Weight factor design and loss optimization method for open-winding motor model predictive control

Abstract

A method for designing weight factors and optimizing losses of model predictive control of an open-winding motor, the method comprising: constructing a mathematical model of an open-winding permanent magnet synchronous motor, designing a cost function of a model predictive control algorithm with constraint losses added according to a control target, and performing preliminary control verification; setting a weight factor range, and obtaining various types of constraint loss data corresponding to different weight factor combinations based on the mathematical model and the cost function; using the weight factors and the corresponding working conditions, speeds and load conditions as input layers, and the corresponding various types of constraint loss data as output layers, training a feedforward neural network with a gray wolf algorithm, using minimization of root mean square error as an evaluation criterion, and using the combination of the minimum index result and its corresponding weight factor as a set of optimal values; establishing a database through the minimum index and its corresponding weight factor under multiple working conditions, and realizing dynamic optimization of the model predictive control weight factor through a table lookup method.

Description

Technical Field

[0001] The present invention relates to the field of motor loss optimization, and in particular to an open-winding motor model predictive control weight factor design and loss optimization method. Background Art

[0002] Under the background of "dual carbon", the energy industry is transforming and upgrading, and new energy vehicles are becoming more and more popular. In the new energy vehicle industry, open-winding permanent magnet synchronous motor drive systems are widely used. Compared with traditional permanent magnet synchronous motors, open-winding permanent magnet synchronous motors have advantages such as wide speed regulation range, high fault tolerance and multi-level output. However, in the model predictive control of open-winding permanent magnet synchronous motors, the design of the weight factor of the cost function is still a challenge. In the motor driving process, it is necessary to weigh the weight factor of each cost function item in the model predictive control. If the weight factor of the cost function is unbalanced, the motor torque fluctuation may increase, the loss may increase, or the system may oscillate and fail to work. Different weights will affect the steady-state operation of the system, which is of great significance to the stability and optimization of the system. In addition, in the context of new energy, the open-winding permanent magnet synchronous motor driven by the dual inverter group, whether it is the loss on the motor or the inverter, is a part that cannot be ignored, but at this stage, there are few model predictive control algorithms that incorporate the loss of the open-winding motor and inverter into the constraint variables.

[0003] In the optimization of the weight of the open-winding permanent magnet synchronous motor model predictive torque control, the commonly used optimization algorithms are the weightless method, the bounded search method, and the heuristic algorithm using artificial intelligence. The weightless method splits the cost function into multiple separate cost functions, and brings different voltage vectors into the calculation to obtain the cost function value, sorts them from small to large, and then adds the sequence values ​​respectively. The smallest one is the optimal voltage vector, avoiding the design of weight factors. However, it is only suitable for control with few cost variables, otherwise its calculation amount will increase exponentially with the number of cost variables. The second is the bounded search method, which limits the weight factor to a certain range, performs traversal optimization, scans all possible combinations, and finally obtains the optimal value. The optimal parameters can be obtained, but its optimization time is long, and once the working conditions are changed, the selected weight factor is invalid. The heuristic algorithm of artificial intelligence can accurately calculate the weight online in real time, but the algorithm is highly complex and the calculation time is long, which is difficult to implement in actual operating conditions.

[0004] Therefore, in the current open-winding motor model predictive torque control, the motor and inverter losses are not taken into consideration as control items, and the commonly used weight optimization methods have the main disadvantages of large amount of calculation, long time consumption and low robustness. Therefore, the open-winding permanent magnet synchronous motor model predictive control drive system needs to be optimized and improved. Summary of the invention

[0005] In view of the technical defects and technical drawbacks in the prior art, the embodiment of the present invention provides a method for designing weight factors and optimizing losses of an open-winding motor model predictive control, which overcomes the above problems or at least partially solves the above problems. The specific scheme is as follows:

[0006] A method for designing weight factors and optimizing losses of an open-winding motor model predictive control, the method comprising:

[0007] Step 1: construct a mathematical model of an open-winding permanent magnet synchronous motor, design a cost function of a model predictive control algorithm with constraint loss based on the control objective, and conduct preliminary control verification;

[0008] Step 2: Set the weight factor range, and obtain the corresponding constraint loss data under different weight factor combinations based on the mathematical model and cost function;

[0009] Step 3: Take the weight factor and the corresponding working condition, speed and load as the input layer, and the corresponding constraint loss data as the output layer. Use the Grey Wolf Algorithm to train the feedforward neural network, use the minimization of the root mean square error as the evaluation criterion, and take the combination of the minimum index result and its corresponding weight factor as a set of optimal values.

[0010] Step 4: Establish a database through the minimum indicators and their corresponding weight factors under multiple working conditions, and realize dynamic optimization of model predictive control weight factors through table lookup method.

[0011] Furthermore, in step 1, the constrained losses include torque error, total motor loss and two sets of inverter losses, and the weight factors in step 2 include torque item weight factors, motor loss item weight factors and inverter loss item weight factors, and various types of constrained losses include torque error, total motor loss and two sets of inverter losses.

[0012] Further, step 1 comprises:

[0013] Establish the mathematical model of open-winding permanent magnet synchronous motor drive system:

[0014]

[0015] Among them, u d and u q are the dq axis electron voltage, i d and i q are dq axis electron current respectively; L d and L q are the dq axis electronic inductance respectively;

[0016] Taking the open-winding permanent magnet synchronous motor as the control object, the first-order forward Euler method is used to discretize the stator voltage equation under the dq axis, and the following is obtained:

[0017]

[0018] Where k is the sampling time, T s is the sampling period, i d (k) and i q (k) represents the dq axis current value at time k, i d (k+1) and i q (k+1) represents the predicted current value of d and q axes at time k+1, u d (k) and u q (k) represents the dq axis voltage value at time k, R is the stator resistance, ω e is the electrical angular velocity, ψ f is the permanent magnet flux;

[0019] The torque at time k+1 can be calculated by formula (1):

[0020]

[0021] Among them, T e (k+1) is the torque prediction value at time k+1, P n is the number of motor pole pairs;

[0022] The expected torque value T e *The speed error at time k can be obtained by collecting it through the PI controller:

[0023]

[0024] Among them, K p , K i are the proportional coefficient and integral coefficient of the controller respectively, s is the integral of the controller; N ref is the set speed, N is the actual speed;

[0025] According to the dq axis current and voltage values, a loss model of motor iron loss and copper loss is established. Copper loss is equivalent to the power consumed when the stator current flows through the stator resistance, and iron loss is equivalent to the power consumed when the iron loss branch current flows through the equivalent iron resistance. When the system is stable, the total motor loss is only related to the active current of the iron loss branch. By derivation, the extreme value can be analyzed, and the minimum loss can be obtained. Then the expected value of the iron loss branch is:

[0026]

[0027] Among them, i ts_ref is the expected value of the iron loss branch current, R c is the equivalent iron resistance;

[0028] According to the circuit principle, the instantaneous value of the iron loss branch current can be calculated as:

[0029]

[0030] Among them, i ts is the instantaneous value of the iron loss branch current;

[0031] The switching loss model is an analytical method to quantify the switching loss of the inverter. The switching device will generate losses when it is turned on and off. The switching loss of a single power switch device can be defined as:

[0032]

[0033] Among them, INVSL sw is the total inverter loss, INVSL on is the turn-on loss, INVSL off is the turn-off loss, t r is the inverter turn-on time, t f is the inverter off time, i r is the current passing through during the opening process, i f is the current passing through during the shutdown process, u r is the turn-on voltage, u f For the convenience of calculation, the equation (6) is approximately transformed into:

[0034]

[0035] Among them, S x (k) is the current x-phase switch state, S x (k-1) is the switch state of phase x at the previous moment, U dc is the bus voltage, i x (k) is the current x-phase current at the current moment, f δ (m) is defined as:

[0036]

[0037] In order to optimize the comprehensive performance of the motor, the torque error, motor loss and inverter loss are included in the constraints, and the torque error, motor loss and inverter loss are defined as the cost function:

[0038]

[0039] Among them, J Te is the torque tracking cost function, J Ploss is the motor loss cost function, J INVSL are the loss cost functions of the two sets of inverters. Based on this, the total cost function can be defined as:

[0040] J=λ Te JTe +λ Loss J Ploss +λ INV J INVSL (10)

[0041] Among them, λ Te is the torque term weight factor, λ Loss is the weight factor of the motor loss term, λ INV is the inverter loss weight factor.

[0042] Furthermore, step 1 also includes:

[0043] Define a fitness function f t To evaluate the control performance, as follows;

[0044] f t =α1Te error +α2P loss +α3INVSL loss (11)

[0045] Among them, Te error , P loss With INVSL loss They are torque error, total motor loss and two sets of inverter losses respectively. α1, α2 and α3 are predefined weights in the fitness function. The larger the weight, the more importance it receives in the objective function.

[0046] Furthermore, step 2 includes:

[0047] Import all models, functions and corresponding motor parameters in step 1 into Matlab, and build a simulation model through Matlab;

[0048] Setting λ Te , Loss and λ INV range and step size, based on the simulation model, to obtain different λ Te , Loss and λ INV The corresponding Te under the combination error , P loss With INVSL loss .

[0049] Further, step 3 includes:

[0050] Different λ Te , Loss and λ INV The corresponding Te under the combination error , P loss With INVSL loss As data samples, each group of λTe , Loss and λ INV And the corresponding working conditions, speed and load conditions as the input layer, the corresponding Te error , P loss With INVSL loss As the output layer, the feedforward neural network is trained with the Levenberg-Marquardt algorithm, the minimization of the root mean square error is used as the evaluation criterion, and the combination of the minimum indicator result and its corresponding weight factor is taken as a set of optimal values.

[0051] Furthermore, step 3 includes: dividing the data samples into a training set and a validation set in a ratio of 7:3.

[0052] Furthermore, step 3 also includes: setting gray wolf optimization algorithm parameters, optimizing the feedforward neural network, setting population size, hunting dimension and hunting boundary value parameters according to the number of data sets, and after the setting is completed, performing iterative optimization, substituting the weights and biases to be optimized into the feedforward neural network, customizing the neural network architecture according to the input and output parameters, using a fully connected method to build neurons and activation functions to construct the overall network, using a Sigmoid activation function between hidden layer 1 and hidden layer 2, and using a Relu activation function between hidden layer 2 and the output layer, and using the minimum root mean square error RMSE as an evaluation indicator for the feedforward neural network.

[0053] Further:

[0054] The Sigmoid activation function is shown below

[0055]

[0056] The Relu activation function is as follows:

[0057] f(x)=max(0,x) (13)

[0058] The RMSE expression is as follows:

[0059]

[0060] Among them, y i That is the actual value, That is the predicted value of the feedforward neural network.

[0061] The present invention has the following beneficial effects:

[0062] (1) The open-winding motor model predictive control weight factor optimization design method proposed in the present invention incorporates the motor and inverter losses as control factors, and uses a feedforward neural network optimized based on the grey wolf optimization algorithm. Compared with the traditional feedforward neural network, the neural network converges faster, has higher prediction accuracy, and improves the robustness of the neural network.

[0063] (2) The present invention reduces the iron loss and copper loss of the open-winding permanent magnet synchronous motor and reduces the switching loss of the dual inverter group.

[0064] (3) The present invention can adjust the weight factor online to ensure stable operation of the system, reduce algorithm complexity, shorten the optimization cycle, and reduce the calculation burden of the control system. Since the present invention finally uses only one database to perform the optimization, it greatly simplifies the internal resources of the digital system and is easy to apply in actual working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flow chart of a method for designing weight factors and optimizing losses of an open-winding motor model predictive control provided by an embodiment of the present invention;

[0066] Figure 2 A diagram of an open-winding permanent magnet synchronous motor drive system provided by an embodiment of the present invention;

[0067] Figure 3 The open-winding permanent magnet synchronous motor loss model provided by the embodiment of the present invention;

[0068] Figure 4 The model prediction control block diagram provided by the embodiment of the present invention;

[0069] Figure 5 A structural diagram of a feedforward neural network provided by an embodiment of the present invention;

[0070] Figure 6 An evolutionary iteration graph of the Grey Wolf Optimization Algorithm provided by an embodiment of the present invention;

[0071] Figure 7 A comparison chart of the effects of a feedforward neural network optimized by the Grey Wolf Optimization Algorithm and a traditional feedforward neural network provided by an embodiment of the present invention;

[0072] Figure 8 Feedforward neural network training, prediction and verification diagram provided by an embodiment of the present invention;

[0073] Fig. 9 A rotation speed curve diagram provided by an embodiment of the present invention;

[0074] Fig.10 A torque curve diagram provided by an embodiment of the present invention;

[0075] Fig.11Comparison diagram of open-winding motor losses provided by an embodiment of the present invention;

[0076] Fig.12 A comparison diagram of the losses of an open-winding motor inverter provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] like Figure 1 As shown, a method for designing weight factors and optimizing losses of an open-winding motor model predictive control is provided in an embodiment of the present invention, and the method includes:

[0079] Step 1: construct a mathematical model of an open-winding permanent magnet synchronous motor, design a cost function of a model predictive control algorithm with constraint loss based on the control objective, and conduct preliminary control verification;

[0080] Step 2: Set the weight factor range, and obtain the corresponding constraint loss data under different weight factor combinations based on the mathematical model and cost function;

[0081] Step 3: Take the weight factor and the corresponding working condition, speed and load as the input layer, and the corresponding constraint loss data as the output layer. Use the Grey Wolf Algorithm to train the feedforward neural network, use the minimization of the root mean square error as the evaluation criterion, and take the combination of the minimum index result and its corresponding weight factor as a set of optimal values.

[0082] Step 4: Establish a database through the minimum indicators and their corresponding weight factors under multiple working conditions, and realize dynamic optimization of model predictive control weight factors through table lookup method.

[0083] The present invention is further described in detail below in conjunction with the accompanying drawings and examples, but the embodiments should not be construed as limiting the present invention.

[0084] The example uses an open-winding permanent magnet synchronous motor, whose parameters are shown in Table 1:

[0085]

[0086] Table 1 Open-winding permanent magnet synchronous motor parameters

[0087] The specific steps include:

[0088] Establish a mathematical model of open-winding permanent magnet synchronous motor drive system;

[0089]

[0090] Among them, ud and uq are the dq axis electron voltages, that is, the electron voltages corresponding to the d axis and q axis in the dq coordinate system, i d and i q are the dq axis electron currents, i.e. the electron currents corresponding to the d axis and q axis in the dq coordinate system; L d and L q are the dq axis equivalent inductance, i.e. the electronic inductance corresponding to the d axis and q axis in the dq coordinate system, R is the stator resistance, ω e is the electrical angular velocity, ψ f is the permanent magnet flux;

[0091] Using the first-order forward Euler method, the stator voltage equation under the dq axis is discretized, and the following is obtained:

[0092]

[0093] Where k is the sampling time, T s is the sampling period, i d (k) and i q (k) represents the dq axis current value at time k, i d (k+1) and i q (k+1) represents the predicted current value of d and q axes at time k+1, u d (k) and u q (k) represents the dq axis voltage value at time k respectively;

[0094] Through the above formula, the state at the current time k can be used to predict the state at the time k+1, and then the torque value at the time k+1 can be calculated;

[0095]

[0096] Among them, T e (k+1) is the torque prediction value at time k+1, P n is the number of motor pole pairs;

[0097] The expected torque value T e *The speed error at time k can be obtained by collecting it through the PI controller:

[0098]

[0099] Among them, K p , K i are the proportional coefficient and integral coefficient of the controller respectively; N ref is the set speed, N is the actual speed;

[0100] According to the dq axis current and voltage values, a loss model of motor iron loss and copper loss is established. Copper loss is equivalent to the power consumed when the stator current flows through the stator resistance, and iron loss is equivalent to the power consumed when the iron loss branch current flows through the equivalent iron resistance. When the system is stable, the total motor loss is only related to the active current of the iron loss branch. By derivation, the extreme value can be analyzed, and the minimum loss can be obtained. Then the expected value of the iron loss branch is:

[0101]

[0102] Among them, i ts_ref is the expected value of the iron loss branch current, R c is the equivalent iron resistance;

[0103] According to the circuit principle, the instantaneous value of the iron loss branch current can be calculated as:

[0104]

[0105] Among them, i ts is the instantaneous value of the iron loss branch current;

[0106] The switching loss model is an analytical method to quantify the switching loss of the inverter. The switching device will generate losses when it is turned on and off. The switching loss of a single power switch device can be defined as:

[0107]

[0108] Among them, INVSL sw is the total inverter loss, INVSL on is the turn-on loss, INVSL off is the turn-off loss, t r is the inverter turn-on time, t f is the inverter off time, i r is the current passing through during the opening process, i f is the current passing through during the shutdown process, u r is the turn-on voltage, u f For the convenience of calculation, the equation (6) is approximately transformed into:

[0109]

[0110] Among them, S x (k) is the current x-phase switch state, S x (k-1) is the switch state of phase x at the previous moment, U dc is the bus voltage, i x (k) is the current x-phase current at the current moment, f δ (m) is defined as:

[0111]

[0112] In order to optimize the comprehensive performance of the motor, the torque error, motor loss and inverter loss are included in the constraints, and the torque error, motor loss and inverter loss are defined as the cost function:

[0113]

[0114] Among them, J Te is the torque tracking cost function, J Ploss is the motor loss cost function, J INVSL are the loss cost functions of the two sets of inverters. Based on this, the total cost function can be defined as:

[0115] J=λ Te J Te +λ Loss J Ploss +λ INV J INVSL

[0116] Among them, λ Te is the torque term weight factor, λ Loss is the weight factor of the motor loss term, λ INV is the inverter loss weight factor.

[0117] Substitute the voltage vector into the model predictive control algorithm and calculate the voltage vector corresponding to the lowest J at the next moment, which is the optimal vector. The fitness function is defined as:

[0118] f t =100Te error +0.01P loss +10000INVSL loss

[0119] Among them, in this embodiment, the balanced system performance is selected, so the torque error, motor loss and inverter loss are normalized to the range of [0-1] by multiplying the order of magnitude of each variable by the corresponding multiple.

[0120] Import the above model and motor parameters into Matlab and follow the steps below. Figure 4 The simulation model is built as shown, and the simulation values ​​of the weight factors are set to λ Te [7-8], step size 0.3; λ Loss [0.005-0.105], step size 0.03; λ INV [10-100], step size 30; then start the simulation.

[0121] The Te corresponding to different combinations of weight factors collected from the simulation error , P lossWith INVSL loss As the data set, and divide the training set and cross-validation set into a ratio of 70% and 30%. Before training, the gray wolf optimization algorithm is first applied to optimize the weights and biases of the feedforward neural network. In this example, the population size is set to 100, the dimension is set to 5, and the boundary values ​​are all [-5,5]. After the setting is completed, optimization is performed, and the weights and biases to be optimized are substituted into the feedforward neural network. According to the above input and output parameters, the neural network architecture is defined as an input layer with 5 neurons, two hidden layers, 11 neurons and 3 neurons respectively, and an output layer with 3 neurons. The fully connected method is used to build neurons and activation functions to construct the overall network. The Sigmoid activation function is used between hidden layer 1 and hidden layer 2, and the Relu activation function is used between hidden layer 2 and the output layer. The training algorithm uses

[0122] Levenberg-Marquardt, the number of trainings is 1000, the learning rate is 0.01, the minimum target is 0.00001, and the minimum root mean square error RMSE is used as the evaluation index of the feedforward neural network. The evolutionary iteration diagram of the gray wolf optimization algorithm is as follows Figure 6 As shown, Figure 7 Comparison of the effects of feedforward neural network optimized by Grey Wolf Optimization Algorithm and traditional feedforward neural network. GWOBP is the feedforward neural network optimized by Grey Wolf Optimization Algorithm, and BP is the traditional feedforward neural network.

[0123] After the training is completed, a neural network model can be obtained. Figure 6 , Figure 7 and Figure 8 The figure is a diagram of the training process of motor loss and weight factor. It can be seen that the feedforward neural network optimized by the Gray Wolf optimization algorithm has smaller error and is more stable than the traditional feedforward neural network. The error of the feedforward neural network optimized by the Gray Wolf optimization algorithm fluctuates within [-2, +2], while the error of the traditional feedforward neural network fluctuates within [-8, +2]. It can be clearly seen that the superiority of the present invention is achieved. After the training is completed, the three variables Teerror, Ploss and INVSLloss are added according to formula (20) to obtain the database based on the open-winding permanent magnet synchronous motor in this example. The verification conditions are shown in Table 2.

[0124]

[0125] Table 2 Verification conditions

[0126] The comparison and verification diagrams are shown in Figures 9-12. Figure 9-11 In the above, Loss MPTC is the model predictive control considering the loss term, and MPTC is the traditional model predictive control. Fig.12In the above, Loss MPTCINV is the model predictive control considering the inverter loss term, and MPTCINV is the traditional model predictive control.

[0127] From above Fig. 9 , Fig.10 It can be seen that the present invention has better dynamic performance than the traditional model predictive control and can quickly reach a balanced state after the working conditions change. Fig.11 It can be seen that the motor loss of the two strategies is compared. Under working condition 1, the loss of the present invention is reduced by about 9W compared with the traditional model predictive control, which is optimized by 23%; under working condition 2, the loss of the present invention is reduced by about 13W compared with the traditional model predictive control, which is optimized by 16%; under working condition 3, the loss of the present invention is reduced by about 8W compared with the traditional model predictive control, which is optimized by 8%; under working condition 4, the loss of the present invention is reduced by about 13W compared with the traditional model predictive control, which is optimized by 16%; Fig.12 It can be seen that within 0.2s, the inverter loss of the present invention is 0.015W lower than that of the traditional model predictive control, which is optimized by 4%. In summary, the effectiveness of the present invention can be seen.

[0128] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for designing weight factors and optimizing losses of open-winding motor model predictive control, characterized in that: The method comprises: Step 1: construct a mathematical model of an open-winding permanent magnet synchronous motor, design a cost function of a model predictive control algorithm with constraint loss based on the control objective, and conduct preliminary control verification; Step 2: Set the weight factor range, and obtain the corresponding constraint loss data under different weight factor combinations based on the mathematical model and cost function; Step 3: Take the weight factor and the corresponding working condition, speed and load as the input layer, and the corresponding constraint loss data as the output layer. Use the Grey Wolf Algorithm to train the feedforward neural network, use the minimization of the root mean square error as the evaluation criterion, and take the combination of the minimum index result and its corresponding weight factor as a set of optimal values. Step 4: Establish a database through the minimum indicators and their corresponding weight factors under multiple working conditions, and realize dynamic optimization of the model predictive control weight factors through the table lookup method; Wherein, step 1 comprises: The switching loss model is an analytical method to quantify the switching loss of the inverter. The switching device will generate losses when it is turned on and off. The switching loss of a single power switch device can be defined as: Among them, INVSL sw is the total inverter loss, INVSL on is the turn-on loss, INVSL off is the turn-off loss, t r is the inverter turn-on time, t f is the inverter off time, i r is the current passing through during the opening process, i f is the current passing through during the shutdown process, u r is the turn-on voltage, u f For the convenience of calculation, the equation (6) is approximately transformed into: Among them, S x (k) is the current x-phase switch state, S x (k-1) is the switch state of phase x at the previous moment, U dc is the bus voltage, i x (k) is the current x-phase current at the current moment, f δ (m) is defined as: In order to optimize the comprehensive performance of the motor, the torque error, motor loss and inverter loss are included in the constraints, and the torque error, motor loss and inverter loss are defined as the cost function: Among them, i ts_ref is the expected value of the iron loss branch current, i ts is the instantaneous value of the iron loss branch current, J Te is the torque tracking cost function, J Ploss is the motor loss cost function, J INVSL are the loss cost functions of the two sets of inverters. Based on this, the total cost function can be defined as: J=λ Te J Te +λ Loss J Ploss +λ INV J INVSL (10) Among them, λ Te is the torque term weight factor, λ Loss is the weight factor of the motor loss term, λ INV is the inverter loss weight factor.

2. The open-winding motor model predictive control weight factor design and loss optimization method according to claim 1, characterized in that: In step 1, the constrained losses include torque error, total motor loss and two sets of inverter losses. In step 2, the weight factors include torque item weight factor, motor loss item weight factor and inverter loss item weight factor. Various types of constrained losses include torque error, total motor loss and two sets of inverter losses.

3. The open-winding motor model predictive control weight factor design and loss optimization method according to claim 1, characterized in that: Step 1 includes: Establish the mathematical model of open-winding permanent magnet synchronous motor drive system: Among them, u d and u q are the dq axis electron voltage, i d and i q are dq axis electron current respectively; L d and L q are the dq axis equivalent inductances respectively; Taking the open-winding permanent magnet synchronous motor as the control object, the first-order forward Euler method is used to discretize the stator voltage equation under the dq axis, and the following is obtained: Where k is the sampling time, T s is the sampling period, i d (k) and i q (k) represents the dq axis current value at time k, i d (k+1) and i q (k+1) represents the predicted current value of dq axis at time k+1, u d (k) and u q (k) represents the dq axis voltage value at time k, R is the stator resistance, ω e is the electrical angular velocity, ψ f is the permanent magnet flux; The torque at time k+1 can be calculated by formula (1): Among them, T e (k+1) is the torque prediction value at time k+1, P n is the number of motor pole pairs; The expected torque value T e *The speed error at time k can be obtained by collecting it through the PI controller: Among them, K p , K i are the proportional coefficient and integral coefficient of the controller respectively, s is the integral of the controller; N ref is the set speed, N is the actual speed; According to the dq axis current and voltage values, a loss model of motor iron loss and copper loss is established. Copper loss is equivalent to the power consumed when the stator current flows through the stator resistance, and iron loss is equivalent to the power consumed when the iron loss branch current flows through the equivalent iron resistance. When the system is stable, the total motor loss is only related to the active current of the iron loss branch. By derivation, the extreme value can be analyzed, and the minimum loss can be obtained. Then the expected value of the iron loss branch is: Among them, R c is the equivalent iron resistance; According to the circuit principle, the instantaneous value of the iron loss branch current can be calculated as:

4. The open-winding motor model predictive control weight factor design and loss optimization method according to claim 3, characterized in that: Step 1 also includes: Define a fitness function f t To evaluate the control performance, as follows; f t =α1Te error +α2P loss +α3INVSL loss (11) Among them, Te error , P loss With INVSL loss They are torque error, total motor loss and two sets of inverter losses respectively. α1, α2 and α3 are predefined weights in the fitness function. The larger the weight, the more importance it receives in the objective function.

5. The method for designing weight factors and optimizing losses of open-winding motor model predictive control according to claim 4, characterized in that: Step 2 includes: Import all models, functions and corresponding motor parameters in step 1 into Matlab, and build a simulation model through Matlab; Setting λ Te , Loss and λ INV range and step size, based on the simulation model, to obtain different λ Te , Loss and λ INV The corresponding Te under the combination error , P loss With INVSL loss .

6. The open-winding motor model predictive control weight factor design and loss optimization method according to claim 5, characterized in that: Step 3 includes: Different λ Te , Loss and λ INV The corresponding Te under the combination error , P loss With INVSL loss As data samples, each group of λ Te , Loss and λ INV And the corresponding working conditions, speed and load conditions as the input layer, the corresponding Te error , P loss With INVSL loss As the output layer, the feedforward neural network is trained with the Levenberg-Marquardt algorithm, the minimization of the root mean square error is used as the evaluation criterion, and the combination of the minimum indicator result and its corresponding weight factor is taken as a set of optimal values.

7. The open-winding motor model predictive control weight factor design and loss optimization method according to claim 5, characterized in that: Step 3 includes: dividing the data samples into a training set and a validation set in a ratio of 7:

3.

8. The open-winding motor model predictive control weight factor design and loss optimization method according to claim 1, characterized in that: Step 3 also includes: setting the parameters of the gray wolf optimization algorithm, optimizing the feedforward neural network, setting the population size, hunting dimension and hunting boundary value parameters according to the number of data sets, and after the setting is completed, performing iterative optimization, substituting the weights and biases to be optimized into the feedforward neural network, customizing the neural network architecture according to the input and output parameters, and using a fully connected method to build neurons and activation functions to construct the overall network. The Sigmoid activation function is used between hidden layer 1 and hidden layer 2, and the Relu activation function is used between hidden layer 2 and the output layer. The minimum root mean square error RMSE is used as the evaluation indicator of the feedforward neural network.

9. The method for designing weight factors and optimizing losses of open-winding motor model predictive control according to claim 8, characterized in that: The Sigmoid activation function is shown below The Relu activation function is as follows: f(x)=max(0,x) (13) The RMSE expression is as follows: Among them, y i That is the actual value, That is the predicted value of the feedforward neural network.

Images