Supercapacitor Direct Connection Energy Feedback Control Method Based on Motor External Characteristics

Abstract

Supercapacitor direct-connected energy feedback control method based on the external characteristics of the motor, the steps include: 1) Analyze the structural parameters of the electro-excited doubly salient motor, establish a field-circuit coupling model and the peripheral drive circuit; 2) Divide the data segments, obtain the external characteristic data of the output voltage and current of the motor under different excitation currents, speeds and loads through multi-condition simulations, and preprocess the data; 3) Initialize the multi-layer perceptron model, take the motor speed, the target feedback current and the supercapacitor voltage as the model inputs, and take the reference value of the excitation current as the output. 4) Deploy the optimal model to the DSP embedded system to construct an energy feedback control system with supercapacitor direct connection based on the external characteristics of the electro-excited doubly salient motor. The control method proposed by the present invention has faster control speed and accuracy compared with the traditional PID controller, effectively improving the energy feedback efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor control, and specifically relates to a supercapacitor direct connection energy feedback control method based on the external characteristics of the motor. Background Technique

[0002] The doubly salient electro-magnetic motor is a switched reluctance motor. Its rotor and stator are both salient pole structures. The rotor is laminated with silicon steel sheets, and at the same time, electro-excitation technology is adopted. The excitation winding and the stator winding are all concentrated on the stator. This special structure makes the DSEM simple and reliable, and is very suitable as the power supporting for special vehicles.

[0003] For the energy feedback of the doubly salient electro-magnetic motor based on supercapacitor direct connection, during braking feedback, without using the traditional DC-DC converter, the supercapacitor is directly connected to the output side of the DSEM rectifier to absorb the feedback energy. At the same time, by adjusting the excitation current to control the excitation intensity, the feedback current is further controlled to stabilize at the target value.

[0004] After retrieval, there is a patent with the Chinese patent application number CN202410167136.5. This application uses a numerical calculation method to obtain the reference value of the excitation current during energy feedback. First, the supercapacitor charging current Idc is deduced according to the state of charge SOCC(S) of the supercapacitor, and further the relationship between the DSEM excitation current and the charging current is deduced. Finally, the relationship between the excitation current If and the state of charge of the supercapacitor is obtained. Based on the above formula, the reference value of the excitation current is calculated according to the state of charge of the supercapacitor, so that the supercapacitor reaches the desired charging state. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a supercapacitor direct connection energy feedback control method based on the external characteristics of the motor. It solves the defects of complex traditional numerical calculation and low accuracy of the DSEM energy feedback based on supercapacitor direct connection, improves the control performance, and increases the energy feedback efficiency.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] The supercapacitor direct connection energy feedback control method based on the external characteristics of the motor is specifically as follows:

[0008] Step 1: Build a simulation platform:

[0009] Analyze the mechanical structure and winding structure of the DSEM prototype, build a motor finite element model in Maxwell, and build the peripheral circuit of the DSEM including a three-phase rectifier bridge and an excitation drive circuit in Simplorer to form a field-circuit coupling simulation platform of Maxwell + Simplorer;

[0010] Step 2: Collect and preprocess data:

[0011] A constant current load was used to fix the current on the output side of the DSEM system. The speed was divided into 3000 rpm and 1000 rpm intervals at 50 rpm. The excitation current was continuously varied from 0 to 6 A at each speed point. The voltage on the output side of the DSEM system was collected. The simulation was repeated at different speed points to obtain mapping data of the excitation current and output voltage at different speeds. The collected data was preprocessed, including filtering, removing outliers, and normalization.

[0012] Step 3: Train the MLP model:

[0013] The speed, target feedback current, and supercapacitor voltage are used as model inputs, and the excitation current reference value is used as output. The data is divided into a training set and a validation set. The model is trained using the training set and the model accuracy is verified on the validation set.

[0014] Step 4: Deploy the model:

[0015] The structural parameters, all weight parameters and bias parameters of the trained MLP model are extracted, and the extracted MLP model data is deployed to the main control DSP of the embedded system.

[0016] As a further improvement of the present invention, the parameters of the mechanical structure and winding structure of the DSEM prototype in step 1 include the inner and outer diameters of the stator and rotor, the number of pole pairs, the air gap, the number of winding turns, and the core material.

[0017] As a further improvement of the present invention, the normalization processing of the collected data in step 2 is specifically as follows:

[0018] Normalize the data, the normalization formula is:

[0019] ;

[0020] x is the original data, x min is the minimum value of the data set, x max is the maximum value of the data set, x norm The data are normalized.

[0021] As a further improvement of the present invention, the specific steps of step 3 are as follows:

[0022] Initialize the input layer to 3 nodes, which are the system output side voltage U C , output side current I C and motor speed n;

[0023] The hidden layer adopts a two-layer structure, the first hidden layer dimension is n1, and the second hidden layer dimension is n2;

[0024] One node in the output layer, i.e., the reference value of the excitation current I fref ;

[0025] Use the training data to train the established MLP model, take minimizing the root mean square error RMSE as the optimization goal, verify the model on the validation set data after training, train the model multiple times, and select the model with the smallest RMSE as the final model of the external characteristic of DSEM;

[0026] The structure of the trained 3-layer MLP model is as follows:

[0027] The structure of the first layer:

[0028] ;

[0029] Z1 is the weighted sum of the first layer, W0 is the weight matrix of the first layer, is the input data of the first layer, b0 is the bias vector of the first layer, and A1 is the output of the first layer;

[0030] The structure of the second layer:

[0031] ;

[0032] Z2 is the weighted sum of the second layer, W1 is the weight matrix of the second layer, b1 is the bias vector of the second layer, and A2 is the output of the second layer;

[0033] The structure of the output layer is:

[0034] ;

[0035] Z3 is the weighted sum of the output layer, W2 is the weight matrix of the output layer, and b2 is the bias vector of the output layer;

[0036] ;

[0037] is the activation function, where x is the original data.

[0038] As a further improvement of the present invention, in step 4, the specific process is as follows:

[0039] S41. Extract the structure parameters of the trained MLP model and deploy the model algorithm and structure parameters to the DSP controller;

[0040] S42. Construct a control method for modeling the external characteristic of DSEM based on MLP: The system collects the supercapacitor voltage, motor speed, and target feedback current in real time. The supercapacitor voltage is used as the voltage on the output side of DSEM, and the target feedback current is used as the current on the output side. The data is input into the MLP model, and the model calculates and outputs the reference value of the excitation current to make the feedback current stable at the target value.

[0041] The advantages brought by this application are as follows:

[0042] (1) Traditional methods rely on mathematical analytical models and usually adopt methods such as local linearization. It is difficult to adapt to a wide range of operating conditions, and the modeling accuracy is limited. While the multi-layer perceptron can effectively process a highly non-linear and strongly coupled motor system, achieving an end-to-end high-precision mapping between the rotational speed, voltage, target feedback current, and excitation current.

[0043] (2) Compared with traditional PID control, using MLP to calculate the reference value of the excitation current does not require the iterative operation process of the traditional PID controller, and the control speed of the feedback current is faster. At the beginning of energy feedback, it can quickly increase the feedback current to the target value.

[0044] (3) In the later stage of energy feedback, as the motor speed decreases and the supercapacitor voltage rises, the demand for the excitation current increases rapidly. The traditional PID controller has a static error, resulting in the actual value of the excitation current being lower than the demand value, causing the feedback current to start dropping before the excitation current reaches the maximum value. However, the control method of this invention can accurately calculate, making the excitation current change synchronously with the demand value.

[0045] (4) The control method proposed in this invention has a faster control speed and higher control accuracy, and can effectively improve the energy feedback efficiency of the doubly salient electro-magnetic motor directly connected to the supercapacitor. Brief Description of the Drawings

[0046] Figure 1 is the system block diagram of the control method proposed in this invention;

[0047] Figure 2 is the DSEM external characteristic data when the feedback current collected in this invention is 20A;

[0048] Figure 3 is the data of using MLP to fit the DSEM external characteristics in this invention;

[0049] Figure 4 is the experimental data when using traditional PID control and the initial voltage of the supercapacitor is 0V;

[0050] Figure 5 is the experimental data when using the control method of this invention and the initial voltage of the supercapacitor is 0V;

[0051] Figure 6 is the experimental data when using traditional PID control and the initial voltage of the supercapacitor is 40V;

[0052] Figure 7 is the experimental data when using the control method of this invention and the initial voltage of the supercapacitor is 40V;

[0053] Figure 8 It is the energy comparison histogram of experimental feedback. Specific implementation manner

[0054] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners:

[0055] Energy feedback method of an electro-excited doubly salient motor based on direct connection of supercapacitors: When the DSEM brakes and feeds back energy, there is no need to rely on the DC-DC converter that is relied on in conventional energy feedback. The supercapacitor is directly connected to the output end of the DSEM rectifier to absorb the energy recovered by braking. The constant feedback current control strategy is adopted. During energy feedback, the motor speed decreases and the supercapacitor voltage increases. The excitation current is adjusted to control the excitation intensity so that the feedback current is stabilized at the target value.

[0056] For current control, the conventional method uses a PID controller. The PID controller makes the current stable at the target value through continuous iterative operations. The traditional PID controller has two key defects. First, the PID controller performs control through iterative operations, and itself has hysteresis. In addition, during energy feedback, the motor speed and supercapacitor voltage change rapidly, and the performance of the traditional single-parameter PID control deteriorates when tracking rapidly changing working conditions, and there is a deviation between the target value and the actual value. When using numerical calculation methods, it is necessary to rely on the accurate modeling of the system, and the DSEM system is strongly nonlinear, and the derivation of traditional mathematical models is relatively difficult.

[0057] A multi-layer perceptron is a feedforward neural network, mainly composed of an input layer, a hidden layer and an output layer. The prediction accuracy of the network can be improved by optimizing the weights and biases of neurons. It is widely used in function approximation, classification, and regression system modeling. The MLP has a very strong non-linear mapping ability and can approximate any continuous function, which is suitable for modeling the complex electromagnetic relationship of the motor. In addition, its strong generalization ability is suitable for different working conditions such as different loads, speeds and voltages of the DSEM.

[0058] The present invention uses the external characteristics of the DSEM system modeled based on a multi-layer perceptron to control the energy feedback of an electro-excited doubly salient motor based on direct connection of supercapacitors. First, a twin model of the DSEM prototype is built in Maxwell, and then the peripheral circuit is built in Simplorer to form a field-circuit coupling simulation platform. In the simulation platform, different motor external characteristic data are obtained by changing the excitation current, speed, load, etc., and the data is preprocessed. The multi-layer perceptron model is initialized, the data is divided into a training set and a validation set, the multi-layer perceptron is used to learn the training data, and the validation set is used to verify the model. Finally, the trained model is deployed to the embedded system after lightweight verification of the control performance and compared with the conventional PID controller.

[0059] According to the control method of the present invention, the specific implementation steps are as follows:

[0060] Step 1: Build a simulation platform:

[0061] Analyze the mechanical structure and winding structure of the DSEM prototype, including parameters such as the inner and outer diameters, number of pole pairs, air gap, number of winding turns, and core material of the stator and rotor. Build a motor finite element model in Maxwell. Build the peripheral circuit (three-phase rectifier bridge, excitation drive circuit) of the DSEM in Simplorer. Form a field-circuit coupling simulation platform of Maxwell + Simplorer.

[0062] Step 2: Collect and preprocess data:

[0063] Use a constant current load to fix the current on the output side of the DSEM system. Taking 20A as an example, divide the rotational speed from 3000rpm to 1000rpm at an interval of 50rpm. Continuously change the excitation current (0 - 6A) at each rotational speed point, collect the voltage on the output side of the DSEM system, and repeat the simulation by changing the rotational speed point to obtain the mapping data of the excitation current and the output side voltage at different rotational speeds. Figure 2 For the collected data, preprocess the collected data, including low-pass filtering and outlier removal.

[0064] Perform normalization processing on the data. The normalization formula is:

[0065] ;

[0066] x is the original data, x min is the minimum value of the data set, x max is the maximum value of the data set, x norm is the data after normalization processing.

[0067] Step 3: Train the MLP model:

[0068] Initialize the input layer to have 3 nodes (respectively the voltage U on the output side of the system C , the current I on the output side C and the rotational speed n of the motor)

[0069] The hidden layer adopts a two-layer structure. The dimension of the first hidden layer is n1, and the dimension of the second hidden layer is n2.

[0070] The output layer has one node (the reference value I of the excitation current fref )

[0071] The established MLP model is trained using training data, and minimizing the root mean square error (RMSE) is used as the optimization objective. After training, the model is verified on the validation set data. The model is trained multiple times, and the model with the minimum RMSE is selected as the final model for the external characteristics of DSEM.

[0072] The structure of the trained model is as follows:

[0073] The structure of the first layer:

[0074] ;

[0075] Among them, Z1 is the weighted sum of the first layer, W0 is the weight matrix of the first layer, b0 is the bias vector of the first layer, A1 is the output of the first layer, and the number of neurons in the first hidden layer is 168.

[0076] The structure of the second layer:

[0077] ;

[0078] Among them, Z2 is the weighted sum of the second layer, W1 is the weight matrix of the second layer, b1 is the bias vector of the second layer, A2 is the output of the second layer, and the number of neurons in the second hidden layer is 83.

[0079] The structure of the output layer is:

[0080] ;

[0081] Among them, Z3 is the weighted sum of the output layer, W2 is the weight matrix of the output layer, b2 is the bias vector of the output layer, and Z2 is the output of the output layer.

[0082] The root mean square error of the trained MLP model on the validation set is 0.005. Figure 3 The surface plot showing the fitting of the external characteristics of DSEM using the trained MLP model is shown.

[0083] Step 4: Deploy the model:

[0084] Extract the structure parameters, all weight parameters, and bias parameters of the trained MLP model, and deploy the extracted MLP model data to the main control DSP of the embedded system.

[0085] For the energy feedback control method based on MLP to model the external characteristics of DSEM, its control system block diagram is as Figure 1 shown. The supercapacitor is directly connected to the output side of the DSEM rectifier bridge to directly absorb the feedback energy. Because the terminal voltage of the supercapacitor cannot change suddenly, the voltage on the output side of DSEM is equal to the voltage of the supercapacitor. First, the system obtains the voltage U C of the supercapacitor through the voltage sensor, obtains the rotational speed n of the motor through the resolver, and sets the target feedback current ICref , input U C , n, and I Cref to the deployed MLP model to obtain the reference value I of the excitation current fref . I fref After being limited, it is sent to the excitation current PID controller, and finally the duty cycle d of the MOS tube on the excitation drive bridge is obtained.

[0086] Experimental comparison:

[0087] Set the comparison test with the traditional PID controller: The motor is driven by the prime mover to drop from 3000 r / min to 1000 r / min in 40 s, the feedback current is 20 A, and the initial voltages of the supercapacitor are 0 V and 40 V respectively. Figure 4 and Figure 5 are the experimental results when the traditional PID controller and the control method proposed in this invention are respectively adopted when the supercapacitor voltage is 0 V. Figure 6 and Figure 7 are the experimental results when the traditional PID controller and the control method proposed in this invention are respectively adopted when the supercapacitor voltage is 40 V.

[0088] When the initial voltage of the supercapacitor is 0 V: As can be seen from Figure 4 , when the PID regenerative current controller is adopted, the feedback current only rises to the target value of 20 A at 3 s, and the feedback current starts to decline at 34.2 s, but the excitation current only reaches the maximum value of 6 A at 37.9 s, and the final charging voltage of the supercapacitor is 61 V. As can be seen from Figure 5 , when the controller based on MLP modeling of the DSEM external characteristics is adopted, the feedback current quickly reaches the target value of 20 A when the energy feedback starts. The feedback current starts to decline at 33.3 s, the excitation current reaches the maximum value of 6 A at 33 s, and the final voltage of the supercapacitor is 67 V.

[0089] When the initial voltage of the supercapacitor is 40 V: As can be seen from Figure 6 , when the PID regenerative current controller is adopted, the feedback current only rises to the target value of 20 A at 3 s, and the feedback current starts to decline at 25 s, but the excitation current only reaches the maximum value of 6 A at 40.7 s, and the final charging voltage of the supercapacitor is 87 V. As can be seen from Figure 7 , when the controller based on MLP modeling of the DSEM external characteristics is adopted, the feedback current quickly reaches the target value of 20 A when the energy feedback starts. The feedback current starts to decline at 25.7 s, the excitation current reaches the maximum value of 6 A at 25.4 s, and the final voltage of the supercapacitor is 91 V.

[0090] It can be concluded from the experimental results that at the beginning of energy feedback, the controller based on the MLP modeling of the DSEM external characteristics has a faster rising feedback current than the traditional PID controller; in the later stage of energy feedback, the demand for excitation current increases rapidly, and the traditional PID controller has an error in tracking the target, resulting in that when the feedback current starts to drop, the excitation current has not reached the maximum value, reducing the stable time of the feedback current. The controller based on the MLP modeling of the DSEM external characteristics calculates the reference value of the excitation current in real time through an accurate model. Thus, at the beginning of feedback, the feedback current rapidly rises to 20 A. In the later stage of feedback, the model can accurately track the demand for excitation current. When the feedback current starts to drop, the excitation current reaches the maximum value at the same time.

[0091] The control method proposed by the present invention can improve the control speed and control accuracy of the feedback current in the energy feedback of the electro-excited doubly salient motor based on the direct connection of supercapacitors, thereby improving the energy feedback rate. Figure 8 It can be seen that the control method proposed by the present invention increases the energy feedback by 20.6% and 11.9% respectively compared with the PID control when the initial voltage of the supercapacitor is 0 V and 40 V.

[0092] The above are only the preferred embodiments of the present invention, and do not limit the present invention in any other form. Any modification or equivalent change made according to the technical essence of the present invention still belongs to the scope protected by the present invention.

Claims

1. A supercapacitor direct-connected energy feedback control method based on motor external characteristics, characterized by: The specific steps are as follows: Step 1: Build a simulation platform: Analyze the mechanical and winding structures of the DSEM prototype, build a motor finite element model in Maxwell, and build the DSEM peripheral circuits, including the three-phase rectifier bridge and excitation drive circuit, in Simplorer to form a Maxwell+Simplorer field-circuit coupling simulation platform. Step 2: Collect and preprocess data: A constant-current load was used to fix the output current of the DSEM system. The speed was divided into 50-rpm intervals from 3000 rpm to 1000 rpm. The excitation current was continuously varied from 0 to 6 A at each speed point. The voltage at the output of the DSEM system was collected. The simulation was repeated at different speed points to obtain mapping data of the excitation current and output voltage at different speeds. The collected data was preprocessed, including filtering, removing outliers, and normalization. Step 3: Train the MLP model: The speed, target feedback current, and supercapacitor voltage are used as model inputs, and the excitation current reference value is used as output. The data is divided into a training set and a validation set. The model is trained using the training set, and the model accuracy is verified on the validation set. The specific steps of step 3 are as follows: Initialize the input layer to 3 nodes, which are the system output side voltage U C , output side current I C and motor speed n; The hidden layer adopts a two-layer structure, the first hidden layer dimension is n1, and the second hidden layer dimension is n2; A node in the output layer is the excitation current reference value I fref ; The established MLP model is trained using the training data, with minimizing the root mean square error (RMSE) as the optimization goal. After the training is completed, the model is verified on the validation set data. The model is trained multiple times, and the model with the smallest RMSE is selected as the final external characteristic model of DSEM. The structure of the trained 3-layer MLP model is as follows: First layer structure: ； Z1 is the weighted sum of the first layer, W0 is the weight matrix of the first layer, is the input data of the first layer, b0 is the bias vector of the first layer, and A1 is the output of the first layer; Second layer structure: ； Z2 is the weighted sum of the second layer, W1 is the weight matrix of the second layer, b1 is the bias vector of the second layer, and A2 is the output of the second layer; The output layer structure is: ； Z3 is the weighted sum of the output layer, W2 is the output layer weight matrix, and b2 is the output layer bias vector; is the activation function, where x is the original data; Step 4: Deploy the model: Extract the structural parameters, all weight parameters, and bias parameters of the trained MLP model, and deploy the extracted MLP model data to the main control DSP of the embedded system; In step 4, the specific process is as follows: 1) Extract the structural parameters of the trained MLP model and deploy the model algorithm and structural parameters to the DSP controller; 2) Construct a control method based on MLP modeling of the DSEM external characteristics: The system collects supercapacitor voltage, motor speed, and target feedback current in real time. The supercapacitor voltage is used as the DSEM output voltage, and the target feedback current is used as the output current. The data is input into the MLP model, and the model calculates the output excitation current reference value to stabilize the feedback current at the target value.

2. The supercapacitor direct-connected energy feedback control method based on motor external characteristics according to claim 1 is characterized in that: The parameters of the mechanical structure and winding structure of the DSEM prototype in step 1 include the inner and outer diameters of the stator and rotor, the number of pole pairs, the air gap, the number of winding turns, and the core material.

3. The supercapacitor direct-connected energy feedback control method based on motor external characteristics according to claim 1 is characterized in that: The step 2 normalizes the collected data as follows: Normalize the data, the normalization formula is: ； x is the original data, x min is the minimum value of the data set, x max is the maximum value of the data set, x norm The data are normalized.

Images