Permanent magnet synchronous motor parameter identification method, recording medium and system

Abstract

The invention belongs to the technical field of motor control, and particularly relates to a permanent magnet synchronous motor parameter identification method, which comprises the following steps of: acquiring voltage, current and speed data of a motor in two states that direct-axis current is equal to 0 and not equal to 0 in the same period as input characteristic quantities through a permanent magnet synchronous motor vector control system; synchronously collecting motor stator resistance, direct-axis inductance, quadrature-axis inductance and permanent magnet flux linkage as output characteristic quantities for training a back propagation neural network model; and after reverse neural network training is completed, motor parameter identification is completed by inputting related characteristic quantities in real time. According to the method, the training efficiency and generalization ability of the neural network model are improved, the problems that a traditional parameter identification scheme depends on high modeling precision and low identification precision are solved, and the method is suitable for parameter identification behaviors of the permanent magnet synchronous motor under various working conditions. The invention also provides a non-transitory readable recording medium storing the program of the method and a system comprising the medium, and the program can be called through a processing circuit to execute the method.

Description

Technical Field

[0001] The invention belongs to the technical field of motor control and discloses a permanent magnet synchronous motor parameter identification method, recording medium and system. Background Art

[0002] Permanent magnet synchronous motors (PMSMs) offer advantages such as high precision, high speed, high dynamic response, high reliability, and high efficiency. With the rapid development of hardware such as permanent magnet materials, integrated chips, and switching devices, as well as the rapid advancement of motor control, variable frequency modulation, and data processing technologies, they have found widespread application in high-end CNC machine tools, high-performance conveying equipment, and high-efficiency rail transit. However, PMSMs are complex, multivariable, tightly coupled, and nonlinear systems, and their parameters are susceptible to temperature, wear, and aging. Failure of control strategies to identify and update motor parameters in real time can lead to reduced control accuracy and stability. Therefore, research on parameter identification for PMSMs is of great significance.

[0003] Currently, the most widely used motor parameter identification strategies include the least squares method, the model reference adaptive method, the Kalman filter method, and artificial intelligence methods. The least squares method has high requirements for the model, assuming that the model is linear and normally distributed. It is sensitive to abnormal data and easily interfered by noise or outliers. It has limited data processing capabilities and is only applicable to models with a small number of parameters and a simple structure. The model reference adaptive method has high requirements for the reference model and controller design, the adaptive rate design is complex, parameter initialization is difficult, and the calculation and debugging workload is large. The Kalman filter method has limited predictive capabilities for nonlinear systems, and modeling requires high accuracy of the system dynamic model and state observation equation.

[0004] Although AI has strong nonlinear fitting capabilities and can be used for parameter identification to fit complex nonlinear relationships in motor systems, and is driven by motor data, it can process information flexibly and conveniently. It has strong self-learning and adaptive capabilities, automatically extracting highly correlated data from massive amounts of data and automatically adapting to various motor operating conditions. It also has high data processing efficiency and computing power. However, using AI to obtain motor parameters also has the following problems:

[0005] If the types of training data are not planned and the AI ​​relies entirely on unsupervised self-learning, the training time will be longer, and there will be problems with low recognition accuracy due to reasons such as lack of rank in the input data, strong coupling of input features, sensitivity to initial training values, and uneven sample distribution.

[0006] Moreover, during actual operation, the motor operates under certain working conditions, and the extracted data is difficult to cover all working conditions. The poor generalization ability of the model is also a problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0007] In view of the above problems, the present invention provides a method for identifying parameters of a permanent magnet synchronous motor, comprising the following steps:

[0008] S1. Using a permanent magnet synchronous motor vector control system, collect the motor's voltage, current, and speed data in the same cycle when the direct-axis current is equal to zero and not equal to zero as input features. Also collect the motor's stator resistance, direct-axis inductance, quadrature-axis inductance, and permanent magnet flux linkage in the same cycle as output features. Multiple input feature / output feature sample pairs from different cycles are used to train a back-propagation neural network model.

[0009] S2. Optimize the backpropagation neural network model using the constrained gradient descent algorithm during training;

[0010] S3. Input the voltage, current, and speed data in the two states of direct-axis current equal to 0 and not equal to 0 into the trained back-propagation neural network model in real time to obtain the corresponding stator resistance, direct-axis inductance, quadrature-axis inductance, and permanent magnet flux data to complete the motor parameter identification.

[0011] Collecting the operating characteristics under dual states to construct input samples can ensure that the training data has the full rank property of the voltage model, realize the joint identification of multiple parameters and provide accuracy guarantee.

[0012] Preferably, the method further includes the step of continuously injecting a current signal with a frequency of 0.1-5 kHz, an amplitude of 5%-20% of the rated current, and a duty cycle of 5%-50% into the d-axis during the sampling process.

[0013] Injecting high-frequency, small-amplitude current can generate clear position-related signals that reflect the characteristics of each motor parameter without significantly affecting the normal operation of the motor, ensuring the quantity and quality of data collection.

[0014] Preferably, the back-propagation neural network model uses a cascade structure to enhance the network topology, and adds a cross-layer connection between the input layer and the output layer on the basis of conventional feedforward connections.

[0015] This network structure is used to strengthen the direct mapping relationship between high-dimensional features and adapt to the mapping requirements of nonlinear strongly coupled operating parameters in permanent magnet synchronous motors.

[0016] Preferably, the method further includes the step of optimizing the initial weights and biases of the neural network through population initialization, fitness evaluation or cross-mutation operation.

[0017] This optimization method improves the global optimization performance of model initialization and reduces the risk of falling into local optimality in complex motor sample training.

[0018] Preferably, a probability threshold is set according to the proportion of the motor's non-steady-state operation, and neurons and their connections that exceed the set probability threshold are discarded during the training process.

[0019] This screening method avoids overfitting of the model to local features, enhances its generalization ability for dynamic operating scenarios such as speed and load changes during motor operation, and reduces the risk of overfitting.

[0020] Another embodiment of the present invention provides a non-transitory readable recording medium for storing one or more programs including a plurality of instructions. When the instructions are executed, the processing circuit will execute the above-mentioned permanent magnet synchronous motor parameter identification method.

[0021] Another embodiment of the present invention is to provide a permanent magnet synchronous motor parameter identification system, comprising a processing circuit and a memory electrically coupled thereto, wherein the memory is configured to store at least one program, wherein the program includes a plurality of instructions, and the processing circuit runs the program to execute the above-mentioned permanent magnet synchronous motor parameter identification method.

[0022] Compared with the prior art, the method, recording medium, and system for identifying parameters of a permanent magnet synchronous motor provided by the present invention have the following beneficial effects:

[0023] The present invention designs a current injection mechanism to enable the permanent magnet synchronous motor to switch between the two states of direct-axis current equal to 0 and not equal to 0, constructs sample pairs containing dual-state operation characteristics, ensures that the training data meets the full-rank condition of the voltage model, and thus supports the simultaneous high-precision identification of multiple key motor parameters (resistance, direct-axis / quadrature-axis inductance, permanent magnet flux).

[0024] The present invention adopts a data-driven approach, breaking away from the traditional identification method's reliance on precise motor modeling, reducing the requirements for professional knowledge, simplifying the parameter identification process, and improving the applicability and engineering deployability of the identification strategy.

[0025] In view of the characteristics of permanent magnet synchronous motor operating data such as high-dimensional coupling, initial value sensitivity and operating condition fluctuation, the present invention uses a cascade structure to enhance the feature expression capability, a genetic algorithm to improve the initialization global optimization performance, and a discarding strategy to improve the generalization ability of the model under variable operating conditions. The three functions complement each other and are optimized synergistically, making the model significantly superior to traditional single optimization strategies in terms of training efficiency, accuracy and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of a method for identifying parameters of a permanent magnet synchronous motor according to an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of data flow of control of a permanent magnet synchronous motor according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of d-axis current setting and data sampling in an embodiment of the present invention;

[0029] Figure 4 It is a schematic diagram of the evolution of the common BP neural network structure to the optimized neural network structure of the present invention;

[0030] Figure 5 is a schematic diagram of a cascade structure in an embodiment of the present invention;

[0031] Figure 6 1 is a schematic diagram of simulation identification results of a CGD-BPNN (constrained gradient descent algorithm-backward neural network) motor parameter identification model according to an embodiment of the present invention;

[0032] Figure 7 It is a schematic diagram of the experimental identification results of the CGD-BPNN motor parameter identification model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings. The described embodiments are part of the embodiments of the present invention, but not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without making any innovative efforts shall fall within the scope of protection of the present invention.

[0034] like Figure 1-5 As shown, the present invention provides a method for identifying parameters of a permanent magnet synchronous motor, comprising the following steps:

[0035] (1) According to the operation requirements of the permanent magnet synchronous motor, in the permanent magnet synchronous motor vector control system, a current excitation signal is given to make the motor run at i d =0 and i d ≠0 state, and synchronously collect the operating parameters in the two states to form a set of sample input X={u d0 、u q0 、i q0 、v0、u d1 、u q1 、i d1 、i q1 , v1}, the corresponding output is Form training data pairs (X, Y) to ensure that the input samples have the full rank characteristics of the voltage model and realize the joint identification of multiple motor parameters;

[0036] Among them, d represents the direct axis, q represents the quadrature axis, and u d0 、u q0 、i q0 、v0 are id = 0 (state 0) when d, q axis voltage, q axis current and speed; u d1 、u q1 、i d1 、i q1 、v1 are i d ≠0 (state 1) d, q axis voltage, d, q axis current and speed; R s 、L d 、L q 、 They are stator resistance, d-axis inductance, q-axis inductance and permanent magnet flux linkage, respectively.

[0037] (2) Using the constructed sample data to train a motor multi-parameter identification model of a multi-strategy optimization neural network, wherein the cascade algorithm, genetic algorithm and Dropout algorithm are used to comprehensively optimize the BP neural network. It should be noted that the cascade structure, genetic algorithm and Dropout strategy are not used in random combination, but are functionally divided and collaboratively optimized for typical features such as high-dimensional coupling, sensitivity to training initial values ​​and sample imbalance in motor operation data, thereby constructing a multi-parameter identification neural network model that is suitable for complex motor operation scenarios and improving the overall training stability and generalization ability;

[0038] (3) The trained CGD-BPNN model is encapsulated as an independent module and can be embedded in the motor control system as an online parameter identification unit. It accepts the current, voltage and speed data generated by the operation of the permanent magnet synchronous motor with unknown motor parameters. The multi-strategy optimization neural network automatically outputs the identification values ​​of the permanent magnet synchronous motor resistance, inductance and flux linkage parameters in real time, and has good control interface adaptability and online deployment capability.

[0039] like Figure 2 As shown, in step (1), the permanent magnet synchronous motor vector control system adopts a three-closed loop structure, including a current loop, a speed loop and a position loop. The trajectory planning module gives the motion trajectory of the permanent magnet synchronous motor rotor. The coordinate transformation module transforms the physical quantities between the three-phase coordinate system, the two-phase rotating coordinate system and the two-phase stationary coordinate system. The SVPWM module converts the voltage in the two-phase stationary coordinate system into a signal for controlling the inverter module through space vector pulse width modulation. The inverter module outputs a three-phase current to control the normal operation of the permanent magnet synchronous motor. The detection module uses a magnetic grid to detect the position and speed of the motor rotor. The CGD-BPNN parameter identification module receives the current, voltage and rotor speed in the two-phase rotating coordinate system, and outputs the resistance, inductance and magnetic flux of the motor parameters to be identified.

[0040] like Figure 3As shown, in step (1), the motor parameters are estimated by collecting easily measurable electrical analog quantities, such as stator voltage and current, rotor position and speed, combined with the motor mathematical model. Usually, current, voltage, speed, position, etc. are used as known quantities to estimate one or more of the parameters to be identified, such as resistance, inductance and flux linkage. Based on i d = 0, the voltage model of the permanent magnet synchronous motor is in a deficient rank state:

[0041]

[0042] in i d = 0 vector control strategy, by continuously injecting i into the d-axis at a certain frequency d The motor equation is expanded using a current signal that does not exceed 0. When injecting d-axis current, the frequency of the injected d-axis current is between 0.1kHz and 5kHz, and is small while ensuring the amount of data collected. The amplitude of the injected d-axis current is between 5% and 20% of the rated current, and is large within the weak magnetic field range allowed by the motor. The duty cycle of the injected d-axis current is between 5% and 50%, and is small while ensuring the amount of samples collected in each sampling cycle. When the motor is running in steady state, the full-rank motor equations for resistance, inductance, and flux are obtained:

[0043]

[0044] Among them, u d0 、u q0 、i q0 i d = 0 (state 0) when the d and q axis voltages and currents are d1 、u q1 、i d1 、i d1 i d ≠0 (state 1) d, q axis voltage, current. When processing the motor parameter identification sample data, the data synthesis frequency is consistent with the current injection frequency. In one cycle, the speed, d, q axis current, voltage and parameters to be identified in state 0 and state 1 form a group of valid samples. d0 、u q0 、i q0 、v0、u d1 、u q1 、i d1 、i q1 , v1} as the predictor variable of the neural network, the parameters to be identified As the characteristic response of the neural network, 9 characteristic inputs and 4 characteristic outputs constitute the sample data for training the neural network parameter identification model.

[0045] like Figure 4 and 5As shown in FIG, in step (2), the cascade algorithm, genetic algorithm and Dropout algorithm are used to comprehensively optimize the BP neural network. The cascade algorithm mainly optimizes the structure of the BP parameter identification model. On the basis of the layer-by-layer connection structure of the BP neural network, a direct connection channel from the input layer to the output layer is added, and more connections are introduced to increase the network capacity, which indirectly increases the weight of the variable that has a greater impact on the motor parameters to be identified among the 9 input variables, facilitates the extraction of effective features from the large amount of sample data generated by the motor operation, and improves the accuracy and expression ability of the parameter identification model:

[0046]

[0047] Among them, ω extra and O extra They represent the connection weights between the lth layer and different layers and the output values ​​of different layers respectively.

[0048] The genetic algorithm mainly searches for the optimal value of the initial weights randomly generated by the BP parameter identification model, iteratively evolves the optimal individual as the initial weight and threshold of the network, and ensures the diversity of the population through the selection, crossover, and mutation process, avoiding the training from tending to assimilation too early and falling into the local extreme value. The optimal weight is assigned to the BP parameter identification model to improve the training efficiency, accuracy and parameter identification precision of the model:

[0049]

[0050] Among them, S is the objective function; V is the fitness function. When S is 0, the initial weight is optimal. On the contrary, the larger S is, the worse the initial weight is. * is the reference value output by the model; last It is the true value output by the last layer of the network.

[0051] The Dropout algorithm mainly optimizes the training process of the BP parameter identification model. On the basis that all neurons in the BP neural network can be activated, the operation of discarding some neurons with a certain probability is added, which improves the training efficiency of the parameter identification model, avoids the identification results from being overly dependent on a certain feature variable, reduces the risk of model overfitting, and improves the generalization ability of the model when processing new data:

[0052]

[0053] Wherein, p is the probability of neuron inactivation; ω′ is the weight of the non-inactivated neuron. In the embodiment of the present invention, the probability of neuron inactivation is set to 1 / 2.

[0054] The BP neural network is the main body for fitting the multi-parameter identification model of the motor. Sample data consisting of different motor parameters and corresponding voltage, current, and speed are fed into the BP neural network model. After forward propagation, the predicted values ​​of resistance, inductance, and flux are obtained. The predicted parameter values ​​are compared with the actual values ​​to obtain the error. The error is then back-propagated layer by layer, and the model weights are updated according to the chain rule. This cycle repeats until the maximum number of iterations is reached or the prediction error of the motor parameters is reduced to the expected level:

[0055]

[0056] Among them, I j l is the input of the jth neuron in the lth layer; O is the output; ω ij l-1 is the connection weight from the i-th neuron to the j-th neuron in the l-1th layer; b is the bias; f is the activation function in the forward propagation of the BP neural network. Loss function and objective function:

[0057]

[0058] Among them, e is the loss function, which is used to measure the difference between the sample prediction result and the true label; E is the objective function, which is used to optimize the entire model. When E is 0, it means that the model is a perfect model with completely accurate prediction values. Conversely, the larger the value of E, the worse the model. Error back propagation process:

[0059]

[0060]

[0061] Among them, a is the learning rate, ω ij (l-1)l (t) is the weight from the i-th neuron in the (l-1)th layer to the j-th neuron in the lth layer at time t.

[0062] like Figure 6 、 7 As shown in step (3), the trained CGD-BPNN motor multi-parameter identification model is modularized and loaded into the permanent magnet synchronous motor control system for simulation verification and experimental verification. In the Simulink simulation verification, the motor parameters of the permanent magnet synchronous motor are set to new motor parameters that are not included in the training sample data, and the motor operating parameters {u d0 、u q0 、i q0 、v0、u d1 、u q1 、i d1 、i q1, v1} are sent to the CGD-BPNN motor multi-parameter identification model, and the permanent magnet synchronous motor resistance Rs, d-axis inductance Ld, q-axis inductance Lq and permanent magnet flux are output. Parameters. Figure 6 It can be seen from the identification results that the identification results of the parameter identification model for different new data always remain close to the given motor parameter values, indicating that the model can accurately identify the resistance, inductance and flux of the motor. Compared with the BP algorithm, the identification accuracy of the CGD-BPNN algorithm has been slightly improved. Taking the mean absolute error as the evaluation index, the resistance error accuracy has been improved by 32%, the inductance error accuracy has been improved by 39%, and the flux error accuracy has been improved by 34%, and the identification results are more accurate. In the semi-physical experiment verification, Figure 7 The identification results show that the mean absolute error between the motor parameters predicted by the parameter identification model and the actual motor parameters is much smaller than the actual values, and can be controlled within 0.8% of the actual values. Therefore, the motor parameter values ​​obtained by the multi-strategy optimization neural network parameter identifier can reflect the actual motor parameters. To maintain generality, the parameter identification model was trained by adding motor operating parameters at different speeds and loads to the training sample data, and identification was performed on new data under different operating conditions:

[0063]

[0064]

[0065] comprehensive Figure 6 and 7 The results show that the permanent magnet synchronous linear motor parameter identification method based on a multi-strategy optimization neural network has higher motor parameter identification accuracy, smaller fluctuations during the identification process, and higher model training efficiency than the conventional BP neural network. This method can more efficiently and accurately identify the resistance, inductance, and flux linkage values ​​of the permanent magnet synchronous linear motor. This confirms that the proposed permanent magnet synchronous linear motor parameter identification method based on a multi-strategy optimization neural network has higher model training efficiency and higher motor parameter identification accuracy.

[0066] Compiling the above-mentioned method steps into a program and then storing it on a hard disk or other non-transitory storage medium constitutes an embodiment of the present invention's "a non-transitory readable recording medium"; and electrically connecting the storage medium to a computer processor and completing the permanent magnet synchronous motor parameter identification through data processing constitutes an embodiment of the present invention's "a permanent magnet synchronous motor parameter identification system".

[0067] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computers containing computer-usable program code, or on available storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.).

[0068] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0069] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0071] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for identifying parameters of a permanent magnet synchronous motor, characterized in that: The following steps are involved: S1. Using a permanent magnet synchronous motor vector control system, collect the motor's voltage, current, and speed data in the same cycle when the direct-axis current is equal to zero and not equal to zero as input features. Also collect the motor's stator resistance, direct-axis inductance, quadrature-axis inductance, and permanent magnet flux linkage in the same cycle as output features. Multiple input feature / output feature sample pairs from different cycles are used to train a back-propagation neural network model. S2. Optimize the backpropagation neural network model using the constrained gradient descent algorithm during training; S3. Input the voltage, current, and speed data in the two states of direct-axis current equal to 0 and not equal to 0 into the trained back-propagation neural network model in real time to obtain the corresponding stator resistance, direct-axis inductance, quadrature-axis inductance, and permanent magnet flux data to complete the motor parameter identification.

2. A permanent magnet synchronous motor parameter identification method according to claim 1, characterized in that: The method also includes the step of continuously injecting a current signal with a frequency of 0.1-5 kHz, an amplitude of 5%-20% of the rated current, and a duty cycle of 5%-50% into the d-axis during the sampling process.

3. A permanent magnet synchronous motor parameter identification method according to claim 2, characterized in that: The back-propagation neural network model uses a cascade structure to enhance the network topology. On the basis of conventional feed-forward connections, cross-layer connections between the input layer and the output layer are added.

4. A permanent magnet synchronous motor parameter identification method according to claim 3, characterized in that: It also includes steps for optimizing the initial weights and biases of the neural network through population initialization, fitness evaluation or cross-mutation operations.

5. A permanent magnet synchronous motor parameter identification method according to claim 4, characterized in that: A probability threshold is set according to the proportion of non-steady-state operation of the motor, and neurons and their connections that exceed the set probability threshold are discarded during training.

6. A non-transitory readable recording medium for storing one or more programs comprising a plurality of instructions, characterized in that: When the instruction is executed, the processing circuit is caused to execute a permanent magnet synchronous motor parameter identification method according to any one of claims 1 to 5.

7. A permanent magnet synchronous motor parameter identification system, comprising a processing circuit and a memory electrically coupled thereto, characterized in that: The memory configuration stores at least one program, the program includes a plurality of instructions, and the processing circuit runs the program to execute a permanent magnet synchronous motor parameter identification method according to any one of claims 1-5.

Images