.

A multi-stage converter can also be implemented by providing more than two series-connected semiconductor switches HSnm per switching strand SLn, as shown in an exemplary embodiment in the form of a two-phase multi-stage converter in FIG. The power converter 1 shown has two switching branches SLn, each with four semiconductor switches HSnm connected in series. Two semiconductor switches HSnm in each case form a switch group, the branch output voltage v _{sn being} tapped between the two switch groups of a switching branch SLn. The center point between the semiconductor switches HSnm of a switch group is in each case switched to the zero point between the two intermediate _{circuit capacitors C dc of} the DC voltage intermediate circuit 12 via a diode. Instead of the diodes, semiconductor switches, such as transistors, can also be provided. This also allows the string output voltage v _{sn to be} three

drei Spannungspegel ermöglichen. Realize voltage level yj. Topologies are also known that have more than

allow three voltage levels.

It should be noted that there are of course a large number of modifications of the topologies shown and other possible topologies of power converters 1, the specific topology not being essential for the present invention.

The basic principle of a model predictive control (MPC) of a converter 1, which supplies an electrical load 2 with electrical energy, is described with FIG. A prediction model 3 is implemented in a control unit 13, which shows the course of at least one variable x _{p to be} controlled of the controlled system (hereinafter referred to as a controlled variable), for example a state variable of the converter 1 or the electrical load 2, over a predetermined prediction horizon N _p ( k + 1, k + 2, ..., k + N _p in discrete time notation). The time-discrete time step is denoted by k, whereby according to the usual notation k stands for current values, (k + ...) for future values and (k -...) for past values. The smallest prediction horizon is N _p = 1. The time step with which the prediction horizon N _{p is} discretized usually (but not necessarily) coincides with the time step of the regulation, so that the prediction horizon N _p specifies how many time steps the regulation is expected to be in the future. The controlled variable x _p is usually a vector that _{contains various controlled variables x p} that are combined in the vector.

The controlled variable x _p can be dependent on the electrical load 2. In the example of an electrical machine (eg asynchronous motor) as electrical load, the controlled variable could be a torque of the electrical machine, or a magnetic stator flux, or also several different variables. However, state variables of the converter 1 can also be used as control variables x _p , also in addition to the state variables of the electrical load 2. For the present invention, in particular, an electrical intermediate circuit current i _{d flowing in the DC voltage intermediate circuit 12 is used} as the control variable x _p , as will be explained in more detail below executed.

The current output variable y (k) of the regulated converter 1, for example an output voltage v _a or an output current i _a , is available to the prediction model 3. However, the electrical phase currents in the chokes L of a nested choke 14 according to FIG. 3 can also be used as the output variable y (k). The output variable y (k) also depends on the prediction model 3. The output variable y (k) can also be a vector of different sizes. The output variable y (k) is usually measured and is available as a measured value. In addition, other current state variables x (k) of the converter 1, for example a phase current i _sn , and / or the electrical load 2, for example a speed of an electric motor, can also be available, for example again in the form of measured values.

Which output variables y (k) and which state variables x (k) are required depends on the topology of the converter 1 and on the electrical load 2, and also on their mathematical modeling by the prediction model 3.

The prediction model 3 determines the controlled variables x _p for various input variables u _p . The input variables u _{p of} a converter 1 are the switching positions S of the semiconductor switches of the switching stage of the converter 1 and thus the manipulated variables of the control. Depending on the topology of the converter 1, there is a certain number of semiconductor switches HSnm in the switching stage 11 and thus a finite possible number of different combinations of the switching positions. The number of possible input variables u _p is thus limited by the number of semiconductor switches HSnm of the switching stage. If N denotes the number of possible combinations of switch positions, then, up to N calculated controlled variables x _p as a function of the N input values u _p (indicated in Figure 5). For a prediction horizon N _p = 1 there are thus N vectors of the controlled variables x _p , but for a prediction horizon N _p > 1 there are already up to N ^Np possible vectors of the controlled variables x _p . From this it can be seen that the computational effort for the calculation of the controlled variables x _p and for the subsequent optimization with the length of the Prediction horizon N _p rises sharply. However, it is of course conceivable not to use all possible input variables u _p , but rather to reduce the number N or N ^Np of possible input variables u _p in order to reduce the possible manipulated quantity.

For example, due to the design of the converter 1, certain impermissible or impossible combinations of the switch positions (which can result from the specific topology of the converter 1 and / or the electrical load 2) can be excluded.

In an embodiment of a three-phase converter 1, for example according to FIG. 2 (also with a choke as in FIG. 3) with two additional phases, in each time step of the regulation N = 64 (2 ⁶ ) possible combinations of switching positions would result and thus also 64 associated possible controlled variables x _p . With a prediction horizon N _p > 1, however, 64 ^Np possible combinations of switch positions.

A cost function J is implemented in an optimization unit 4, which is a function of the time course of the controlled variables x _p over the prediction horizon N _p , that is to say J (x _p ). With a prediction horizon N _p = 1, the course over time is naturally reduced to the next time step. The optimization unit 4 is also supplied with setpoint variables SW of the control system. Which setpoint values SW are required can in turn depend on the design of the converter 1, on the electrical load 2 and / or on the regulation objective and can vary.

The cost function J is optimized in the optimization unit 4 (usually minimized, but maximization is also possible). This essentially means that the value of the cost function J is determined with all the determined control variables x _p. The optimal controlled variables x _{p, opt} are those controlled variables x _p that result in the minimum (or maximum) value of the cost function J. The input variables u _p that are assigned to these optimal controlled variables x _{p, opt} are the optimal input variables u _opt that are used to control the converter 1. These input variables u _ot are set in the next time step (k + 1) of the control with the converter 1.

If the prediction model 3 is used to determine a time profile of the controlled variables x _p over the prediction horizon N _p , then the value of the cost function J is determined with this time profile. From that temporal course of the controlled variable x _p over the prediction horizon N _p that yields the smallest (or largest) value of the cost function J, the controlled variable of the temporal course that is closest to the current point in time is used as the optimal controlled variable x _{p, opt} . _{This again results in} the associated optimal input variables u opt.

The control unit 13, the optimization unit 4 and the prediction model 3 can be implemented as microprocessor-based hardware, for example as a computer or digital signal processor (DSP), on the corresponding software for performing the respective function is carried out. The control unit 13, the optimization unit 4 or the prediction model 3 can also be an integrated circuit, for example an application-specific integrated circuit (ASIC) or a field programmable gate array (FPGA), also with a microprocessor. The control unit 13, the optimization unit 4 and the prediction model 3 can, however, also be implemented as an analog circuit or an analog computer. Mixed forms are also conceivable. It is also possible that different functions are implemented on the same hardware, for example the optimization unit 4 and / or the prediction model 3 as software on the control unit 13, which is implemented as microprocessor-based hardware or an integrated circuit.

The cost function J (x _p ) usually contains at least one cost term Jsw (x _P ) in a known manner, which evaluates a deviation of at least one specified target variable SW for a controlled variable x _p from the associated controlled variable x _p , i.e. from the actual value of the controlled variable x _p . The actual value can be measured, for example, or can also be calculated by the prediction model 3. For example, the deviation can be as simple

Difference or the amount of the difference can be evaluated, i.e. as a ^ norm. For the

Stability of the regulation, it can be advantageous to express the deviation as the square of the

(^₂-Norm) zu bewerten. Auch andere Normen sind möglich. Difference (SW -x

(^ ₂ norm). Other standards are also possible.

If the controlled variable x _{p is present} over a prediction horizon N _p > 1, then the cost term Jsw can be the sum of the individual cost terms Jsw at the respective times

(^C _{R Ϊ}) Anstelle einer derartigen einfachen Linearkombination der Kostenterme Jsw zu den jeweiligen Zeitpunkten könnte natürlich auch eine andere Verknüpfung verwendet werden, insbesondere auch eine nichtlineare Kombination der Kostenterme Jsw zu den jeweiligen Zeitpunkten. Np k + 1, k + 2, ..., k + N _p, i.e. J _SW ( ^X _P ) =

( ^C _{R Ϊ} ) Instead of such a simple linear combination of the cost terms Jsw at the respective points in time, another link could of course also be used, in particular also a non-linear combination of the cost terms Jsw at the respective points in time.

In the case of an electric machine as the electric load 2, for example, the torque T _{e of} the electric motor or a speed n _{e of} the electric motor can be specified as a setpoint value SW. In the prediction model 3, the actual values of the torque or the speed are then determined as controlled variables x _p (depending on the possible input variables u _p ). However, a stator flux y _d can also be used as the setpoint value SW, for example in order to avoid magnetic saturation of the stator teeth. In the prediction model 3, the actual values of the stator fluxes would then be determined as a controlled variable x _p as a function of the possible input variables u _p . Of course, several different setpoints SW can also be specified and _{taken into account in the cost function J (x p} ) or in the cost term Jsw (x _P ), for example as the sum of the individual differences in the form of the ^ norm or ^ ₂ norm. The cost function J (x _p ) contains a cost term J _d (x) which evaluates _{the alternating component i d, A} c of the total intermediate circuit current i _{d flowing in the DC voltage intermediate circuit 12.} The DC link current i _d flowing through the DC link capacitors C _DC of the DC intermediate circuit 12. In an embodiment according to Figure 2 with split intermediate circuit with a plurality of intermediate circuit capacitors C _dc, the entire intermediate circuit current is i _d composed of the current flowing in the upper and lower branch of the intermediate circuit electricity. The intermediate _{circuit current i d is accordingly the sum of the electrical phase currents i sn} taken from the intermediate circuit 12 by the switching branches SLn of the switching stage 11. Because of the high switching frequencies of the semiconductor switches HSnm of the switching stage, the intermediate circuit current i _d can also change at high frequencies. This leads to high-frequency recharging processes on the intermediate circuit capacitors C _dc and thus to a thermal load on the intermediate _{circuit capacitors C dc} . According to the invention, this alternating component i _{d, A} c should be minimized, which is done _{with the cost term J d} (x) in the cost function J (x _p).

The alternating component i _{d, A} c of the intermediate _{circuit current i d} as a controlled variable x _p can of course be evaluated in the most varied of ways. A very simple approach would be to evaluate the distance between the maximum and minimum amplitudes of the intermediate circuit current i _d , possibly also as an amount or as a square. Since the intermediate circuit _{current i d is} not a periodic variable, statistical measures can also be used for evaluation. A statistical measure is applied to the intermediate circuit _current i d, for example a variance or a standard deviation. A mean value of the intermediate circuit _current i d over time could be used as the expected value and the deviation of the intermediate circuit _current i d from the expected value could be assessed, which leads to the variance. The standard deviation then results from the square root of the variance.

The cost term J _d (x) can then, for example, in the form

Length N _d , which corresponds to the effective value of the alternating component i _{d, A} c of the intermediate circuit _current i d in this time window. For this purpose, the prediction model 3 would determine the intermediate circuit current i _d as the controlled variable x _p . The number N _d of past current values could be stored in a memory, for example the control unit 13 or the prediction model 3. The mean value i _{d, v of} the intermediate circuit _current i d can can also be determined over a sliding time window of length N _v (where N _d = N _v can, but does not have to be), for example as j).

If the controlled variable x _{p is present} over a prediction horizon N _p > 1, then the cost term J _d can be the sum of the individual cost terms Jsw at the respective times k + 1,

(x_{p i}) Anstelle einer derartigen einfachen Linearkombination der Kostenterme J_d zu den jeweiligen Zeitpunkten könnte natürlich auch eine andere Verknüpfung verwendet werden, insbesondere auch eine beliebige nichtlineare Kombination der Kostenterme J_d zu den jeweiligen Zeitpunkten. Np k + 2, k + N _p , _i.e. J d (x _p ) =

(x _pi ) Instead of such a simple linear combination of the cost terms J _d at the respective points in time, another link could of course also be used, in particular any non-linear combination of the cost terms J _d at the respective points in time.

= max in der Form formuliert

werden, mit einem vergebenen Grenzdifferenzstrom i_sat, mit dem Sättigung sicher vermieden wird. Alternativ kann der Kostenterm auch in der Form Further cost terms can also be used in the cost function J (x _p ). In an implementation of the converter 1 with a nested choke 14 (FIG. 3), undesirable saturation in the core of the choke 14 can occur, in particular if the differential currents (i _si - L2) of the nested choke 14 via the chokes L flowing currents become too large. The phase currents i _si , i _S 2 are the controlled variables x _p . In such an implementation, therefore, it may be advantageous to use a further cost term J _sat (x) which limits the differential currents (i _si − i _S 2). It must be taken into account here that there can be such a restriction for each phase of the converter 1 with such a nested choke 14. The restriction can be set with the maximum norm

= max formulated in the form

are, with an assigned limit differential _{current i sat} , with which saturation is reliably avoided. Alternatively, the cost term can also be in the form

stärkerem Messrauschen als vorteilhaft herausgestellt hat. Darin wird der Parameter a vorgegebenen, vorzugsweise kleiner als eins gewählt, beispielsweise a=0,8. Eine aktive Minimierung der Differenzströme wird erreicht, wenn a=0 gesetzt wird, während bei a > 0 der Differenzstrom nur außerhalb des durch i_sat festgelegten Bandes minimiert wird. Der Parameter A_sat wird vorteilhafterweise groß genug gewählt, sodass diese Kostenfunktion J_sat bei Verletzung der Bedingung stark in den Wert der Kostenfunktion J(x_p) eingeht, sodass Eingangsgrößen u_p, die zu einer Sättigung führen können, sicher ausgeschlossen werden.J sat to be formulated, which is particularly important

has shown greater measurement noise to be advantageous. The parameter a is specified therein, preferably selected to be less than one, for example a = 0.8. An active minimization of the differential currents is achieved when a = 0 is set, while with a> 0 the differential current is only minimized outside the band defined _{by i sat.} The parameter A _sat is advantageously chosen large enough that this cost function J _{sat is} heavily included in the value of the cost function J (x _p ) if the condition is violated, so that input variables u _p that can lead to saturation are reliably excluded.

If the controlled variable x _{p is present} over a prediction horizon N _p > 1, then the cost term J _sat can be the sum of the individual cost terms J _sat at the respective times Np k + 1, k + 2, k + Np, that is, J _sat {x _p ) = ^' ^ _J J _sat {x _pi ) Instead of such a simple linear combination of the cost terms J _sat at the respective times, a Other links can be used, in particular also a non-linear combination of the cost terms J _sat at the respective points in time.

The differential _{currents (i si} - i _S 2) of a converter with several interleaved switching lines, also in the form of a multi-stage converter, i.e. the difference between the _{switching lines connected via the chokes L of the phase currents i si} , i _S 2, can be included in a cost term J _bai (x) of the cost function J are taken into account, for example for power balancing. The difference can in turn in the cost term as a simple difference or as a norm of the difference (^ -norm or ^ ₂ -norm)

The differential currents (i _si - i _S 2) occur in this cost term J _sat. These can advantageously be determined, specifically measured, with a measuring arrangement 15 in the form of a bridge circuit according to FIG.

A first branch of the bridge connects the outputs of the two chokes L of the nested choke 14 with a series circuit of resistors RMI, RM2. This connection can consist of two symmetrical resistors R _M 2 which are connected in series. A second bridge branch connects the outputs of the two chokes L of the nested choke 14 via preferably low-resistance measuring resistors R _M 3 (usually a few mQ) to a common output terminal BP of the respective phase of the converter 1. The measuring resistors R _M 3 should have significantly lower resistance values than the Resistances R _M 2 of the first bridge branch to ensure that the current in the first bridge branch is negligibly small. Alternatively, a compensating factor can be implemented in a higher-level evaluation logic according to the current distribution. In such a measuring arrangement 15, _{a voltage sensor MS1 can be switched via the resistors R MI of} the first bridge branch, the voltage thus measured being proportional to the differential current (i _si - i _S 2). In a shunt arm that connects the midpoint AP of the first bridge arm to the common output terminal BP of the second bridge arm, another voltage sensor MS2 can be used to measure a voltage that is proportional to the total current (i _si + i _S 2).

An advantage of this measuring arrangement 15 is that the input voltage ranges of the respective voltage sensors MS1, MS2 can be selected separately for the expected differential or total currents by selecting the resistance values for the resistance values RMI, RM3. In addition, the measured voltage in the first bridge branch can be over an additional voltage divider can be scaled down via the resistors RMI before it is fed to the voltage sensor MS1.

Such a measuring arrangement 15 can of course also be used in the case of nested switching branches SL1, SL2 of a converter 1 without nested choke 14 (for example as shown in FIG. 2).

The individual cost terms of the cost function J (x _p ), or just certain cost terms, can also be weighted by a weighting factor A. The weighting factor can then be used to give preference to a specific control objective over others, for example a higher control accuracy through a high weighting factor Asw for the cost term Jsw and a low weighting factor A _d for the cost term J _d , or vice versa.

The cost function J (x _p ) can then be defined, for example, as follows, with optional terms being given in square brackets. Usually, at least the cost function Jsw is also taken into account.

This cost function J (x _p ) is optimized (minimized or maximized), that is, that controlled variable x _{p is used} as the solution of the optimization which gives the optimal (smallest or largest) value of the cost function J (x _p ). This controlled variable x _p is assigned an input variable u _p which led to the controlled variable x _p . This input variable is used as the optimal input variable u _opt for the regulation and is used in the next time step (k + 1) in order to control the converter 1 with it.

In the optimization, boundary conditions g (x _p ) can also be taken into account, which are usually formulated as inequalities, for example g (x _p ) <g _max.

The implementation of the optimization will take a certain amount of time, which can also be longer than a time step of the control. This can have the consequence that the determined optimal input variable u _opt no longer relates to measured values at time (k), but to previous measured values at time (k-1). The controller lags behind the controlled system by one time step. This can have a negative impact on the control quality. Time compensation can therefore be provided.

For time compensation, the input variable u _p (k) at the current point in time (k), which is known, can be used _{to calculate the predicted controlled variable x p} (k + 1) with the prediction model 3. With this predicted controlled variable x _p (k + 1) and the N or up to N ^Np possible input variables u _p (k + 1) that can be applied at the point in time (k + 1), the prediction model 3 can be used resulting N or up to N ^Np Controlled variables x _p (k + 2) can be determined. With these controlled variables x _p (k + 2), the optimization can be carried out as described above in order to determine the optimal input variable u _ot (k + 1), which is then applied to the converter 1 for the next time step (k + 1).

For the prediction model 3, the converter 1 in the respective topology and the electrical load 2 are to be mathematically described by system equations, as exemplified by the example of a converter 1 according to FIG. 2 with a nested choke 14 according to FIG. 3 and an asynchronous motor as electrical load 2 is explained.

The nested choke 14 can be modeled with an equivalent circuit diagram as shown in FIG. 6, with the leakage inductances I_i _s , I_2 _s known for the choke 14, the known mutual inductance M and the known loss resistances R1, R2. the

Input voltages vio, V20 can depending on the switch position of the semiconductor switch HSnm die

die folgenden Systemgleichungen.

Accept values. Applying the knot and mesh rule one obtains

the following system equations.

From the above system equations it can be seen that the output voltage v _{a can} also assume the value zero. For an implementation of the prediction model 3 on a microprocessor or an integrated circuit (ASCI, FPGA), a discrete-time formulation of the system equations is required. This can be obtained simply by applying the Euler forward method by adding the differential quotients dx / dt in the system equations by [x (k + 1) -x (k)] / T with the specified sampling time T and the instantaneous values x (t ) can be replaced by x (k). Of course, another discretization method can also be used. This leads to the following discrete-time system equations.

The above system equations naturally result for each phase of the output voltage v _a , so that a total of nine system equations result for a three-phase converter 1.

The system equations are supplemented by the system equations of the electrical load 2. These are well known for generally used electrical loads 2, such as an asynchronous machine.

The torque T _{e of} an asynchronous machine can be described, for example, with the following system equation in the stator-fixed ab coordinate system.

The number of pole pairs z _p , the mutual inductance M _sr , the stator inductivity L _s , the rotor inductance L _r and the scatter coefficient c are known parameters of the asynchronous machine. Y describe the magnetic fluxes in the rotor (index r) and in the stator (index s) in the aß coordinate system. The torque T _{e of} the electric motor is determined from this. The magnetic fluxes in the stator and rotor of the asynchronous machine can be given by the following equations, whereby only the discrete-time equations are written.

und /_s = /_{sa sß}^ Der Strom i_a in den obigen Gleichungen ist der Wcklungsstrom (der den Ausgangsströmen i_a der Phasen des Stromrichters 1 entspricht) im ab-Koordinatensystem und stellt die Verbindung der Systemgleichungen des Stromrichters 1 und der elektrischen Last 2 her. Die Transformation der Ausgangsströme i_a der Phasen des Stromrichters 1 in den Strom im aß- 6 Koordinatensystem erfolgt bekanntermaßen mittels der bekannten Clarke-Transformation

Therein, the rotor resistance R _r is again a known parameter of the asynchronous machine, T is the sampling time (the not be the same need as the sampling time for the above system equations of the power converter 1), oo _r is the angular velocity (which can be measured), and the magnetic fluxes y _G , ijJ _s in the rotor and stator contain the fluxes in the ab coordinate system _{, i.e. y g} = \ y _ga

and / _s = / _{sa sß} ^ The current i _a in the above equations is the winding current (which _{corresponds to the output currents i a of} the phases of the converter 1) in the ab coordinate system and represents the connection of the system equations of the converter 1 and the electrical load 2 here. The transformation of the output _{currents i a of} the phases of the converter 1 into the current in the aß- 6 As is known, the coordinate system takes place by means of the known Clarke transformation

The system equations of the converter 1 and the electrical load 2 can be dx (t)

- ~ - = A x (t) + B u (t) more compact than state space model in the form dt or in y (t) = Cx _p (t) x (k + l) = (A + l) -Tx _{n (} k) + BTu _{n (} k) discrete-time notation in the form (with Euler- y (k) = CTx _p (k)

Forward method discretized, other discretization methods are also conceivable) with the unit matrix I and the sampling time T, the system matrices A, B, C result from the concrete modeling. The vector of the state variables x _p contains, for example, the magnetic fluxes y _G , ijJ _s and the _{phase currents i si} , i _S2 (of course over all phases of the converter 1), both also in the ab coordinate system, and the vector of the output variables y contains that, for example Torque T _{e of} the electric motor and the output _{currents i a} (or the individual phase currents i _si , i _S2 ). _{The intermediate circuit current i d} can of course also be determined from the phase currents i _si , i _S2, possibly together with the known switching positions of the semiconductor switches HSnm (input variables u _p).

Claims

Claims 1. A method for regulating a converter (1), preferably a multi-stage converter, which supplies an electrical load (2) with electrical energy, wherein a switching stage (11) with a number of switching strands (SLn) with a DC voltage intermediate circuit (12) with at least one Intermediate circuit capacitor (Cdc) is connected, with a plurality of semiconductor switches (HSnm) being connected in series in each switching strand (SLn) and each switching strand (SLn) being switched in parallel to the DC voltage intermediate circuit (12), whereby the switching of the semiconductor switches (HSnm) results in the Switching stage (11) of the converter (1) an output voltage (v _a ) of the converter (1) is generated, and _{the switching positions of the semiconductor switches (HSnm) to be set as the manipulated variables u p of} the control of the converter (1) in each time step of the control Switching stage (11) can be determined with a model predictive control by a cost function in each time step of the control Function J (x _p ) is optimized as a function of a time curve of at least one controlled variable x _{p of} the converter (1) and / or the electrical load (2) _{to be regulated and dependent on the manipulated variables u p over a predetermined prediction horizon N p} > 1 , in order to determine the optimal manipulated variables u _ot which are set on the semiconductor switches (HSnm) of the switching stage (11), characterized in that the cost function J (x _p ) contains a cost term J _d (x) which has an alternating component i _{d, A} c of an electrical intermediate circuit current (id) flowing via the at least one intermediate circuit capacitor (Cdc) of the DC voltage intermediate circuit (12) is minimized. 2. The method according to claim 1, characterized in that a statistical measure of the intermediate _{circuit current (i d} ) is used _{as the cost term J d (x).} 3. The method according to claim 2, characterized in that a mean value of the intermediate _{circuit current (i d} ) over a sliding time window with length N _{v is determined} as the expected value of the intermediate _{circuit current (i d} ) and a variance over a sliding one as the cost term J _{d (x)} Time window with the length N _{d is used} as the deviation of the intermediate circuit current (i _d ) from this expected value or a standard deviation is used as the square root of this variance. 4. The method according to claim 1, characterized in that the cost function J (x _p ) contains a further cost term Jsw (x) which evaluates a deviation of at least one target variable (SW) for a controlled variable x _p from this associated controlled variable x _p . 5. The method according to claim 1, characterized in that in an implementation of the converter (1) with a nested choke (14) or in one Implementation of the converter (1) with nested switching lines (SLn), in which the outputs of two switching lines (SL1, SL2) of the switching stage (11) of the converter (1) are connected to one another to form a voltage output of one phase of the converter (1), another is used cost-term J _sat (x), J _bai (x) in the cost function J (x _p) of the differential current (i _si - i _S 2) as the difference of the electric phase currents (i _si, i _S 2) to the Outputs of the switching lines (SL1, SL2) are limited. 6. The method according to claim 5, characterized in that the cost term J _sat (x) in the form or in the form

^ sat is used with the maximum norm

= max | x _; | , a predetermined parameter a and a predetermined parameter A _sat . i = l, ..., n 7. The method according to claim 5 or 6, characterized in that the differential current (i _si - i _S 2) is measured with a measuring arrangement (15) in the form of a bridge circuit, in which a first bridge branch with a series circuit of two resistors (R _M 2), whereby the two resistors (R _M 2) are connected between the outputs of chokes (L) of the switching lines (SL1, SL2) and a second bridge branch with a series connection of two measuring resistors (R _M 3) between the outputs of the chokes (L) of the switching strands (SL1, SL2) are switched so that a midpoint (AP) between the resistors (R _M 2) of the first bridge branch with a shunt branch of the bridge circuit with a common output terminal (BP) between the two measuring resistors (R _M 3 ) of the second bridge branch is connected, and that a voltage sensor (MS1) is used to measure the _{electrical voltage applied to the two resistors (R M} 2) of the first bridge branch, which voltage is proportional to the difference current (i _si - i _S 2). 8. The method according to claim 7, characterized in that a further voltage sensor (MS2) is connected in the shunt arm, which measures the electrical voltage between the center point (AP) and the output terminal (BP), which is proportional to the total current (i _si + i _S 2) is the sum of the line currents (i _si , i _S 2) of the switching lines (SL1, SL2). 9. The method according to any one of claims 1 to 8, characterized in that with a prediction model (3) for different possible time courses of the manipulated variables u _p over the prediction horizon N _p , in particular up to N ^Np possible time courses with N as the number of possible Combinations of the switching positions of the semiconductor switches (HSnm) of the switching stage (11) of the converter (1), the associated time profile of at least one controlled variable x _{p is} determined and for each of these time profiles of the controlled variable x _p the Value of the cost function J (x _p ) is determined and that from the temporal course of the manipulated variable u _p which is assigned to the temporal course of the controlled variable x _p , which minimizes the value of the cost function J (x _p ), the optimal manipulated variables u _{ot are} determined . 10. The method according to claim 9, characterized in that from the temporal course of the manipulated variable u _p , which minimizes the value of the cost function J (x _p ), the temporally following manipulated variable is determined as the optimal manipulated variable u _ot . 11. Converter, preferably multi-stage converter, for supplying an electrical load (2) with electrical energy, with a control unit (13) for regulating the converter (I), a switching stage (11) with a number of switching branches (SLn) being provided in the converter (1), which is connected to a DC voltage intermediate circuit (12) with at least one intermediate circuit capacitor (Cdc), with each switching branch (SLn) a plurality of series-connected semiconductor switches (HSnm) is provided and each switching strand (SLn) is connected in parallel to the DC voltage intermediate circuit (12), the converter (1) by switching the semiconductor switches (HSnm) of the switching stage (II) an output voltage (v _a ) is generated, and the control unit (13) uses the switching positions of the semiconductor switches (HSnm) of the switching stage (11) to be set _{as manipulated variables u p of the regulation of the converter (1) with a model predictive in each time step of the regulation} scheme determined by in an optimization unit (4), a cost function J (x _p) as a function of a time characteristic of at least one to be controlled and from the manipulated variables u _p dependent controlled variable x _p of the power converter (1) and / or the electrical load (2 ) is implemented over a predetermined prediction horizon N _p > 1 and the optimization unit (4) optimizes the cost function J (x _p ) in each time step of the control in order _{to determine the optimal manipulated variables u opt} to be set, characterized in that in the cost function J ( x _p ) a cost term J _d (x _P ) is included, which includes an alternating component i _{d, A} c of the DC voltage intermediate via the at least one intermediate circuit capacitor (Cdc) enkreises (12) flowing electrical intermediate circuit current (id) minimized. 12. Converter according to claim 11, characterized in that a prediction model (3) is provided which determines the associated temporal profiles of the controlled variables x _{p for various possible time profiles of the manipulated variables u p} over the prediction horizon N _p and the optimization unit (4) for each of these temporal curves of the controlled variable x _{p determines} the value of the cost function J (x _p ) and that the optimization unit (4) is assigned from the temporal curve of the manipulated variable u _p to the temporal curve of the controlled variable x _p, which is the value of the cost function J. (x _p ) minimized, the optimal manipulated variables u _opt determined. 13. Converter according to claim 11 or 12, characterized in that the converter (1) is designed with interleaved switching lines (SLn), at least two switching lines (SL1, SL2) being connected to one another and another in the cost function J (x _p ) Cost term J _Sat (x _P ), J _bai (x) is implemented, which _{calculates the differential current (i si} - i _S 2) as the difference between the electrical phase currents (i _si , i _S 2) at the outputs of the switching phases (SL1, SL2 ) limited. 14. Converter according to claim 13, characterized in that in the converter (1) at least one nested choke (14) is provided which has two chokes (L) which are wound with opposing windings on a common core of the nested choke (14) , and the nested choke (14) connects outputs of two switching lines (SL1, SL2) of the switching stage (11) of the converter (1) to one another to form a voltage output of one phase of the converter (1). 15. Converter according to claim 13 or 14, characterized in that a measuring arrangement (15) in the form of a bridge circuit is provided in the converter (1), in which a first bridge branch with a series circuit of two resistors (R _M 2) between the outputs of Chokes (L) of the switching lines (SL1, SL2) is connected and a second bridge branch with a series connection of two measuring resistors (R _M 3) is connected between the outputs of the chokes (L) of the switching lines (SL1, SL2) so that a center point ( AP) between the two resistors (R _M 2) of the first bridge branch is connected to a shunt branch of the bridge circuit with a common output terminal (BP) between the two measuring resistors (R _M 3) of the second bridge branch, and that a Voltage sensor (MS1) is provided, which _{measures the electrical voltage applied to the two measuring resistors (R M} 2) of the first bridge arm, which is proportional to the differential current (i _si - i _S 2).