A new energy vehicle permanent magnet motor high-precision torque control method

Abstract

The application discloses a high-precision torque control method for a permanent magnet motor of a new energy vehicle, constructs a neural network torque observer by taking motor voltage, current, temperature and electric angular velocity as input, performs online parameter identification on the motor by introducing historical data of the motor under different working conditions, and calculates partial derivative information of torque to current by using the identified parameters. The application substitutes the partial derivative information of torque to current and the observed torque of the torque observer into a Taylor formula of the historical data to determine whether the network output is correct, and adjusts the network interior by using a least mean square algorithm, which improves the torque observation precision of the network, solves the shortcoming that the fixed parameters in the traditional network cannot adapt to time-varying working conditions, improves the accuracy and robustness of the network, and is suitable for working conditions of parameter change caused by motor temperature change, torque control precision reduction and MTPA trajectory deviation.

Description

Technical Field

[0001] This invention belongs to the field of motor control technology, specifically relating to a high-precision torque control method for permanent magnet motors in new energy vehicles. Background Technology

[0002] Permanent magnet motors (PMMs) are widely used in new energy vehicles due to their high efficiency factor and high torque accuracy. However, the operating environment of PMMs is complex and variable. The control system mainly adopts torque mode, and the vehicle does not have a torque sensor installed, so the true value of the shaft torque cannot be obtained during control. At the same time, the temperature of the motor body varies greatly, and parameters such as resistance and flux linkage are greatly affected by temperature, which leads to inaccuracies in the controlled model. This results in reduced torque control accuracy and makes it difficult for the motor to operate according to the MTPA (Maximum Torque Per Ampere) trajectory. At present, algorithms that consider temperature changes are roughly divided into online methods and offline methods. Existing offline methods (such as the literature [Wu Zhihong, Li Gensheng, Zhu Yuan, et al. Maximum Torque Per Ampere Control of Vehicle Permanent Magnet Synchronous Motor Considering Parameter Variation [J]. Beijing: Annual Meeting of China Electrotechnical Society, 2011, 1]) require calibration of the motor through experiments or finite element simulation and the establishment of a corresponding torque current lookup table. However, offline methods not only require a large number of complicated experiments, but also have poor versatility and cannot be ported. Existing online methods control the motor by identifying motor parameters or by using signal injection. However, existing parameter identification methods (such as the literature [Sun Shuyuan, Li Xiaoqing. Parameter identification of permanent magnet synchronous motor and research on MTPA and feedforward control [J]. Modular Machine Tool & Automated Manufacturing Technology, 2024, (10): 115-119+125]) have the problems of complex calculation process and difficulty in convergence of results. Signal injection methods include high-frequency signal injection method and virtual signal injection method. High-frequency signal injection method (such as the literature [Liu TH, Chen Y, Dai B C. MTPA control for an IPMSM drive system using high frequency injection method [C] / / 2016IEEE International Conference on Industrial Technology (ICIT).IEEE, 2016: 181-186]) can achieve MTPA control without relying on motor parameters, but due to the need for continuous signal injection, there will be additional losses in the control system.The virtual signal injection method (such as the literature [Sun T, Wang J, Koc M, et al. Self-learning MTPA control of interior permanent magnet synchronous machine drives based on virtual signal injection[J]. IEEE Transactions on Industry Applications, 2016, 52(4): 3062-3070]) solves the loss problem caused by real physical signal injection, but due to the lack of torque observation, its torque control accuracy under temperature changes is not high. Summary of the Invention

[0003] In view of the above, the present invention provides a high-precision torque control method for permanent magnet motors in new energy vehicles, which can maintain torque control accuracy and enable the motor to achieve MTPA operation.

[0004] A high-precision torque control method for permanent magnet motors in new energy vehicles includes the following steps:

[0005] (1) Construct a torque observer based on a gated recurrent unit neural network to observe the electromagnetic torque of the motor in real time;

[0006] (2) For the current electromagnetic torque data of the motor, add two sets of historical electromagnetic torque data under the same working conditions, and calculate the inductance parameters and flux linkage parameters of the motor under the working conditions by adding and subtracting elimination methods on the two sets of historical electromagnetic torque data.

[0007] (3) Calculate the partial derivative of the electromagnetic torque with respect to the stator current based on the inductance and flux linkage parameters of the motor.

[0008] (4) Expand the historical electromagnetic torque data using the Taylor formula, substitute the electromagnetic torque output by the torque observer and the partial derivatives calculated in step (3) into the Taylor formula, and determine whether the output of the torque observer is correct. If it is incorrect, adjust the torque observer using the least mean square algorithm.

[0009] (5) The current reference value generated by the MTPA controller is compensated based on the electromagnetic torque output by the adjusted torque observer and the partial derivative calculated in step (3), and then the compensated current reference value is used to realize high-precision torque control of the motor.

[0010] Furthermore, the specific implementation of step (1) is as follows: First, collect motor data under different operating conditions, including i d i q ω e ud u q T e and T d , where i d and i q These are the d-axis stator current and q-axis stator current of the motor, respectively. d and u q These are the d-axis stator voltage and q-axis stator voltage of the motor, respectively, ω e T is the electric angular velocity of the motor. e T is the electromagnetic torque of the motor. d The motor temperature is then determined; a torque observer consisting of multiple gated recurrent unit neural networks and a fully connected layer is then established. The outputs of each gated recurrent unit neural network are weighted and summed in the fully connected layer to obtain the electromagnetic torque of the motor; subsequently, all motor data are divided into training and test sets, with the i-th motor data in the training set being used as the basis for the measurement. d i q u d u q ω e and T d As input to a gated recurrent unit neural network, T e As truth labels, the neural networks of each gated recurrent unit are trained, and the trained torque observer is tested using motor data from the test set.

[0011] Furthermore, the specific implementation of step (2) is as follows: First, define the current electromagnetic torque data of the motor as T. e (i d i q The two sets of historical electromagnetic torque data are T e (i d +A, i q +B) and T e (i d +C, i q +D), where i d and i q These are the d-axis stator current and q-axis stator current of the motor (corresponding to the current electromagnetic torque data), respectively. A, B, C, and D are all constants and represent the stator current difference between the historical electromagnetic torque data and the current electromagnetic torque data. Then, based on the elimination method, the inductance parameters and flux linkage parameters of the motor are solved by the following calculation expressions.

[0012]

[0013] Where: L d and L q These are the d-axis and q-axis stator inductances of the motor, respectively, where p is the number of rotor pole pairs, and ψ is the number of rotor pole pairs. fThis refers to the rotor flux linkage of the motor.

[0014] Further, in step (3), the partial derivative of the electromagnetic torque with respect to the stator current is calculated using the following formula;

[0015]

[0016] Wherein: T e i represents the electromagnetic torque of the motor. d and i q These are the d-axis stator current and q-axis stator current of the motor, respectively, where p is the number of rotor pole pairs and ψ is the number of stator currents. f L is the rotor flux linkage of the motor. d and L q These are the d-axis stator inductance and q-axis stator inductance of the motor, respectively. For T e to i d The partial derivatives, For T e to i q The partial derivatives, for to i d The partial derivatives of .

[0017] Furthermore, in step (4), a set of historical electromagnetic torque data T is taken. e (i d +A, i q +B), its Taylor expansion is as follows:

[0018]

[0019] in: electromagnetic torque T e to i d The partial derivatives, electromagnetic torque T e to i q The partial derivatives, for to i d The partial derivatives of .

[0020] Furthermore, the specific implementation method for determining whether the torque observer output is correct in step (4) is as follows: First, the electromagnetic torque T output by the torque observer is... GRU As T e (i d i q Substituting these values ​​into the Taylor formula, we can calculate the error signal ε(k) between the observer output and the target output at time k:

[0021]

[0022] Then, determine whether the following relationship holds true. If not, adjust the network parameters of the torque observer using the least mean square algorithm.

[0023]

[0024] Where: W(k) is the internal weight parameter of the fully connected layer in the torque observer at time k, W(k+1) is the internal weight parameter of the fully connected layer in the torque observer after adjustment at time k+1, H(k) is the input of the fully connected layer in the torque observer at time k, η is the iteration step size, and k is a natural number.

[0025] Furthermore, the specific implementation of step (5) is as follows: firstly, the current compensation amount is calculated using the following expression;

[0026]

[0027] Wherein: T e * T is the reference value for electromagnetic torque. GRU The electromagnetic torque output by the torque observer, i d and i q These are the d-axis stator current and q-axis stator current of the motor, respectively, where p is the number of rotor pole pairs and ψ is the number of stator currents. f L is the rotor flux linkage of the motor. d and L q These are the d-axis stator inductance and q-axis stator inductance of the motor, respectively. electromagnetic torque T e to i d The partial derivatives, electromagnetic torque T e to i q The partial derivatives of Δi d and Δi q These are the d-axis current compensation and the q-axis current compensation, respectively.

[0028] Then the d-axis stator current reference value i generated by the MTPA controller d * and q-axis stator current reference value i q * respectively with Δi d and Δi q After superposition and PI (Proportional-Integral) control and decoupling, the stator voltage reference values ​​of the d-axis and q-axis are obtained. These are then converted into stator voltage reference values ​​of the α-axis and β-axis, and then a three-phase PWM (Pulse Width Modulation) signal is generated by the SVPWM (Space Vector Pulse Width Modulation) module to control the on / off state of the power switching devices in the motor inverter, thereby achieving high-precision torque control of the motor.

[0029] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-mentioned high-precision torque control method for permanent magnet motors in new energy vehicles.

[0030] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned high-precision torque control method for permanent magnet motors in new energy vehicles.

[0031] Based on the above technical solution, the present invention has the following beneficial technical effects:

[0032] 1. The high-precision torque control method for permanent magnet motors in new energy vehicles of the present invention constructs a neural network torque observer by taking motor voltage, current, temperature, and electric angular velocity as inputs. Since various parameters of the motor are considered, the accuracy of the network observation is higher. By introducing historical data of the motor under different operating conditions, the motor is identified online, and the partial derivative information of torque with respect to current is calculated through the identified parameters.

[0033] 2. This invention substitutes the calculated partial derivative of torque with respect to current and the observed torque of the neural network torque observer into the Taylor expansion formula of historical data. It judges whether the network output is correct by checking whether both sides of the equation are equal, and uses the least mean square algorithm to adjust the network internally. This method improves the torque observation accuracy of the network and solves the shortcomings of traditional networks with fixed internal parameters that cannot adapt to time-varying operating conditions, thus improving the accuracy and robustness of the network.

[0034] 3. This invention achieves MTPA control by using the calculated partial derivative of torque with respect to current and the observed torque of the neural network torque observer to compensate for the d-axis and q-axis currents of the MTPA in real time. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the structure of a gated recurrent unit neural network.

[0036] Figure 2 This is a block diagram of the torque observer structure based on a gated recurrent unit neural network in this invention.

[0037] Figure 3 This is a block diagram of the adjusted torque observer structure in this invention.

[0038] Figure 4 This is a block diagram of the high-precision torque control system for the permanent magnet motor of the present invention. Detailed Implementation

[0039] To describe the present invention in more detail, the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Built-in permanent magnet synchronous motors (PMSMs) are widely used in new energy vehicles due to their high efficiency and high-precision torque control. Precise torque control and efficient torque-to-pause-motion (MTPA) control are crucial for maintaining the stability and efficiency of new energy vehicles. However, the enclosed housing and integrated powertrain design of new energy vehicles make motor heat dissipation difficult. When the motor operates for extended periods, its temperature rises, causing changes in relevant motor parameters and leading to inaccuracies in the motor control model. This results in decreased torque control accuracy and MTPA trajectory deviation. To address this problem, this invention proposes a high-precision torque control method for permanent magnet motors in new energy vehicles, comprising the following steps:

[0041] (1) Adopting as follows Figure 1 The gated recurrent unit network shown establishes a neural network torque observer to observe motor torque. The torque observer is as follows: Figure 2 As shown, Figure 1 ZhongX t H represents the input information at the current moment. t-1 H represents the hidden state from the previous time step. The hidden state acts as the neural network's memory, containing information about the data seen by previous nodes. t R represents the output of the current cell. t and Z t These represent the states of resetting and updating the door, respectively. W represents the candidate hidden state. z W r W and W represent the weight values ​​of the reset gate, update gate, and candidate hidden state, respectively. Figure 2 In this diagram, W1 represents the first weight parameter of the fully connected layer, W2 represents the second weight parameter of the fully connected layer, and W6 represents the sixth weight parameter of the fully connected layer.

[0042] (2) First, collect data information of the motor under different operating conditions, and define i d The stator current along the d-axis and the current along the i-axis in the rotating coordinate system (dq-axis coordinate system) of the motor. q Let ω be the q-axis stator current in the dq-axis coordinate system of the motor. e T is the electric angular velocity of the motor. e u is the electromagnetic torque of the motor. d The d-axis stator voltage and u in the dq-axis coordinate system of the motor q T is the q-axis stator voltage in the dq-axis coordinate system of the motor. d For motor temperature, i is collected under different operating conditions. d i q ω e u d u q T e and Td Based on the collected motor data, i d i q u d u q ω e and T d As network input, T e The gated recurrent unit neural network is generated as the network output. After determining the input and output, the collected data is divided into training set and test set. The gated recurrent unit network is trained using the training set and tested using the test set.

[0043] (3) Add two historical data points under the same operating condition, and calculate the motor inductance and flux linkage parameters under this operating condition by adding, subtracting, and eliminating variables from the historical data. First, define the current motor output torque as T. e (i d i q ), taking two torque data points under the same operating conditions as the current motor output torque as historical data, i.e., T e (i d +A, i q +B) and T e (i d +C, i q +D), where A, B, C, and D are constants representing the current difference between the historical torque data and the current torque. Since the two historical data are from the same operating condition, the motor parameters of these two historical data are the same. The motor parameters can be obtained by the addition and subtraction elimination method:

[0044]

[0045] In the formula: ψ f For rotor flux linkage, L d For the d-axis stator inductance, L q is the q-axis stator inductance, and p is the number of rotor pole pairs.

[0046] Historical data changes according to the operating conditions. When the operating conditions of the motor change, the torque under the changed operating conditions is taken as historical data.

[0047] (4) Calculate the partial derivative of torque with respect to current based on the calculated inductance and flux linkage parameters. The specific formula for calculating the partial derivative of torque with respect to current is as follows:

[0048]

[0049] In the formula: Let be the partial derivative of the electromagnetic torque with respect to the d-axis current. This is the partial derivative of the electromagnetic torque with respect to the q-axis current. It is the second-order partial derivative of the electromagnetic torque with respect to the dq-axis current.

[0050] (5) Expand the historical data using the Taylor series. Substitute the observed torque and the calculated partial derivative of the torque with respect to the current from the neural network torque observer into the Taylor series. Determine if the network output is correct by checking if both sides of the equation are equal. If the two sides are not equal, adjust the neural network torque observer using the least mean square algorithm. The specific results are as follows: Figure 3 As shown. The historical data, expanded using the Taylor formula, is as follows:

[0051]

[0052] In the formula: Δx represents the influence of temperature rise and magnetic saturation effect on the motor parameters under the current operating conditions. Since the selected historical data is different from T... e (i d i q Since the conditions are the same, Δx can be ignored.

[0053] The output torque of the neural network is defined as T. GRU T GRU Substitute the motor torque under the current operating condition into T. e (i d i q In formula (2), the partial derivative information calculated by formula (2) is substituted into formula (3). By judging whether the two sides of the equation are equal, the correctness of the output torque of the neural network can be determined. If the two sides of the equation are not equal, the neural network torque observer is adjusted by the least mean square algorithm. The adjustment principle is as follows: Figure 3 As shown, the formula is expressed as:

[0054]

[0055] In the formula: η is the step size for weight calculation, ε(k) is the error signal between the network output and the target output at time k, H(k) is the input of the neural network to the fully connected network at time k, W(k) is the internal weight parameter of the neural network at time k, and W(k+1) is the adjusted internal weight parameter of the neural network at time k+1.

[0056] (6) The MTPA current reference value is adjusted using the obtained partial derivative of torque with respect to current and the observed torque of the neural network torque observer. The optimal current adjustment scheme for the q-axis, as can be seen from the torque formula, allows for compensation of the q-axis current once accurate information on the partial derivative of current is available. The specific formula is as follows:

[0057]

[0058] In the formula: T e* For torque reference value, Δi q This is the q-axis current compensation value.

[0059] One of the core features of the MTPA control strategy is that the partial derivative of torque with respect to the current vector angle β is zero. Therefore, the d-axis current can be corrected by ensuring that the partial derivative of torque with respect to the current vector angle β is zero. First, the relationship between the d-axis current and the q-axis current and the current vector angle β is as follows:

[0060]

[0061] Secondly, the partial derivative of torque with respect to the current vector angle can be expanded into the following formula:

[0062]

[0063] Substituting formula (6) into formula (7) yields the following formula:

[0064]

[0065] Where: dT e / dβ is the partial derivative of the torque with respect to the current vector angle.

[0066] To meet the requirements of MTPA operation, the current-compensated dT is required. e Since / dβ is 0, and the compensation value of the q-axis current is already known, substituting formula (7) into formula (8) yields the d-axis current compensation formula:

[0067]

[0068] In the formula: Δi d This is the d-axis current compensation value.

[0069] The control system implemented based on the method of this invention is as follows: Figure 4 As shown, the d-axis stator current reference value i generated by the MTPA controller is... d * and q-axis stator current reference value i q * respectively with Δi d and Δi q After superposition and PI control and decoupling, the stator voltage reference values ​​of the d-axis and q-axis are obtained. These are then converted into stator voltage reference values ​​of the α-axis and β-axis, and then a three-phase PWM signal is generated by the SVPWM module to control the on and off of the power switching devices in the motor inverter, thereby achieving high-precision torque control of the motor.

[0070] In summary, the high-precision torque control method for permanent magnet motors in new energy vehicles of this invention inputs the motor's current, voltage, temperature, and electrical angular velocity information into a gated recurrent neural network to observe torque. It uses two historical data points to identify motor parameters online, and calculates the partial derivative of torque with respect to current based on these identified parameters. The partial derivative information and the observed torque are substituted into the Taylor expansion of the historical data. The neural network is adjusted based on whether both sides of the equation are equal; if not, the weight parameters within the neural network are adjusted using a least mean square algorithm to improve the observation accuracy of the torque observer. Simultaneously, the MTPA current reference value is adjusted online using the partial derivative information and the observed torque, thus achieving MTPA operation. This invention can accurately observe motor torque under temperature variations, improves the dynamic performance and robustness of the neural network torque observer, and achieves MTPA operation.

[0071] The above description of the embodiments is provided to enable those skilled in the art to understand and apply the present invention. Those skilled in the art can readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made to the present invention by those skilled in the art based on the disclosure thereof should be within the scope of protection of the present invention.

Claims

1. A high-precision torque control method for a permanent magnet motor in a new energy vehicle, characterized in that, Includes the following steps: (1) Construct a torque observer based on a gated recurrent unit neural network to observe the electromagnetic torque of the motor in real time; (2) For the current electromagnetic torque data of the motor, add two sets of historical electromagnetic torque data under the same operating conditions. By performing addition and subtraction elimination on these two sets of historical electromagnetic torque data, calculate the inductance parameters and flux linkage parameters of the motor under this operating condition. The specific implementation method is as follows: First, define the current electromagnetic torque data of the motor as... T e ( i d , i q The two sets of historical electromagnetic torque data are respectively T e ( i d + A , i q + B )and T e ( i d + C , i q + D ),in i d and i q These are the d-axis stator current and q-axis stator current of the motor, respectively. A , B , C , D All are constants and represent the stator current difference between historical electromagnetic torque data and current electromagnetic torque data; Then, based on the elimination method, the inductance and flux linkage parameters of the motor are solved using the following calculation expressions; in: L d and L q These are the d-axis stator inductance and q-axis stator inductance of the motor, respectively. p This represents the number of rotor pole pairs of the motor. For the rotor flux linkage of the motor; (3) Calculate the partial derivative of the electromagnetic torque with respect to the stator current based on the inductance and flux linkage parameters of the motor; (4) Expand the historical electromagnetic torque data using the Taylor formula, substitute the electromagnetic torque output by the torque observer and the partial derivatives calculated in step (3) into the Taylor formula, and determine whether the output of the torque observer is correct. If it is incorrect, adjust the torque observer using the least mean square algorithm. The specific implementation method for determining whether the torque observer output is correct is as follows: First, the electromagnetic torque output by the torque observer is... T GRU As T e ( i d , i q Substituting into Taylor's formula, we can calculate... k Error signal between the time observer output and the target output ε ( k ): Then, determine whether the following relationship holds true. If not, adjust the network parameters of the torque observer using the least mean square algorithm. in: W ( k )for k Internal weight parameters of the fully connected layer in the time-major torque observer. W ( k+ 1) For k+ After adjusting the internal weight parameters of the fully connected layer in the torque observer at time 1, H ( k )for k The input of the fully connected layer in the moment-to-moment observer, η The iteration step size, k It is a natural number; (5) The current reference value generated by the MTPA controller is compensated based on the electromagnetic torque output by the adjusted torque observer and the partial derivative calculated in step (3). Then, the high-precision torque control of the motor is achieved by using the compensated current reference value. The specific implementation method is as follows: First, the current compensation amount is calculated by the following expression. ， in: This is the reference value for electromagnetic torque. T GRU The electromagnetic torque output by the torque observer. i d and i q These are the d-axis stator current and q-axis stator current of the motor, respectively. electromagnetic torque T e right i d The partial derivatives, electromagnetic torque T e right i q The partial derivatives of Δ i d and Δ i q These are the d-axis current compensation and the q-axis current compensation, respectively. Then the d-axis stator current reference value generated by the MTPA controller and q-axis stator current reference value respectively with Δ i d and Δ i q After superposition and PI control and decoupling, the stator voltage reference values ​​of the d-axis and q-axis are obtained. These are then converted into stator voltage reference values ​​of the α-axis and β-axis, and then a three-phase PWM signal is generated by the SVPWM module to control the on and off of the power switching devices in the motor inverter, thereby achieving high-precision torque control of the motor.

2. The high-precision torque control method for permanent magnet motors in new energy vehicles according to claim 1, characterized in that: The specific implementation method of step (1) is as follows: First, collect motor data under different operating conditions, including i d , i q , ω e , u d , u q , T e and T d ,in u d and u q These are the d-axis stator voltage and q-axis stator voltage of the motor, respectively. ω e The electric angular velocity of the motor. T e The electromagnetic torque of the motor. T d The motor temperature is then determined; a torque observer is then built consisting of multiple gated recurrent unit neural networks and a fully connected layer. The outputs of each gated recurrent unit neural network are weighted and summed in the fully connected layer to obtain the electromagnetic torque of the motor; subsequently, all motor data are divided into training and test sets, with the training set containing the motor data... i d , i q , u d , u q , ω e and T d As input to a gated recurrent unit neural network, T e As truth labels, the neural networks of each gated recurrent unit are trained, and the trained torque observer is tested using motor data from the test set.

3. The high-precision torque control method for permanent magnet motors in new energy vehicles according to claim 1, characterized in that: In step (3), the partial derivative of the electromagnetic torque with respect to the stator current is calculated using the following formula; in: T e The electromagnetic torque of the motor. for right i d The partial derivatives of .

4. The high-precision torque control method for permanent magnet motors in new energy vehicles according to claim 1, characterized in that: In step (4), a set of historical electromagnetic torque data is taken. T e ( i d + A , i q + B The Taylor expansion of this is as follows: in: for right i d The partial derivatives of .

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: The processor is used to execute the computer program to implement the high-precision torque control method for permanent magnet motors in new energy vehicles as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the high-precision torque control method for permanent magnet motors in new energy vehicles as described in any one of claims 1 to 4.

Images