A switched reluctance motor control method based on neural network algorithm

Abstract

The present invention discloses a switched reluctance motor control method based on a neural network algorithm. First, the rated speed of the motor is determined according to the motor power and starting time, the motor is modeled and analyzed in finite element analysis software to obtain the electromagnetic parameters of the motor, the output torque of the motor under different chopping currents is analyzed by finite element analysis, and the chopping current amplitude is determined. On this basis, the product of the maximum average output torque and the torque-current ratio is maximized, and the optimal opening and closing angles are obtained through finite element simulation scanning. A set of nonlinear mappings with the motor speed as input and the opening and closing angles as output are obtained. The mapping is fitted through a BP neural network learning algorithm to obtain the optimal control curve at each speed. Finally, the control parameters are adjusted by the curve to control the operation of the motor. The method of the present invention simplifies the control model by flexibly adjusting the control parameters, and has higher output torque and smaller torque pulsation.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor control, and in particular relates to a switched reluctance motor control method based on a neural network algorithm. Background Art

[0002] Switched reluctance motors (SRMs), with their simple structure, reliable operation, wide speed regulation range, and flexible control strategies, have attracted considerable attention from scholars both domestically and internationally over the past three decades. They are widely used in a variety of common applications, including household appliances, electric vehicles, and aerospace. Furthermore, their superior performance also finds application in specialized applications.

[0003] The switched reluctance motor, with its inherent advantages, is very suitable as the drive motor of the energy storage pulse generator. It has a simple structure, low failure rate and high reliability; it has strong starting ability and can quickly start large inertia loads from a standstill; it has the ability to drive quickly and can drag the rotor to the rated speed as quickly as possible, and has the ability to quickly recover when the speed drops. It can also work reliably under the condition of frequent speed changes of the motor. Summary of the Invention

[0004] The present invention provides a switched reluctance motor control method based on a neural network algorithm, which can optimize control parameters to maximize the motor output and minimize the driving time while considering that the motor is quickly driven to a given speed with a large inertia load.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a switched reluctance motor control method based on a neural network algorithm, comprising the following steps:

[0006] Step 1: Determine the rated speed of the motor based on the switched reluctance motor power and starting time:

[0007]

[0008] Where ω max Indicates the maximum speed of the motor, ω N Indicates the rated speed of the motor, t indicates the time required for the motor to reach the maximum speed at different rated speeds, P N is the rated power of the motor, J represents the motor moment of inertia;

[0009] Step 2: Based on the actual physical and electrical parameters of the switched reluctance motor, the motor is modeled and analyzed in finite element analysis software. The electromagnetic parameters of the motor are obtained based on the finite element simulation model. The parameters are used to simulate and analyze the motor in MATLAB.

[0010] Step 3, determining the chopping current amplitude by analyzing the output torque of the switched reluctance motor under different chopping currents through finite element analysis;

[0011] Step 4: With the goal of maximizing the product of the maximum average output torque and the torque-to-current ratio, the optimal turn-on and turn-off angles at different speeds are obtained through finite element simulation scanning. First, different speeds, turn-on angles, and turn-off angles are set in the finite element software, and the output torque under different combinations of the three is simulated. The smaller the input step size of the speed, turn-on angle, and turn-off angle, the larger the simulation volume, and the more accurate the optimal control angle obtained.

[0012] Step 5: Fitting is performed through the BP neural network learning algorithm to establish the angle-speed model of the switched reluctance motor using the BP neural network learning algorithm

[0013]

[0014]

[0015] Where y j Indicates the value of the next layer of neurons, f() represents the transfer function, and if it is to the output layer, it represents the output after training, y' j represents the actual value, n represents the number of neurons in the current layer, w i Represents the weight of the current layer unit to the next layer, x i Indicates the number of input variables, E k Indicates calculation error;

[0016] The obtained BP neural network model consists of an input layer, a hidden layer and an output layer. The BP neural network model is trained and learned using training group data and sample data. Each actual motor speed input to the input layer obtains a set of corresponding opening and closing angles, thereby determining the number of hidden layers and hidden units of the BP neural network model. By continuously adjusting the threshold so that the calculation error is less than the threshold, the output layer finally obtains a set of nonlinear mappings with motor speed as input and opening and closing angles as output. After obtaining the optimal control curve at each speed, the controller adjusts the control parameters through the curve to control the operation of the switched reluctance motor.

[0017] Furthermore, the step 3 is specifically as follows: first, according to the relationship between torque, speed and power, the torque of the motor when it reaches the rated speed is determined to be T N =P N / ω N Then, according to the required rated torque, the motor output torque value at different currents is obtained through finite element simulation, which is used as the current value when calculating the rated torque to determine the chopping current amplitude.

[0018] Furthermore, step 5 also includes the forward propagation of sample data and the back propagation process of data error; the forward propagation is that the data passes through the input layer and the hidden layer, and finally outputs the result to the output layer; the back propagation of the data error is that after the forward propagation calculates the error between the fitting output and the true output, the error passes through the output layer and the hidden layer, and finally outputs the result to the input layer.

[0019] Furthermore, the step 5 is specifically as follows:

[0020] Step 1, data preprocessing: through the formula For the input variable x of the sample data ij and the output variable y i Perform standardization, where n ij is the normalized result. The normalized data range is [-1,1]. ij is the value of the jth feature of the i-th sample, x jmax and x jmin are the maximum and minimum values ​​of the data under the jth feature respectively;

[0021] Step 2, forward propagation of sample data: The output of the previous layer in the BP neural network model is first multiplied by the corresponding weight, then added to the threshold, and finally transformed by the neuron activation function to obtain the output of each neuron: y i =f(net)f(∑ j w ij n ij +b i ), where y i is the output value, f(net) is the activation function, n ij is the normalized result, w ij is the connection weight, b i is the threshold;

[0022] Step 3, back propagation of data error: The error starts from the output layer and propagates layer by layer to the relevant weights and thresholds between the hidden layer and the input layer using the error gradient descent method: △w ij =(l)E i , w ij =△w ij +w ij , where E i is the error, l is the learning rate, y i is the output value, T i is the actual value;

[0023] Repeat the above steps Step 1 to Step 3 until the termination condition is reached.

[0024] Furthermore, step 6 is also included: building a switched reluctance motor control model based on the MATALB platform, adjusting the control parameters according to the BP neural network learning algorithm in step 5, formulating a simulation control strategy, and observing the control effect.

[0025] The beneficial effects of the present invention are as follows: the control method of the present invention takes the product of the maximum average output torque of the switched reluctance motor at each speed and the torque-to-current ratio as the control target, uses the BP neural network learning algorithm to determine the optimal control strategy of the motor, and fully considers the working characteristics of the motor at different speeds. Compared with the existing control strategy, it not only simplifies the control model, but also has higher output torque and smaller torque pulsation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a modeling diagram of the switched reluctance motor of the present invention in finite element software;

[0027] Figure 2 are the electromagnetic parameters of the switched reluctance motor in an embodiment of the present invention;

[0028] Figure 3 is the current of the switched reluctance motor of the present invention under different chopping current amplitudes in the finite element analysis;

[0029] Figure 4 The phase current waveforms corresponding to different turn-on and turn-off angles at a certain speed in the finite element analysis of the switched reluctance motor of the present invention are shown;

[0030] Figure 5 The optimal control curve and the calculation curve obtained by the BPNN algorithm in the embodiment of the present invention are obtained;

[0031] Figure 6 The torque and torque ripple are obtained by MATLAB simulation analysis in the embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] An embodiment of the present invention provides a switched reluctance motor control method based on a neural network algorithm, comprising the following steps.

[0034] Step 1: Determine the rated speed of the motor based on the switched reluctance motor power and starting time (acceleration time):

[0035]

[0036] Where ω max Indicates the maximum speed of the motor, ω N Indicates the rated speed of the motor, t indicates the time required for the motor to reach the maximum speed at different rated speeds, P N is the rated power of the motor, and J is the motor's moment of inertia.

[0037] Step 2: Based on the actual physical parameters and electrical parameters of the switched reluctance motor, the motor is modeled and analyzed in finite element analysis software, and the electromagnetic parameters of the motor are obtained according to the finite element simulation model. The parameters are used to simulate and analyze the motor in MATLAB.

[0038] Finite element model such as Figure 1 The electromagnetic parameters obtained are shown as Figure 2 shown.

[0039] Step 3: Determine the chopping current amplitude by analyzing the output torque of the switched reluctance motor under different chopping currents through finite element analysis.

[0040] First, according to the relationship between torque, speed and power, the torque of the motor when it reaches the rated speed is determined to be T N =P N / ω N Then, according to the required rated torque, the motor output torque value under different currents is obtained through finite element simulation. The simulated output torque is equal to the current value when the rated torque is calculated, so as to determine the chopping current amplitude. The current waveform is as follows: Figure 3 shown.

[0041] Step 4: With the goal of maximizing the product of the maximum average output torque and the torque-to-current ratio, the optimal opening and closing angles at different speeds are obtained through finite element simulation scanning.

[0042] First, set different speed, turn-on angle, and turn-off angle inputs in the finite element software, and simulate the output torque under different combinations of the three. The smaller the speed, turn-on angle, and turn-off angle input step size, the larger the simulation amount, and the more accurate the optimal control angle obtained. The phase current waveform corresponding to different turn-on and turn-off angles of the motor at a certain speed is as follows: Figure 4 shown.

[0043] Step 5: Fitting is performed through the BP neural network learning algorithm to establish the angle-speed model of the switched reluctance motor using the BP neural network learning algorithm

[0044]

[0045]

[0046] Where y jIndicates the value of the next layer of neurons, f() represents the transfer function, and if it is to the output layer, it represents the output after training, y' j represents the actual value, n represents the number of neurons in the current layer, w i Represents the weight of the current layer unit to the next layer, x i Indicates the number of input variables, E k Indicates calculation error;

[0047] The resulting BP neural network model consists of an input layer, a hidden layer, and an output layer. The BP neural network model is trained using training data and sample data. Each actual motor speed input to the input layer generates a corresponding set of switch-on and switch-off angles. This determines the number of hidden layers and hidden units in the BP neural network model. By continuously adjusting the threshold to keep the calculation error below the threshold, the output layer ultimately generates a nonlinear mapping with motor speed as input and switch-on and switch-off angles as output. After obtaining the optimal control curve for each speed, the controller uses this curve to adjust the control parameters to control the operation of the switched reluctance motor. The specific steps are shown below.

[0048] Step 1, data preprocessing: through the formula For the input variable x of the sample data ij and the output variable y i Perform standardization, where n ij is the normalized result. The normalized data range is [-1,1]. ij is the value of the jth feature of the i-th sample, x jmax and x jmin are the maximum and minimum values ​​of the data under the jth feature respectively.

[0049] Step 2, forward propagation of sample data: data passes through the input layer, hidden layer, and finally outputs the result to the output layer. In the BP neural network model, the output of the previous layer is first multiplied by the corresponding weight, then added to the threshold, and finally transformed by the neuron activation function to obtain the output of each neuron: y i =f(net)f(∑ j w ij n ij +b i ), where y i is the output value, f(net) is the activation function, n ij is the normalized result, w ij is the connection weight, b i is the threshold.

[0050] Step 3, back propagation of data error: After forward propagation calculates the error between the fitted output and the true output, the error passes through the output layer, hidden layer, and finally outputs the result to the input layer. Starting from the output layer, the error is propagated layer by layer to the relevant weights and thresholds between the hidden layer and the input layer using the error gradient descent method: △w ij =(l)E i , w ij =△w ij +w ij , where E i is the error, l is the learning rate, y i is the output value, T i is the actual value.

[0051] Repeat the above steps Step 1 to Step 3 until the termination condition is reached.

[0052] The present invention obtains the optimal control angle of the switched reluctance motor at different speeds through a large number of simulations. This angle and speed are a strongly nonlinear mapping, which needs to be modeled through a BP neural network learning algorithm so that a set of corresponding opening and closing angles can be obtained for each motor speed input.

[0053] In this training, a three-layer BPNN model was used to build the angle-speed model. The input signal is the actual motor speed, and the output is the corresponding turn-on and turn-off angles. This is a single-input, dual-output nonlinear model. The neurons in the first and second hidden layers use a hyperbolic tangent sigmoid transfer function, while the output neurons use a linear (Purelin) transfer function.

[0054] This embodiment takes into account both modeling accuracy and convergence speed. The number of neurons in the first hidden layer and the second hidden layer are selected to be 6 and 8 respectively. After multiple training and iterations, the error MSE is 0.009, which meets the training requirements. The optimal control curve and the curve obtained by the neural network algorithm are as follows Figure 5 shown.

[0055] Step 6: Based on the MATALB platform, build a motor control model, adjust the control parameters according to the BP neural network learning algorithm obtained in step 5, formulate a simulation control strategy, and observe the control effect; the control results are as follows: Figure 6 shown.

[0056] Finally, from the simulation results, it can be seen that compared with the traditional control strategy, the starting time of this control strategy is advanced by 20s and the torque ripple is greatly reduced.

[0057] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A switched reluctance motor control method based on a neural network algorithm, characterized by: Includes the following steps Step 1: Get the rated speed of the motor based on the switched reluctance motor power and starting time: Where ω max Indicates the maximum speed of the motor, ω N Indicates the rated speed of the motor, t indicates the time required for the motor to reach the maximum speed at different rated speeds, P N is the rated power of the motor, J represents the motor moment of inertia; Step 2: Based on the actual physical and electrical parameters of the switched reluctance motor, a finite element simulation model is established to obtain the electromagnetic parameters of the motor for simulation analysis of the motor in MATLAB. Step 3, determining the chopping current amplitude by analyzing the output torque of the switched reluctance motor under different chopping currents through finite element analysis; Step 4: With the goal of maximizing the product of the maximum average output torque and the torque-to-current ratio, the optimal turn-on and turn-off angles at different speeds are obtained through finite element simulation scanning. First, different speed, turn-on angle, and turn-off angle inputs are set, and the output torque under different combinations of the three is simulated. The smaller the speed, turn-on angle, and turn-off angle input step size, the larger the simulation volume, and the more accurate the optimal control angle obtained. Step 5: Use BP neural network learning algorithm to establish the angle-speed model of the switched reluctance motor Where y j Indicates the value of the next layer of neurons, f() represents the transfer function, and if it is to the output layer, it represents the output after training, y' j represents the actual value, n represents the number of neurons in the current layer, w i Represents the weight of the current layer unit to the next layer, x i Indicates the number of input variables, E k Indicates calculation error; The obtained BP neural network model consists of an input layer, a hidden layer and an output layer. The BP neural network model is trained and learned using training group data and sample data. Each actual motor speed input to the input layer obtains a set of corresponding opening and closing angles, thereby determining the number of hidden layers and hidden units of the BP neural network model. By continuously adjusting the threshold so that the calculation error is less than the threshold, the output layer finally obtains a set of nonlinear mappings with motor speed as input and opening and closing angles as output, and obtains the optimal control curve at each speed. The control parameters are adjusted through this curve to control the operation of the switched reluctance motor.

2. The switched reluctance motor control method based on a neural network algorithm according to claim 1, characterized in that: In step 3, first, according to the relationship between torque, speed and power, the torque of the motor when it reaches the rated speed is determined to be T N =P N / ω N Then, according to the required rated torque, the motor output torque value at different currents is obtained through finite element simulation, which is used as the current value when calculating the rated torque to determine the chopping current amplitude.

3. The switched reluctance motor control method based on a neural network algorithm according to claim 2, characterized in that: The step 5 also includes the forward propagation of sample data and the back propagation process of data error; the forward propagation is that the data passes through the input layer and the hidden layer, and finally outputs the result to the output layer; the back propagation of the data error is that after the forward propagation calculates the error between the fitting output and the true output, the error passes through the output layer and the hidden layer, and finally outputs the result to the input layer.

4. The switched reluctance motor control method based on a neural network algorithm according to claim 3, characterized in that: The step 5 is specifically as follows: Step 1, through the formula For the input variable x of the sample data ij and the output variable y i Perform standardization, where n ij is the normalized result. The normalized data range is [-1,1]. ij is the value of the jth feature of the i-th sample, x jmax and x jmin are the maximum and minimum values ​​of the data under the jth feature respectively; Step 2: The output of the previous layer in the BP neural network model is first multiplied by the corresponding weight, then added to the threshold, and finally transformed by the neuron activation function to obtain the output of each neuron: y i =f(net)f(∑ j w ij n ij +b i ), where y i is the output value, f(net) is the activation function, n ij is the normalized result, w ij is the connection weight, b i is the threshold; Step 3: The error starts from the output layer and propagates layer by layer to the relevant weights and thresholds between the hidden layer and the input layer using the error gradient descent method: Δw ij =(l)E i , w ij =Δw ij +w ij , where E i is the error, l is the learning rate, y i is the output value, T i is the actual value; Repeat the above steps Step 1 to Step 3 until the termination condition is reached.

5. The switched reluctance motor control method based on a neural network algorithm according to claim 4, characterized in that: It also includes step 6: building a switched reluctance motor control model based on the MATALB platform, adjusting the control parameters according to the BP neural network, formulating a simulation control strategy, and observing the control effect.

Images