Induction motor efficiency optimization control method based on physical information neural network
By using physical information neural network and adaptive loss function in induction motor control, the limitations of traditional control algorithms in dealing with complex motor models and variable working conditions are solved, and efficient optimization and robustness of induction motors are achieved.
Patent Information
- Application Number
- CN202411777699.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-05
- Publication Date
- 2025-03-25
AI Technical Summary
Traditional induction motor control algorithms have limitations when dealing with complex motor models and variable operating conditions, especially when dealing with nonlinear characteristics and high-dimensional data, it is difficult to achieve efficient optimization.
Using an induction motor efficiency optimization control method based on physical information neural network, a physical information neural network including an input layer, a plurality of hidden layers and an output layer is designed, and the weight parameters in the loss function are adaptively adjusted using an uncertainty-based weighting method.
It improves the efficiency and robustness of the induction motor, enhances the training efficiency and generalization capabilities of the model, and ensures physical consistency.
Smart Images
Figure CN119696453A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of induction motor control, and in particular relates to an induction motor efficiency optimization control method based on physical information neural network. Background Art
[0002] Researching and developing new energy vehicles is a challenge and opportunity for the current automotive industry. Induction motors are widely used in industries, transportation, and household appliances due to their simple structure, low cost, and durability. The efficiency level of induction motors directly affects energy utilization and operating costs. Traditional induction motor control algorithms are mostly based on analytical methods, which minimize losses by solving the steady-state model of the motor. However, with the complexity of the motor model and the variability of operating conditions, traditional methods have limitations in dealing with nonlinear characteristics and high-dimensional data. Summary of the invention
[0003] The purpose of the present invention is to propose an induction motor efficiency optimization control method based on physical information neural network, which can improve the efficiency and robustness of the induction motor in the process of controlling the induction motor.
[0004] The present invention is achieved through the following technical solutions: An induction motor efficiency optimization control method based on physical information neural network includes the following steps: Step S1, obtaining an initial data set reflecting the operating state of the induction motor, and preprocessing the initial data set to obtain a data set, and dividing the data set into a training set, a verification set, and a test set; Step S2, designing a physical information neural network, the physical information neural network comprising an input layer, a plurality of hidden layers and an output layer, the input of the input layer comprising partial derivative characteristics of the total loss of the induction motor with respect to the excitation current; Step S3: Design the loss function of the physical information neural network as ,in, represents the data loss term related to the excitation current, represents the first physical loss term associated with the total losses of the induction motor, represents the second physical loss term related to the physical residual, , , Respectively represent , , The corresponding uncertainty parameter is , j =1,2,3; Step S4: Use the training set to train the physical information neural network to optimize the physical information neural network parameters and uncertainty parameters, and use the validation set and test set to evaluate the performance of the physical information neural network.
[0005] Furthermore, in step S1, the initial data set is simulation data or experimental data. The simulation data is generated by simulating the operating state of the induction motor under different working conditions using a mathematical model of the induction motor and professional simulation software. The experimental data is real induction motor operating data collected through an actual induction motor experimental platform.
[0006] Furthermore, in step S1, the preprocessing includes data cleaning and normalization, wherein data cleaning is used to fill missing values and remove outliers, and normalization is used to eliminate dimensional differences between different features.
[0007] Furthermore, in step S2, the number of neurons in the input layer is 6, and the input of the input layer also includes five features obtained according to the samples in the data set, namely timestamp, synchronous acceleration, speed, torque current and total loss, and the total loss includes stator copper loss, iron loss and rotor copper loss.
[0008] Furthermore, the physical information neural network includes three hidden layers, each of which has 8 neurons, the activation function used in the hidden layer is the Tanh function, the activation function used in the output layer is the Softplus function, and the predicted excitation current is output. .
[0009] Furthermore, in step S3, the first physical loss term is expressed by the formula Calculate, where N represents the number of samples, Indicates i The power loss of the induction motor considering the iron loss resistance of the sample, , , represents the stator resistance of the induction motor, represents the rotor resistance of the induction motor, Indicates i The synchronous angular velocity of samples, represents the inductance of the induction motor, represents the iron loss resistance of the induction motor, Indicates i The speed of the samples, Indicates i The rotor resistance of the samples.
[0010] Furthermore, in step S3, the second physical loss term is expressed by the formula Calculate, where represents the physical residual, Indicates that the i The actual excitation current of each sample.
[0011] Furthermore, in step S3, the data loss term is given by the formula calculate.
[0012] Furthermore, in step S3, in order to ensure , parameterized using an exponential function, namely , then the loss function is expressed as ,in, s j is to ensure that the parameters can be learned A learnable real parameter greater than 0, , , .
[0013] The present invention has the following beneficial effects: The present invention utilizes a physical information neural network with an input layer, multiple hidden layers and an output layer for optimization control, and designs a loss function that combines a data loss term, a first physical loss term and a second physical loss term. Each loss term is correspondingly provided with an uncertainty parameter for representing the noise level or uncertainty of the loss term. An uncertainty-based weighting method is used to adaptively adjust the weight parameters in the loss function according to the characteristics of each loss term, thereby effectively improving training efficiency and robustness, and having stronger generalization ability and physical consistency. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] The present invention is further described in detail below with reference to the accompanying drawings.
[0015] Figure 1 It is a flow chart of the present invention.
[0016] Figure 2 It is a principle block diagram of the present invention. DETAILED DESCRIPTION
[0017] like Figure 1 and Figure 2 As shown, the induction motor efficiency optimization control method based on physical information neural network is characterized by comprising the following steps: Step S1, obtaining an initial data set reflecting the operating state of the induction motor, and preprocessing the initial data set to obtain a data set, and dividing the data set into a training set, a verification set, and a test set; More specifically, the initial data set is simulation data or experimental data. The simulation data uses the mathematical model of the induction motor and professional simulation software (such as MATLAB / Simulink) to simulate the operating state of the induction motor under different working conditions to generate a large amount of simulation data. The experimental data is the real induction motor operating data collected through an actual induction motor experimental platform, including voltage, current, speed, torque, input power and output power, etc.
[0018] The dataset is saved in a CSV file and contains the following fields: Time: Timestamp used to record the sampling time, in seconds (s).
[0019] we (synchronous angular velocity ): Synchronous angular velocity, in radians per second (rad / s).
[0020] wr (speed ): Rotational speed, in radians per second (rad / s).
[0021] iqs (torque current ): Stator q-axis current, in amperes (A), used to generate electromagnetic torque.
[0022] ids (excitation current ): The stator d-axis current, in amperes (A), determines the size of the rotor magnetic flux.
[0023] Ptotal (total loss ): includes stator copper loss, iron loss and rotor copper loss, the unit is watt (W).
[0024] (Total loss right The simulation uses PI control to give A sine wave, giving A constant value, changing The values of (The same group is constant), then the total loss is , .
[0025] Preprocessing includes data cleaning and normalization. Data cleaning is used to fill missing values, remove outliers, and ensure data quality. Normalization is used to eliminate the dimensional differences between different features and prevent numerical instability caused by feature values that are too large or too small. Standardization or minimum-maximum normalization can be used. The normalization formula is ,in, is the mean of the feature, is the standard deviation of the feature.
[0026] For the division of data sets: Training set: used for model training, accounting for 70% of the total data.
[0027] Validation set: used for model parameter adjustment and to prevent overfitting, accounting for 15% of the total data.
[0028] Test set: used for final model evaluation, accounting for 15% of the total data.
[0029] Step S2, designing a physical information neural network, the physical information neural network comprising an input layer, a plurality of hidden layers and an output layer, the input of the input layer comprising partial derivative characteristics of the total loss of the induction motor with respect to the excitation current; The number of neurons in the input layer is 6. The input of the input layer also includes five features obtained based on the samples in the data set, namely timestamp, synchronous acceleration, speed, torque current and total loss. The total loss includes stator copper loss, iron loss and rotor copper loss. This derivative is particularly important in optimization problems because it tells us how to adjust To minimize , this derivative value is obtained by the formula. As input, physical information neural networks can learn about ’s changing trend, so as to more accurately predict the optimal ; The physical information neural network includes three hidden layers, each of which has 8 neurons. The activation function used in the hidden layer is the Tanh function, which has nonlinear approximation capability. The output layer has 1 output node, which outputs the predicted excitation current , the activation function used is the Softplus function, which ensures that the output is positive and consistent with the physical meaning.
[0030] Step S3: Design the loss function of the physical information neural network as , using an uncertainty-based weighting method, adaptively adjust the weight parameters in the total loss function to improve the training efficiency and performance of the model. For each loss term, a learnable uncertainty parameter is introduced , represents the uncertainty of the loss term; among them, represents the data loss term related to the excitation current, represents the first physical loss term associated with the total losses of the induction motor, represents the second physical loss term related to the physical residual, , , Respectively represent , , The corresponding uncertainty parameters (i.e., the loss terms , , ), is a regularization term to prevent Too large to ensure the numerical stability of the loss function. ,j =1,2,3; reflects the noise level or uncertainty of the corresponding loss term. If If the loss term is larger, it means that the noise of the loss term is larger, and the model will pay less attention to it (i.e. the corresponding becomes smaller), if Smaller means that the noise of the loss term is smaller, and the model will pay more attention to this loss term (that is, the corresponding To prevent Too large or too small, you can set The upper and lower limits of prevents the value from exceeding a reasonable range.
[0031] Specifically, the first physical loss term is given by the formula Calculate, where N Indicates the number of samples (a sample refers to a data point in the dataset that contains all features and the corresponding target variable. Each sample is an independent data point. During training, the number of samples here is the number of samples in the training set. During validation and testing, the number of samples here is the number of samples in the validation set and the number of samples in the dataset, respectively). Indicates i The power loss of the induction motor considering the iron loss resistance of the sample, , , represents the stator resistance of the induction motor, represents the rotor resistance of the induction motor, Indicates i The synchronous angular velocity of samples, represents the inductance of the induction motor, represents the iron loss resistance of the induction motor, Indicates i The speed of the samples, Indicates i The rotor resistance of the samples.
[0032] The second physical loss term is given by the formula Calculate, where represents the physical residual, Indicates that the i The actual excitation current of the samples. The specific calculation steps are: calculate :According to the current and , and the known induction motor parameters, calculate the , the process is prior art; Calculate physical residuals: Use the predicted and known Calculating physical residuals ; Calculating physical losses: Calculation right The partial derivative of is taken and its square is minimized to ensure that the power loss is as close to the minimum as possible.
[0033] The data loss term is given by the formula Calculation. The specific calculation steps are: Get the real excitation current : Get the true value of each sample from the data set; In the specific implementation, in order to ensure , parameterized using an exponential function, namely , then the loss function is expressed as ,in, s j is to ensure that the parameters can be learned A learnable real parameter greater than 0, , , .
[0034] Step S4: Use the training set to train the physical information neural network to optimize the physical information neural network parameters and uncertainty parameters, and use the validation set and test set to evaluate the performance of the physical information neural network.
[0035] During specific training, first initialize , in each training iteration, the total loss function is calculated, back propagation is performed, and the model parameters are updated and s j , and monitor s j to ensure that the changes are within a reasonable range.
[0036] The above description is only a preferred embodiment of the present invention, and therefore cannot be used to limit the scope of implementation of the present invention. That is, equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the specification should still fall within the scope covered by the patent of the present invention.
Claims
1. An induction motor efficiency optimization control method based on physical information neural network, characterized in that: The steps include: Step S1, obtaining an initial data set reflecting the operating state of the induction motor, and preprocessing the initial data set to obtain a data set, and dividing the data set into a training set, a verification set, and a test set; Step S2, designing a physical information neural network, the physical information neural network comprising an input layer, a plurality of hidden layers and an output layer, the input of the input layer comprising partial derivative characteristics of the total loss of the induction motor with respect to the excitation current; Step S3: Design the loss function of the physical information neural network as ,in, represents the data loss term related to the excitation current, represents the first physical loss term associated with the total losses of the induction motor, represents the second physical loss term related to the physical residual, , , Respectively represent , , The corresponding uncertainty parameter is , j =1,2,3; Step S4: Use the training set to train the physical information neural network to optimize the physical information neural network parameters and uncertainty parameters, and use the validation set and test set to evaluate the performance of the physical information neural network.
2. The induction motor efficiency optimization control method based on physical information neural network according to claim 1 is characterized in that: In step S1, the initial data set is simulation data or experimental data. The simulation data is generated by simulating the operating state of the induction motor under different working conditions using a mathematical model of the induction motor and professional simulation software. The experimental data is real induction motor operating data collected through an actual induction motor experimental platform.
3. The induction motor efficiency optimization control method based on physical information neural network according to claim 1 is characterized in that: In the step S1, the preprocessing includes data cleaning and normalization. Data cleaning is used to fill missing values and remove outliers, and normalization is used to eliminate dimensional differences between different features.
4. The induction motor efficiency optimization control method based on physical information neural network according to claim 1, 2 or 3, characterized in that: In step S2, the number of neurons in the input layer is 6, and the input of the input layer also includes five features obtained according to the samples in the data set, namely timestamp, synchronous acceleration, speed, torque current and total loss.
5. The induction motor efficiency optimization control method based on physical information neural network according to claim 1, 2 or 3, characterized in that: The physical information neural network includes three hidden layers, each of which has 8 neurons. The activation function used in the hidden layer is the Tanh function, and the activation function used in the output layer is the Softplus function. The predicted excitation current is output. .
6. The induction motor efficiency optimization control method based on physical information neural network according to claim 4 is characterized in that: In step S3, the first physical loss term is given by the formula Calculate, where N represents the number of samples, Indicates i The power loss of the induction motor considering the iron loss resistance of the sample, , , represents the stator resistance of the induction motor, represents the rotor resistance of the induction motor, Indicates i The synchronous angular velocity of samples, represents the inductance of the induction motor, represents the iron loss resistance of the induction motor, Indicates i The speed of the samples, Indicates i The rotor resistance of the samples.
7. The method for optimizing the efficiency of an induction motor based on a physical information neural network according to claim 6, characterized in that: In step S3, the second physical loss term is given by the formula Calculate, where represents the physical residual, Indicates that the i The actual excitation current of each sample.
8. The method for optimizing the efficiency of an induction motor based on a physical information neural network according to claim 7, characterized in that: In step S3, the data loss term is given by the formula calculate.
9. The induction motor efficiency optimization control method based on physical information neural network according to claim 8, characterized in that: In step S3, an exponential function is used for parameterization, that is, , then the loss function is expressed as ,in, s j is to ensure that the parameters can be learned A learnable real parameter greater than 0, , , .