DE102022200868A1 - Control of a flow valve - Google Patents

Abstract

Control device (160) for a current valve (155) in a half-bridge (150) of a power converter (120), the half-bridge (150) controlling a current flow through an electrical machine (110); wherein the control device (160) is set up to determine an operating point of the power converter (120); to determine a gate resistance as a function of the operating point; and driving the current valve (155) with the gate resistance. The gate resistance is determined by means of an artificial neural network that is trained to determine a gate resistance that has an optimized compromise between a low oscillation of a current flowing through the current valve (155) and a small time delay until it is reached of a predetermined flow through the flow control valve (155).

Description

The present invention relates to the control of a flow control valve. In particular, the invention relates to the control of a current valve in a half-bridge that controls current flow through an electrical machine.

An electrical machine can be fed from an intermediate circuit by means of a power converter. The intermediate circuit carries a DC voltage and the electrical machine is designed, for example, as a permanently excited synchronous machine (PSM). The power converter usually includes a number of half-bridges, each of which includes a first and a second flow control valve. The first current valve can control a current flow between a high potential of the intermediate circuit and a phase of the electrical machine, and the second current valve between the phase and a low potential of the intermediate circuit. If the current valves of a half-bridge are opened and closed alternately with a suitable pulse duty factor, a predetermined potential is established at the phase.

The electrical machine can be controlled by the power converter to assume a predetermined direction of rotation, to provide a predetermined torque or to rotate at a predetermined speed. The electric machine can be used in particular as a drive machine for a vehicle, for example a passenger car. In order to carry out a low-loss, precise and efficient control, the current valve, when it is activated, must establish the current flow between the intermediate circuit and the electrical machine as quickly as possible.

The flow control valve can in particular include a field effect transistor or a bipolar transistor, in particular an IGBT. A control terminal of such a transistor is called the gate and is connected to a current source or a voltage source by means of a gate resistor in order to control the current valve. The lower the selected gate resistance, the faster a current flowing through the flow control valve can reach a predetermined threshold value after the first actuation. However, the current flowing through the flow control valve tends to oscillate. The larger the gate resistance is chosen, the faster the oscillation can decay.

The vibration behavior of a flow control valve can depend on a large number of parameters. For example, a junction temperature of the flow control valve, a frequency with which the flow control valve is controlled, or a voltage in the intermediate circuit or a leakage inductance of a circuit can influence the oscillation behavior. It can therefore be difficult to choose a gate resistance that is large enough to allow current flow oscillations to decay in an acceptable time, yet small enough to ensure that the current flowing through the flow control valve reaches a predetermined threshold soon enough after the initial gating.

It has been proposed to provide different gate resistances and to use them depending on a parameter for driving the current valve. However, due to the large number of relevant parameters, it can be difficult to make an appropriate choice.

One of the objects on which the present invention is based is to specify an improved technique for controlling a flow control valve in a half-bridge of a power converter. The invention solves this problem by means of the subject matter of the independent claims. Subclaims reflect preferred embodiments.

A power converter includes a half-bridge with a current valve, the half-bridge controlling a current flow through an electrical machine. According to a first aspect of the present invention, a control device for the flow control valve is set up to determine an operating point of the power converter; to determine a gate resistance as a function of the operating point; and driving the current valve with the gate resistor. The gate resistance is determined using an artificial neural network (ANN), which is trained to determine a gate resistance that represents an optimized compromise between low oscillation of a current flowing through the flow valve and a small time delay until a predetermined current is reached through the flow valve.

The correspondingly trained artificial neural network can allow the gate resistance to be determined quickly and reliably with regard to an operating point, which can include a large number of physical parameters that can affect operation of the power converter, in particular the half-bridge and more particularly the current valve. The flow control valve can be controlled periodically with a predetermined timing, for example by means of pulse width modulation (PWM), pulse-pause modulation (PPM) or another pulse modulation method. The transistor can be controlled for a very short time so that it becomes conductive and switches through. By varying the timing, the provided Current follow a predetermined course, for example a sine wave. The time in which a current valve is switched on can be very short and the course of the current flowing through the current valve after the initial switch-on can be very fast, so that a conventional determination of an optimized gate resistance with regard to the multitude of parameters would be too slow or disproportionately expensive. By controlling the gate resistance as a function of prevailing conditions, a look-ahead control can be implemented that can yield good results with little effort, without requiring regulation at the time of execution. The artificial neural network can be trained in particular by means of reinforcement learning (Reinforcement Learning, DRL, deep RL).

It is preferred that a predetermined number of different gate resistances are provided, one of which is selected for the drive. In this case, the gate resistors can have different resistance values. This embodiment can be implemented in a simple manner in terms of circuitry and has the effect that the value of the gate resistance assumes one of a predetermined multiplicity of predetermined values. In a further development, several gate resistances of different sizes can be selected in parallel, so that the effective value of the gate resistance can be determined according to Kirchhoff's laws.

It is also preferred that a time profile of the gate resistance is determined. In other words, the value of the gate resistance can be changed after the current valve has been driven. For example, a low gate resistance can initially be chosen in order to cause the current flow to increase rapidly, followed shortly thereafter by a larger gate resistance in order to allow the current flow oscillation to decay quickly. The gate resistance can be determined for a predetermined time step or a time course of gate resistances can be determined, which is then controlled. The curve can include multiple time steps and, in one embodiment, include the entire duration of the activation of the flow control valve.

It is further preferable that the gate resistance is changed only at predetermined times after the first drive. There may be a predetermined number of times after the first drive at which the effective gate resistance may be changed. As a result, a solution space for finding an optimum gate resistance can be reduced.

Time intervals between consecutive points in time can increase successively. In this way, it can be taken into account that different values for the gate resistance have a major influence on the oscillation behavior of the current flow, particularly immediately after the first actuation. The longer the current valve is driven, the less influence the size of the gate resistance can have. The solution space can be further reduced by the non-uniform intervals between the predetermined points in time.

The training of the artificial neural network is preferably carried out before productive use of the control device. However, the control device can also be set up to continue training the artificial neural network during ongoing operation. For this purpose, the control device can be set up to determine a time profile of the current flowing through the controlled flow control valve; and to further train the artificial neural network with regard to the operating point, the determined current profile and the gate resistance used for control.

Through the continued training of the artificial neural network during ongoing operation of the control device, an operating point that is frequently assumed can be learned in an improved, reliable manner. Operating points that were only trained a little or not at all during previous training can be improved or additionally learned. A gate resistance with an optimized resistance value can be determined dynamically for each of these states. In addition, the control device can adapt itself to an individual circuitry environment, which can be characterized, for example, by component tolerances or aging effects.

According to a further aspect of the present invention, a half-bridge comprises at least one flow control valve, which is controlled by means of a control device described herein. The half-bridge preferably includes at least two flow control valves, which can be controlled in alternation by the same control device. A first current valve can control a current flow between a high potential of an intermediate circuit and the electrical machine and a second current valve can control a current flow from the electrical machine to a low potential of the intermediate circuit. The current valves can be formed in particular by transistors, for example FET or IGBT. Optionally, several first and/or several second flow control valves can also be connected in parallel, so that they essentially act like a first or second flow control valve.

According to yet another aspect of the present invention, a power converter includes at least one half-bridge as described herein. The power converter preferably includes at least three half-bridges that can be connected to separate phases of the electrical machine.

According to yet another aspect of the present invention, an axle drive includes an electric machine and a power converter described herein. The final drive can be provided for a vehicle, in particular for a motor vehicle such as a passenger car. The electrical machine can include a permanent magnet synchronous machine (PSM). The electric machine can act on wheels mounted on different sides of the axle. Alternatively, the axle drive can include two electrical machines, which can each act on a wheel on different sides of the axle drive. A power converter is preferably assigned to each electrical machine. If more than one power converter is included in the axle drive, associated control devices can be combined with one another or designed to be integrated.

According to yet another aspect of the present invention, a vehicle includes an electric machine and a power converter as described herein. The electrical machine is designed in particular as a drive machine and is set up to move the vehicle on a surface by means of a drive wheel.

According to yet another aspect of the present invention, a first method for training an artificial neural network comprises the steps of determining an operating point of a power converter comprising a half-bridge that controls a current flow through an electrical machine; wherein the half-bridge comprises a flow control valve; determining a gate resistance as a function of the operating point; driving the current valve with the determined gate resistance; detecting a flow of current through the flow control valve; evaluating the course of the current with regard to an oscillation and a time delay until a predetermined current flow is reached; and adjusting the artificial neural network to enhance at the operating point determining a gate resistance that optimizes a trade-off between the criteria.

The first method can be carried out on a predetermined electrical machine using a power converter described herein. For this purpose, the control device can include a programmable microcomputer or microcontroller, which in particular forms the artificial neural network. The method can be in the form of a computer program product with program code means. Optionally, the computer program product is stored on a computer-readable data carrier. Features or advantages of the method can be transferred to a device described herein or vice versa.

The artificial neural network can be trained using reinforcement learning, in particular using deep reinforcement learning. For this purpose, at least two target criteria can be specified, one of which can relate to a low level of oscillation in the current profile after the flow control valve has closed and the other to a high rate of increase in the current flow through the flow control valve. The ANN can be trained to optimize both criteria if possible. To this end, the criteria can be weighted in a predetermined ratio to one another.

The operating point can be determined using one or more sensors on the power converter. Additional parameters of the operating point can emerge from time derivatives of a measured parameter. The operating point can also include a parameter that is derived from a requirement for the power converter or a behavior of the connected electrical machine. For example, the operating point can include one or more temperatures at an element of the power converter, in particular one of the flow control valves, a requested or existing torque of the electrical machine, a requested or existing speed of the electrical machine, a requested or existing rotational acceleration of the electrical machine, a rotational position of the electrical machine, a voltage of the intermediate circuit, a state of charge or a temperature of an electrical energy store connected to the intermediate circuit, which can in particular include a rechargeable battery, a current flow through one of the phases of the electrical machine or another, in particular electrical, parameter.

According to yet another aspect of the present invention, a second method for controlling a current valve in a half-bridge of a power converter, the half-bridge controlling a current flow through an electrical machine, comprises steps of determining an operating point of the power converter; determining a gate resistance as a function of the operating point; and driving the current valve with the gate resistor. The gate resistance is determined using an artificial neural network that is trained to determine a gate resistance that offers an optimized compromise between a low oscillation of a represents the current flowing through the flow control valve and a small time delay until a predetermined current is reached through the flow control valve.

The second method can also be carried out using a power converter described herein on a predetermined electrical machine. The first and the second method are connected to each other by the present inventive idea. Features or advantages may also be transferred among the methods or between the second method and an apparatus described herein.

The processing device can be set up to carry out a method described herein in whole or in part. For this purpose, the processing device can include a programmable microcomputer or microcontroller, and the method can be in the form of a computer program product with program code means. The computer program product can also be stored on a computer-readable data carrier. Features or advantages of the method can be transferred to the device or vice versa.

• 1 ein System mit einem Stromrichter;
• 2 eine Halbbrücke in einem Stromrichter;
• 3 Stromverläufe in Abhängigkeit unterschiedlicher Gate-Widerstände an einem Stromventil; und
• 4 Ablaufdiagramme zweier Verfahren darstellt.

The invention will now be described in more detail with reference to the attached figures, in which:

• 1 a system with a power converter;
• 2 a half-bridge in a power converter;
• 3 Current curves as a function of different gate resistances on a current valve; and
• 4 Shows flowcharts of two methods.

1 shows a system 100 on board a vehicle 105, which in particular can comprise a motor vehicle. The system 100 includes an electric machine 110 which is preferably set up to act on a drive wheel 115 of the vehicle 105 . A power converter 120 is set up to control currents through phases of the electrical machine 110 . For this purpose, the power converter 120 is connected to an intermediate circuit 125 which is usually connected to an electrical energy store on board the vehicle 105 . An intermediate circuit capacitor 130 is preferably provided between a high potential 135 and a low potential 140 of the intermediate circuit 125 .

The power converter 120 can be part of an electric axle drive 145 together with the electric machine 110 . The final drive 145 can also include the drive wheel 115 .

In the present schematic representation, power converter 120 includes three half-bridges 150, each of which includes two flow control valves 155, one of which controls a current flow between high potential 135 and an associated phase of electrical machine 110, and the other controls a current flow from electrical machine 110 to low potential 140 of intermediate circuit 125. The flow control valves 155 of the half-bridges 150 can each be activated by means of a control device 160 . In the embodiment shown, three half-bridges 150 are provided, which are connected to three phases of the electrical machine 110 and can be controlled by means of a common control device 160 .

A current flowing between electrical machine 110 and a half-bridge 150 of power converter 120 can be determined using a current sensor 165 . The individual currents through the electrical machine 110 are controlled as a function of a request which the control device 160 can accept via an interface 170 . A state of the electrical machine 110 can be taken into account, which can be defined, for example, by a rotation angle and currents flowing through the individual phases.

To control the flow control valves 155, the control device 160 can also take into account other parameters, which can be determined in particular by means of associated sensors that are 1 are not marked. Further parameters can also be derived from the state of the electric machine 110 or the request at the interface 170.

2 shows a half-bridge 150 in a power converter 120 in a schematic representation. The two flow control valves 155 are controlled in a corresponding manner by means of a driver 205, which can be controlled by the control device 160. Driver 205 can be actuated, in particular, with regard to a current flowing through flow control valve 155; current sensors 165 are therefore provided for determining these currents. The task of driver 205 is, depending on a request from control device 160, to control associated current valve 155 in such a way that it allows the greatest possible current flow between high potential 135 and electrical machine 110 or between electrical machine 110 and low potential 140 of intermediate circuit 125 as quickly as possible. In this case, the current flowing through the flow control valve 155 should be subjected to an oscillation which is as small as possible and/or which decays as quickly as possible after the first actuation. The driver 205 can also close the flow control valve 155 at the request of the control device 160 by controlling the flow control valve in the opposite direction or not at all. Ideally, the flow of current can be terminated practically without delay and without any oscillation behavior.

The driver 205 may include a plurality of resistors 210 one of which may be selected to act as a gate resistor for the associated power valve 155 . The individual resistors 210 are preferably of different sizes, so that depending on the selection of a resistor 210 a larger or smaller control current flows through the gate connection of the flow control valve 155 and enables a corresponding flow of current between the remaining two connections of the flow control valve 155.

3 shows exemplary current curves as a function of different gate resistances at a current valve 155. The times, resistance values and current strengths shown are exemplary; other values can result in a real circuit. Exemplary current curves are shown in a first diagram 305 in an upper area. A time is drawn in the horizontal direction and a current flowing through a flow control valve 155 is drawn in the vertical direction. In a lower range of 3 a second diagram 310 shows a time profile of a gate resistance of a flow control valve 155. A time is shown in the horizontal direction and an ohmic resistance is shown in the vertical direction. The time scales of charts 305 and 310 correspond to each other.

If the flow control valve 155 is controlled in order to open it, the resulting flow of current depends on an effective gate resistance. A first current course 315 is brought about by a first gate resistance 320, which is deliberately chosen to be small. Although the first current curve 315 rises very quickly after a latency period of approximately 8 ns in the present example, so that it has reached a maximum approximately 12 ns after activation, it then transitions into an aperiodic oscillation, which has not yet decayed approximately 50 ns after the first activation.

A second current profile 325 corresponds to a second gate resistance 330, which is deliberately chosen to be large. The second current curve 325 only reaches its maximum approximately 14 ns after the first actuation, but then does not fall as far as the first curve 315, oscillates with relatively strong damping and has almost reached the quality of a direct current after approximately 50 ns.

A third current profile 335 forms a compromise or an optimized combination of properties of the profiles 315 and 325 in that the current flowing through the flow control valve 155 increases as quickly as possible after switching on, but is then subject to the lowest possible oscillation. For this purpose, a third gate resistor 340 is formed, which initially lies between the resistance values of the first gate resistor 320 and the second gate resistor 330 and is variable over time in the present illustration. The gate resistance can assume different discrete resistance values.

It is proposed to train an artificial neural network to determine the gate resistance of the current valve 155 as a function of an existing operating point of a surrounding converter 120, which can include an operating point of a half-bridge 150 and in particular an operating point of a current valve 155, in such a way that two criteria are optimized, which can be seen in the current curves 315 and 325. The first current course 315 rises very rapidly when the flow control valve 155 opens for the first time.

A first criterion can therefore be that a predetermined flow of current through the flow control valve 155 should be achieved as quickly as possible after the flow control valve 155 is actuated for the first time. In other embodiments, the first criterion can also be that the current flow should rise as steeply as possible after the first actuation. In yet another specific embodiment, the first criterion can state that the current flow should have reached a predetermined current value after a predetermined time after the flow control valve 155 was activated for the first time.

A second criterion can be that the current flow, after the first actuation, should be subject to as little oscillation as possible, which should decay as quickly as possible. The vibration can be evaluated, for example, with regard to its amplitude or with regard to its damping. The second criterion can require the highest possible damping, which is associated with rapid decay of the vibration below a predetermined amplitude. The second criterion can also require that a predetermined current flow should not be undershot or should be undershot as seldom as possible. For example, it can be required that the current flow does not fall below 0. This is not the case for the first current flow 315 in the present illustration. Another threshold value, in particular greater than 0, can also be used. In yet another embodiment, the second criterion may require that a current integrated over time flowing through the flow control valve 155 is maximized. For the second criterion, a predetermined time Licher area after the first activation of the flow control valve 155 are considered.

4 FIG. 4 shows flowcharts of a first exemplary method 400 and a second exemplary method 450. The methods 400 and 450 can be implemented using a power converter 120 in particular. The first method 400 aims to train an ANN for a controller described herein, while the second method 450 may use an appropriately trained ANN to control a flow valve in a half-bridge.

In a step 405 of the first method 400, an operating point of the power converter 120 can be determined. To this end, a large number of parameters can be determined which reflect the function and/or the state of components of the power converter 120 or components connected thereto. With regard to a single flow control valve 155, 10 or more, preferably 20 or more, parameters can be taken into account. The operating point may be expressed by a vector that includes a value for each of the specified parameters.

In a step 410, a gate resistance for a current valve 155 under consideration can be determined. For this purpose, an optimized gate resistance can be determined with respect to the specific operating point using an ANN. In a further embodiment, a predetermined course of the gate resistance over time can be determined. The gate resistance or its course can be varied systematically or pseudo-randomly in a solution space defined by degrees of freedom of the gate resistance. For example, the gate resistance may be between a predetermined minimum and a predetermined maximum value. If the gate resistance changes, the resistance value of the gate resistance can change in a predetermined quantitative and/or temporal pattern. The resistance can be determined using the still untrained or only partially trained ANN.

In a step 415, the flow control valve 155 can be driven with the specific gate resistance. In the case of a specific time profile of the gate resistance, the flow control valve 155 is preferably controlled according to the specific profile; a reaction of a parameter changing during the course is preferably not provided.

In a step 420, a current running through the flow control valve 155 can be detected. In this case, a time profile of the current is preferably recorded in a predetermined time range.

In a step 425, the artificial neural network can be trained with respect to one or more criteria that can be determined on the basis of the determined current profile. The criteria are usually competing in that they involve a rapid increase in current flow and a rapid decay of an oscillation in current flow. Optionally, the criteria can be weighted in a predetermined manner. With regard to the criteria described, a quality of the specific gate resistance or its profile can be determined and the ANN can be made through reinforcement learning to preferentially determine a solution for the gate resistance that has proven to be successful at a similar operating point.

The method 400 can very often be carried out cyclically, with the determination quality of the ANN being able to be successively improved. The first method 400 is preferably carried out with regard to a large number of operating points. In this case, in particular, operating points can be specified which were observed on a real power converter 120 and were found to be relevant. The operating points can also be selected, for example, at regular intervals in the vector space of all possible operating points.

The second method 450 can be used to operate a power converter 120 . In a step 455, an operating point of the power converter 120 can be determined. This step can correspond to step 405 of the first method 400 .

In a step 460 a gate resistance for the current valve 155 under consideration can be determined by applying the vector of the operating point to the artificial neural network that was trained in the first method 400 . The ANN can then provide an optimized gate resistance or an optimized time profile of a gate resistance.

In a step 465, the flow control valve 155 can be controlled with the specific gate resistance or its time profile. This step can correspond to step 420 of the first method 400 .

As indicated by the block arrow, the second method 450 requires a trained artificial neural network, which may have been trained using the first method 400, for example. Procedures 400 and 450 differ in just a few steps. In one embodiment, the first method 400 can also be carried out further in terms of the second method 450 after the artificial neuro nal network has already been trained to a certain point. In this way, the ANN can determine further optimized gate resistances for specific operating points during operation, while controlling an electrical machine 110 .

The ANN can allow a large state space to be mapped onto a small action space. The state space st can include a current I flowing through the current valve 155 and its derivative or derivatives, as well as a state Ω of the currently used gate resistor 210: $$s_t\left(I\left(t\right),\mathrm{i.e}I\left(t\right)/\mathrm{i.e}t,\mathrm{i.e}I\left(t\right)/\mathrm{i.e}{t}^2,\Omega \left(t\right)\right)$$

The action space at preferably represents design decisions for the state of the gate resistor 210 at a time step subsequent to a current time step. This state can be determined as an absolute value. Alternatively, the state can be determined as an incremental value of the ohmic resistance of the gate resistor 210, for example as one of the indications “keep gate resistance”, “decrease gate resistance” and “increase gate resistance”. The reduction or enlargement can be done by a predetermined amount. $$a_t\left(\Omega \left(t+1\right)\right)$$

The training of the ANN preferably follows the central approach of Deep Reinforcement Learning (DRN) to determine the value of the action function based on the Bellman equation as an iterative update: $$Q\ast \left(s,a\right)=E_{\left(s\text{'}\right)}\left[\mathrm{right}+\underset{\mathrm{i.e}\text{'}}{g}Q\ast \left(s\text{'},a\text{'}\right)\vee s,a\right].$$

Reference List

100

system

105

vehicle

110

electric machine

115

drive wheel

120

power converter

125

intermediate circuit

130

intermediate circuit capacitor

135

high potential

140

low potential

145

electric axle drive

150

half bridge

155

flow valve

160

control device

165

current sensor

170

interface

205

driver

210

gate resistance

305

first diagram

310

second chart

315

first current flow (fast)

320

first gate resistor (low)

325

second current flow (low vibration)

330

second gate resistor (large)

335

third current flow (optimized)

340

third gate resistor (optimized)

400

first procedure

405

Determining the operating point

410

Determining a gate resistance

415

Driving a current valve with the gate resistor

420

Detecting a flow of current through the flow control valve

425

Training an ANN

450

second procedure

455

Determining the operating point

460

Determining a gate resistance using KNN

465

Driving a current valve with the gate resistor

Claims (12)

Control device (160) for a current valve (155) in a half-bridge (150) of a power converter (120), the half-bridge (150) controlling a current flow through an electrical machine (110); wherein the control device (160) is set up to determine an operating point of the power converter (120); to determine a gate resistance as a function of the operating point; and driving the current valve (155) with the gate resistance; characterized in that the gate resistance is determined by means of an artificial neural network which is trained to determine a gate resistance which has an optimized compromise between a small oscillation of a current flowing through the current valve (155) and a small time delay until a predetermined current through the flow control valve (155). Control device (160) after claim 1 , wherein a predetermined number of different gate resistances are provided, one of which is selected. Control device (160) after claim 1 or 2 , where a time profile of the gate resistance is determined. Control device (160) after claim 3 , wherein the gate resistance is changed only at predetermined times after the first drive. Control device (160) after claim 4 , where distances between consecutive points in time increase successively. Control device (160) according to one of the preceding claims, wherein the control device (160) is set up to determine a time profile of the current flowing through the controlled flow valve (155); and to further train the artificial neural network with regard to the operating point, the determined current profile and the gate resistance used for control. Half-bridge (150) comprising at least one flow control valve (155) which is activated by means of a control device (160) according to one of the preceding claims. Power converter (120), comprising at least one half-bridge (150). claim 7 . Final drive (145), comprising an electric machine (110) and a power converter (120). claim 8 . Vehicle (105) comprising an electric machine (110) and a power converter (120). claim 8 . A method (400) for training an artificial neural network, the method (400) comprising the steps of: determining an operating point of a power converter (120) that includes a half-bridge (150) that controls a current flow through an electrical machine (110); wherein the half-bridge (150) comprises a flow control valve (155); determining a gate resistance as a function of the operating point; driving the current valve (155) with the determined gate resistance; detecting a flow of current through the flow control valve (155); Evaluation of the current course with regard to an oscillation and a time delay until a predetermined current flow is reached; and adjusting the artificial neural network to enhance at the operating point determining a gate resistance that optimizes a trade-off between the criteria. Method (450) for controlling a current valve (155) in a half-bridge (150) of a power converter (120), wherein the half-bridge (150) controls a current flow through an electrical machine (110); wherein the method (450) comprises the following steps: determining an operating point of the power converter (120); determining a gate resistance as a function of the operating point; and driving the current valve (155) with the gate resistor; characterized in that the gate resistance is determined by means of an artificial neural network that is trained to determine a gate resistance that represents an optimized compromise between a low oscillation of a current flowing through the flow valve (155) and a small time delay until a predetermined current is reached through the flow valve (155).

Images