Method for Identifying Parameters of a Synchronous Motor Based on a Neural Network

Abstract

A method for parameter identification of a synchronous motor based on a neural network, the method includes: (S1) establishing and training a first neural network model, and using the trained first neural network model to determine an intermediate result of a parameter to be identified of a synchronous motor; (S2) forming a second neural network model by adding a compensation module to the trained first neural network model; and (S3) training a second neural network model based on the intermediate result of the parameter to be identified, and using the trained second neural network model to determine a final result of the parameter to be identified. The technical solution also relates to a device and a computer program product for parameter identification of a synchronous motor based on a neural network.

Description

Technical Field

[0001] The present invention relates to a method for identifying parameters of a synchronous motor based on a neural network, a device for identifying parameters of a synchronous motor based on a neural network, and a computer program product.

Background Art

[0002] Synchronous motors have been attracting increasing attention in industrial applications due to their low cost, high efficiency, and high stability. However, when a synchronous motor operates under load, various motor parameters exhibit non-linear characteristics due to reasons such as magnetic saturation, temperature drift, and cross-coupling, and conventional linear models cannot accurately reflect the true dynamic characteristics of the motor. In such cases, parameter identification of the synchronous motor is necessary to achieve better control performance.

[0003] In the prior art, motor parameters are often modeled in the form of a look-up table, which is time-consuming and expensive and requires traversing the entire range of motor operating points. Furthermore, there are methods that use mathematical models or special equations to fit the non-linear inductance surface. However, these mathematical models or equations are often very complex, their coefficients are parameterized, and the curve plotting algorithm requires corresponding database support.

[0004] In this context, there is a need for alternative motor parameter identification methods that achieve good parameter fitting at a limited number of operating points.

Summary of the Invention

Problems to be Solved by the Invention

[0005] The object of the present invention is to provide a method for identifying parameters of a synchronous motor based on a neural network, a device for identifying parameters of a synchronous motor based on a neural network, and a computer program product, which solve at least some of the problems of the prior art.

Means for Solving the Problems

[0006] According to a first aspect of the present invention, a method for identifying parameters of a synchronous motor based on a neural network is provided. The method includes: S1) Establishing and training a first neural network model, and using the trained first neural network model to determine an intermediate result of the parameters to be identified of the synchronous motor; S2) Forming a second neural network model by adding a compensation module to the trained first neural network model; S3) Training the second neural network model based on the intermediate result of the parameters to be identified, and using the trained second neural network model to determine the final result of the parameters to be identified. The method includes the steps above.

[0007] In particular, the present invention relates to the following technical concept, that is, a method for establishing a parameter model of a motor using a neural network. This utilizes the self-learning ability and fuzzy logic of an artificial neural network to optimize and estimate model parameters when there are limited observation data and a small number of operating points, thus improving the efficiency and accuracy of parameter identification. Furthermore, a two-stage model training process is proposed for parameter estimation, which enables the assignment of different task priorities and better adaptation to various characteristics of unknown parameters.

[0008] Optionally, the intermediate result is a linear approximation value of a first parameter among the parameters to be identified, and The first estimated value of the second parameter, which is less than the linear approximation of the first parameter, and is included.

[0009] Optionally, the final result is the linear approximation value of the first parameter among the parameters to be identified, after compensating for the non-linearity, and the second estimated value of the second parameter among the parameters to be identified, after compensating for the non-linearity of the linear approximation value of the first parameter, and is included.

[0010] The present invention achieves the technical advantage that the decoupling analysis of the linear characteristics and non-linear characteristics of the unknown parameters of the motor is realized by using the parameter identification process of the first neural network model and the second neural network model, and the identification principle is simplified.

[0011] Optionally, different training data is used for training the first neural network model and the second neural network model.

[0012] The linear characteristics of the unknown parameters are adapted based on the observation data under magnetic unsaturation, and the observation data under magnetic saturation is used for non-linear compensation, so the advantage of accelerating the network convergence process is particularly achieved.

[0013] Optionally, the parameters to be identified of the synchronous motor include the stator resistance, d-axis inductance, q-axis inductance and permanent magnet flux of the synchronous motor.

[0014] Different from the parameter identification method using a look-up table, the proposed scheme can achieve the synchronous identification of multiple motor parameters, thus particularly achieving the technical advantage of eliminating the need for additional sensors for pre-detecting some fixed parameters and saving time and cost.

[0015] Optionally, step S1 includes establishing an input-output relationship of the first neural network model based on the static voltage state equation of the synchronous motor in the d-q coordinate system, where the q-axis current, d-axis current, and electrical angular velocity of the synchronous motor are used as inputs to the first neural network model, and the q-axis voltage and d-axis voltage are used as outputs of the first neural network model, and the weights of the first neural network model are at least partially represented through the parameters to be identified.

[0016] Through supervised learning, artificial neural networks have the technical advantage of having a good input-output mapping relationship, generating reasonable outputs for new inputs, continuously optimizing their performance during the learning process, effectively utilizing existing data patterns with very limited samples, and thereby accelerating the parameter identification process.

[0017] Optionally, step S1 is to obtain known observation data of the synchronous motor, where the known observation data includes the d-axis current, q-axis current, d-axis voltage, q-axis voltage, and electrical angular velocity of the synchronous motor measured at different operating points, and to train the first neural network model using the known observation data, adjusting the weights of the first neural network model until the loss function satisfies a first predetermined condition, and when the loss function satisfies the first predetermined condition, determining an intermediate result of the parameters to be identified of the synchronous motor based on the weights of the first neural network model, and including.

[0018] During this period, the unknown parameters of the system are reflected in the internal weights, thereby particularly achieving the technical advantage of avoiding direct identification of the unknown parameters of the controlled object.

[0019] Optionally, adjusting the weights of the first neural network model includes iteratively updating the weights of the first neural network model using a gradient descent algorithm, particularly the backpropagation algorithm.

[0020] Through the above algorithms, "reasonable" solution rules are automatically extracted based on known input-output relationships, thus particularly achieving the technical advantage of automatically determining internal weights, solving mapping problems with complex internal mechanisms, and enabling the network to have good generalization and fault tolerance capabilities.

[0021] Optionally, step S2 includes using the q-axis current and the d-axis current as inputs to a third neural network model, and using the non-linear characteristic part of the parameter to be identified as the output of the third neural network model to establish the third neural network model as a compensation module.

[0022] By introducing the compensation module, it is possible to reasonably compensate for the non-linear factors included by assuming parameter linearity, and it is also possible to create a dedicated mathematical model for the non-linear characteristics of the parameters, thus particularly achieving the technical advantage of making the final parameter identification result more reliable.

[0023] Optionally, step S2 includes adding the compensation module to the first neural network model by sharing a part of the input layer of the first neural network model with the input layer of the compensation module, and connecting the output layer of the compensation module to the output layer of the first neural network model.

[0024] By embedding a compensation module into a first neural network model through common inputs and outputs, a technical advantage is particularly achieved in that the compensation module is passively trained during the overall training process of a second neural network model. As a result, the unknown non-linear characteristics of the parameters are converted into the internal parameters of the solution model. When there is no data available for direct training, the input-output mapping relationship of the compensation module is indirectly established through the overall backpropagation ability of the network.

[0025] Optionally, step S3 is training a second neural network model using known observation data, adjusting the weights of the trained first neural network model and the weights of the compensation module until the loss function satisfies a second predetermined condition, determining a final result of the parameters to be identified for the synchronous motor based on the weights of the trained second neural network model and the weights of the compensation module when the loss function satisfies the second predetermined condition, and including.

[0026] The training of the second neural network model is based on the results of the training in the first stage, which particularly achieves the technical advantages of enabling improvement based on the intermediate results of the parameters to be identified, improving time efficiency, and accelerating the convergence process.

[0027] Optionally, during the adjustment of the weights of the trained first neural network model, the intermediate result of the first parameter among the parameters to be identified is kept unchanged, and the intermediate result of the second parameter among the parameters to be identified is changed.

[0028] The parameter estimation in the first stage is completed with less than a linear approximation of a specific parameter. Therefore, in order to distinguish well from the non-linear characteristic part during the compensation stage, it can be considered constant and not participate in further training processes, and a technical advantage is particularly achieved. However, since this linear approximation is not accurate enough, the intermediate results of specific fixed parameters may inaccurately include the non-linear contributions of specific parameters. In order to compensate for the parts that do not belong to these fixed parameters, it is necessary to adjust these fixed parameters in the second stage to remove interference.

[0029] Optionally, the loss function is represented by the following equation.

Number

[0030] In the formula, N is the number of known observation data used, and U sd is the d-axis voltage of the synchronous motor in the known observation data,

Number

Number

[0031] A technical advantage is particularly achieved that the error between the model output and the actual output can be effectively considered and utilized, enabling the model to continuously self-learn in the direction of approaching the actual value, and providing a reliable evaluation criterion and termination condition for the training process.

[0032] Optionally, the method To control the operation of the synchronous motor based on the final result, a step of transmitting the final result of the parameter to be identified to the controller of the synchronous motor; and / or When applying the weights of the second neural network model, enabling the internal neural network to determine the online final result of the parameter to be identified in real time based on the operating parameters of the synchronous motor, and the controller controlling the operation of the synchronous motor based on the online final result, a step of transmitting the weights of the trained second neural network model to the internal neural network of the controller of the synchronous motor further comprises.

[0033] By directly transmitting the motor parameters to the controller of the synchronous motor, the parameter generation process is kept offline, ensuring that the computational overhead and hardware requirements of the controller are low, and particularly achieving the technical advantage of achieving an overall cost-effective control scheme. By integrating the internal neural network into the controller and providing the trained weight values to the internal neural network, the controller can update the parameter identification results in real time according to different operating conditions, enabling more reliable motor control.

[0034] According to a second aspect of the present invention, a device for parameter identification of a synchronous motor based on a neural network is provided. The device is used to implement the method according to the first aspect of the present invention, a first identification module configured to establish and train a first neural network model and use the trained first neural network model to determine an intermediate result of the parameter to be identified of the synchronous motor; a modification module configured to form a second neural network model by adding a compensation module to the trained first neural network model; Based on the intermediate results of the parameters to be identified, train a second neural network model, and use the trained second neural network model to determine the final results of the parameters to be identified, a second identification module configured as such; comprising.

[0035] According to a third aspect of the present invention, a computer program product is provided, and the computer program product comprises a computer program configured to implement the method according to the first aspect of the present invention when executed by a computer.

Brief Description of the Drawings

[0036] The present invention will be described in more detail with reference to the accompanying drawings in order to provide a better understanding of its principles, features and advantages. The drawings include the following.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10a

Figure 10b

[0037] In order to provide a clearer understanding of the technical problems, technical solutions and beneficial technical effects of the present invention, the following description will be presented in more detail in conjunction with the accompanying drawings and a plurality of exemplary embodiments. Naturally, the specific embodiments described herein are provided only for the purpose of explaining the present invention and do not limit the scope of protection of the present invention.

[0038] FIG. 1 shows a flowchart of a method for identifying parameters based on a neural network of a synchronous motor according to an exemplary embodiment of the present invention.

[0039] In step S1, a first neural network model is established and trained to determine an intermediate result of the parameters to be identified of the synchronous motor.

[0040] The synchronous motor may include, in the context of the present invention, a synchronous reluctance motor, a permanent magnet synchronous motor, an induction motor, and other types of synchronous motors having parameter non-linearity.

[0041] In the context of the present invention, the parameters to be identified of the synchronous motor are, for example, the stator resistance R, the d-axis inductance L d , the q-axis inductance L q , and / or the permanent magnet flux linkage φ fmay be included. Among these parameters, the stator resistance R and the permanent magnet flux linkage φ f can be regarded as the inherent parameters of the synchronous motor, and thus do not change with the control of the external temperature. However, due to the iron core saturation phenomenon and material factors that are difficult to avoid, the d-axis inductance L d of the synchronous motor and the q-axis inductance L q change with the change of current. Treating them as constants will inevitably affect the control accuracy and stability of the motor.

[0042] In the context of the present invention, the intermediate results of the parameters to be identified include, for example, the linear approximation values of the first parameters to be identified (for example, the d-axis inductance L d and the q-axis inductance L q and). Furthermore, these intermediate results also include the first estimated values of the second parameters among the parameters to be identified (for example, the stator resistance R and the permanent magnet flux linkage φ f and), considering the linear approximation of the first parameters.

[0043] In this step, the input-output relationship of the first neural network model is established based on, for example, the static voltage state equation of the synchronous motor in the d-q coordinate system. Considering an example of a permanent magnet synchronous motor, the static voltage state equation used is as follows.

Equation

[0044] In the formula, U sd is the d-axis voltage of the synchronous motor, R is the stator resistance of the synchronous motor, ω1 is the electrical angular velocity of the synchronous motor, L q is the q-axis inductance of the synchronous motor, i sq is the q-axis current of the synchronous motor, |φ f | is the amplitude of the permanent magnet flux linkage of the synchronous motor, θ is the zero angle of the synchronous motor, U sqis the q-axis voltage, and i sd is the d-axis current of the synchronous motor, and L d is the d-axis inductance.

[0045] To establish the first neural network model, for example, the q-axis current, d-axis current, and angular velocity of the synchronous motor are used as inputs to the first neural network model, and the q-axis voltage and d-axis voltage are used as outputs of the first neural network model. As parameters to be identified, R, L d 、L q and φ f characterize the adjustable weights of the first neural network model using them.

[0046] In step S2, the second neural network model is formed by adding a compensation module to the trained first neural network model. To account for the non-linearity of the parameters to be identified, for example, a third neural network model uses the q-axis current and d-axis current as inputs and uses the non-linear characteristic portions of the d-axis inductance and q-axis inductance to be identified as outputs to establish the compensation module.

[0047] In step S3, the second neural network model is trained based on the intermediate results derived from the parameters to be identified. The final result of the parameters to be identified is determined using the trained second neural network model. In the context of the present invention, the term "final result" refers to a result that can characterize the actual value or actual functional relationship of the parameters to be identified. For example, the final result of the parameters to be identified may include the linear approximation values of the first parameters (e.g., d-axis inductance and q-axis inductance) after being compensated for non-linearity. Further, the final result, after compensating for the non-linearity of the linear approximation values of the first parameters, for the second estimated values of the second parameters (e.g., stator resistance R and permanent magnet flux linkage φf) among the parameters to be identified, compared with the previously determined first estimated values, also includes that the contribution of the non-linear characteristic portion of the first parameters is substantially eliminated or only slightly included. By excluding this portion, the second estimated values of the parameters approach their true values more closely. In this step, for example, by using the second neural network model to update the intermediate results of the already determined parameters to be identified, the identification equation for each unknown parameter can be determined based on the updated results.

[0048] Figure 2 shows a flowchart of the method steps of the method shown in Figure 1. In this embodiment, method step S1 of Figure 1 includes steps S11 to S18 as an example.

[0049] In step S11, observation data is obtained. Considering a permanent magnet synchronous motor as an example, the stable operation of the permanent magnet synchronous motor is first controlled. Next, by changing the reference current, load torque, motor speed, and other conditions, N sets of different operating points are traversed in as wide an operating range as possible, and the d-axis voltage U sd , q-axis voltage U sq , d-axis current i sd , q-axis current i sqAnd the electrical angular velocity ω1 is recorded at these operating points. Here, the value of N can usually be freely adjusted within the range from 15 to 30 according to the time consumption and the desired model accuracy.

[0050] In step S12, the d-axis current i in the data set sd , the q-axis current i sq and the electrical angular velocity ω1 are extracted from the observation data and input into the first neural network model through forward propagation.

[0051] In step S13, the output of the first neural network model is calculated based on the initial estimated values of the weights between each layer of neurons. Here, at least some of the initial estimated values of the weights of the first neural network model describe the values of the parameters R, L d , L q and φ f of the motor to be identified under the condition of ignoring the non-linear characteristics of the motor. These initial estimated values of the weights may be derived by engineering experts based on experience, or may be any mathematically and physically meaningful values input randomly from the outside.

[0052] In step S14, the output error is calculated. For example, in step S12, the deviation between the d-axis voltage

Number

Number

[0053] In step S15, the connection weights are adjusted using the backpropagation algorithm based on the obtained error. Here, the error update value is propagated layer by layer so as to reach the first layer through backpropagation, and all weights are updated together at the end of the backpropagation process. The learning rate used in the backpropagation algorithm can be freely adjusted according to the required model accuracy and time consumption.

[0054] In step S16, it is checked whether the observation data used to train the model has been exhausted. For example, it is checked whether the number of traversed observation data is N or more.

[0055] If it has not been exhausted, the process returns from step S16 to step S12 and continues to extract the next training sample.

[0056] If it has been exhausted, in step S17, the total loss function is calculated for all training samples (i.e., all observation data), and it is determined whether the output result of the loss function satisfies the first predetermined condition. For example, it is checked whether the output result of the loss function is smaller than the first limit value.

[0057] If the output result of the loss function is greater than the first limit value, it indicates that the performance of the model still does not meet the standard and training needs to be continued. Therefore, the process returns from step S17 to step S12 and starts a new iteration loop.

[0058] When the output result of the loss function is smaller than the first limit value, it indicates that the model has already achieved satisfactory performance in the predicted result. In this case, in step S18, the current weights of the first neural network model are output, and based on this, the linear approximation part of the parameters determined by the synchronous motor is determined. Since the non-linear factors of the synchronous motor are not considered at this stage, it is worth noting that the parameters determined here are represented in a certain form. Regarding the first parameters that should be affected by non-linear factors (such as d-axis and q-axis inductances), the results obtained at this stage can be regarded as the average values of these parameters over all observation data, so they are not absolutely accurate. They need to be corrected through subsequent non-linear compensation processes. Regarding the second parameters (such as stator resistance and permanent magnet flux linkage) that should be constant, the results determined at this stage also include a part of the non-linear contribution from the first parameters.

[0059] Figure 3 shows the flowcharts of the two method steps in Figure 1. In this embodiment, method step S3 in Figure 1 is exemplified by steps S31 to S38.

[0060] In step S2, a compensation module in the form of a third neural network model is added to the trained first neural network model to form a second neural network model.

[0061] In step S31, a set of data consisting of d-axis current (i sd ), q-axis current (i sq ) and electrical angular velocity (ω1) is extracted from the acquired observation data and input into the second neural network model through forward propagation. For example, d-axis current (i sd ), q-axis current (i sq ) and electrical angular velocity (ω1) are input into the trained first neural network model, and d-axis current (i sd ) and q-axis current (i sq ) are also input into the compensation module.

[0062] In step S32, the output of the second neural network model is calculated. For example, the corresponding voltage

Number

[0063] In step S33, the error between the output voltage of the second neural network model and the voltage of the observation data is calculated.

[0064] In step S34, the weights of the trained first neural network model are adjusted based on this error using the backpropagation algorithm. The weights of the first neural network model are the motor parameters R, L d , L q and φ f still partially describe. However, L d , L q Regarding the intermediate results L d0 , L q0 Since they already represent the linear characteristic part, they are no longer involved in further training. However, the values of R and φ f incorrectly include a part of the non - linear contribution from L d , L q Therefore, they need to be compensated. Thus, when adjusting the weights of the trained first neural network model, only R and φ f are changed.

[0065] In step S35, the network weights of the third neural network model (compensation module) are adjusted. Here, for example, the overall output error of the second neural network model indirectly affects the output value of the third neural network model, and thereby is used to indirectly change the internal weights of the third neural network model.

[0066] In step S36, it is checked whether the observation data used to train the model has been exhausted.

[0067] If it has not been exhausted, the process returns from step S36 to step S31, and the extraction of the next training sample is continued.

[0068] If it has been exhausted, in step S37, the total loss function is calculated for all training samples (i.e., all observation data), and it is determined whether the output result of the loss function satisfies a second predetermined condition, that is, whether it is smaller than a second threshold value.

[0069] If the output result of the loss function is greater than the second threshold value, the process returns from step S37 to step S31, and a new iteration loop is started.

[0070] If the output result of the loss function is smaller than the second threshold value, it indicates that satisfactory performance has been achieved in the result predicted by the model. In this case, in step S38, the current weights of the second neural network model are output, and based on this, the final values of the parameters R and φ of the synchronous motor to be identified f are determined. Further, with the assistance of the output of the third neural network model, the non-linear characteristic portions ΔL d and L q of L d0 and ΔL q0 are obtained. By combining these with the linear characteristic portions L d0 and L q0 determined using the first neural network model, the functional relationship between L d and L q and the current can finally be determined.

[0071] FIG. 4 shows a schematic diagram of a device for parameter identification based on a neural network of a synchronous motor according to an exemplary embodiment of the present invention.

[0072] As shown in FIG. 4, the device 1 includes a first identification module 10, a correction module 20, and a second identification module 30. Both the first identification module 10 and the second identification module 30 are used to identify unknown parameters of the synchronous motor, but have different task focuses. For example, the first identification module 10 is mainly used to determine the linear characteristic part of the parameter to be identified or the parameter value less than the linear approximation, while the second identification module 20 is mainly used to determine the result considering the non-linear factor or the result after non-linear compensation.

[0073] Specifically, the first identification module 10 is used, for example, to establish and train a first neural network model and use the trained model to determine an intermediate result of the parameter to be identified of the synchronous motor.

[0074] The correction module 20 is used, for example, to form a second neural network model by adding a compensation module to the trained first neural network model.

[0075] The second identification module 20 is used, for example, to train a second neural network model based on the intermediate result of the parameter to be identified and use the trained second neural network model to determine the final result of the parameter to be identified.

[0076] FIG. 5 shows a schematic diagram of the operating points that need to be traversed for parameter identification using the lookup table method.

[0077] In order to comprehensively reflect the non-linear characteristics of specific parameters of the synchronous motor (for example, d-axis inductance and q-axis inductance), it is necessary to cover the entire operating range of the synchronous motor with operating points as comprehensively as possible. Therefore, it can be seen from FIG. 5 that it is necessary to collect and test a large number of operating points. Furthermore, since the magnetic saturation phenomenon is more important at low currents, it is particularly necessary to intensively increase the sampling speed at low currents, which significantly increases the time consumption of parameter identification. Furthermore, the accuracy of the sampled data itself is also limited by the accuracy of the measurement software / hardware.

[0078] FIG. 6 shows a conceptual schematic diagram of the non-linear characteristics of the d-axis flux of the synchronous motor.

[0079] In an ideal situation, the d-axis flux should change linearly with the corresponding current and form a plane. However, as shown in FIG. 6, the d-axis flux exhibits a non-linear relationship with the d-axis and q-axis currents due to saturation, which is visually represented as an irregular surface. This non-linearity further changes the d-axis inductance with the change of the current. Treating it as a constant will inevitably affect the control accuracy and stability of the motor.

[0080] FIG. 7 shows an exemplary schematic diagram of the first neural network model used in the method according to the present invention.

[0081] The first neural network model 60 has a two-layer neuron structure and is fully connected to each layer. The input-output relationship is established using the following static voltage state equations in the d-q coordinate system by the synchronous motor.

Equation

[0082] Where U sd is the d-axis voltage of the synchronous motor, R is the stator resistance of the synchronous motor, ω1 is the electrical angular velocity of the synchronous motor, L qis the q-axis inductance of the synchronous motor, and i sq is the q-axis current of the synchronous motor, and |φ f | is the amplitude of the permanent magnet flux linkage of the synchronous motor, θ is the zero angle of the synchronous motor, and U sq is the q-axis voltage, and i sd is the d-axis current of the synchronous motor, and L d is the d-axis inductance of the synchronous motor.

[0083] Using the first neural network model 60, to determine the linear characteristic part of the parameters to be identified of the synchronous motor, the q-axis current i sq and the d-axis current i sd and the electrical angular velocity ω1 of the synchronous motor are used as inputs to the first neural network model 60, while the q-axis voltage U sq and the d-axis voltage U sd are used as outputs of the first neural network model 60. Based on the mapping relationship defined by the static voltage state equation, the weights between the first input node and the two output nodes are R and ω1, the weights between the second input node and the two output nodes are ω1 and R, and the weights between the third input node and the two output nodes are |φ f |sinθ and |φ f |cosθ. When the zero angle of the motor is correctly calibrated, θ is 0. Here, R, L d , L q and φ f are adjustable weights of the first neural network model 60, and their values can be iteratively updated in the backpropagation process.

[0084] For each given input, the first neural network model 60 outputs estimated values of the q-axis voltage and the d-axis voltage. The loss function can be constructed based on the difference between the measured voltage and the estimated value.

Equation

[0085] where N is the number of known observation data used, and U sd is the d-axis voltage of the synchronous motor of the known observation data,

Number

Number

[0086] The weights of the network are at least partially modified using the backpropagation algorithm until the value of the loss function J satisfies the first predetermined condition. In this regard, the parameters R, L d , L q and φ f The intermediate results of can be determined based on the internal weights of the first neural network model 60.

[0087] FIG. 8 shows an exemplary schematic diagram of a compensation module used in the method according to the present invention.

[0088] The compensation module 70 in the form of a third neural network model includes, illustratively, an input layer, an output layer, and at least one hidden layer. Depending on the desired accuracy and the amount of observation data, the number of hidden layers and the number of neurons contained therein can be adjusted. The q-axis inductance L q and the d-axis inductance L d To fit the non-linear characteristics of, L q and L d Are each assumed to be composed of two parts represented as follows.

Number

[0089] where Lq and L d are the q - axis inductance and the d - axis inductance respectively, and L d0 and L q0 represent the linear characteristic parts of the q - axis inductance and the d - axis inductance, and ΔL d0 and ΔL q0 represent the non - linear characteristic parts of the q - axis inductance and the d - axis inductance. For example, ΔL d0 and ΔL q0 are determined in the form of intermediate results using the first neural network model. Therefore, the objective is to use the non - linear compensation module 70 to determine ΔL d0 and ΔL q0 The magnetic circuit saturation of the synchronous motor is very related to the d - axis and q - axis currents i sd and i sq Therefore, since i sd and i sq are very related to ΔL d0 and ΔL q0 a direct relationship between them is established within the compensation module. In other words, the q - axis current i sq and the d - axis current i sd are used as inputs to the third neural network model, and the non - linear characteristic parts ΔL d0 and ΔL q0 of the parameters to be identified are used as outputs of the third neural network model.

[0090] FIG. 9 shows an exemplary schematic diagram of the second neural network model used in the method according to the present invention.

[0091] Here, the compensation module 70 shown in FIG. 8 is such that the compensation module 70 shares a part of the input layer with the first neural network model 60, that is, "the d, q - axis currents i sd and i sqIt is added to the first neural network model 60 shown in FIG. 7 so as to share "". The output layer of the compensation module 70 is completely connected to the output layer of the first neural network model 60, thus forming the overall second neural network model 80.

[0092] Overall, the input of the second neural network model 80 still remains the q-axis current i sq , the d-axis current i sd , and the electrical angular velocity ω1 of the synchronous motor. The output of the second neural network model 80 still remains the q-axis voltage U sq and the d-axis voltage U sd . However, due to the consideration of non-linear factors, the input-output relationship of the second neural network model 80 is established based on the following static voltage state equation.

Equation

[0093] In the formula, U sd is the d-axis voltage of the synchronous motor, R is the stator resistance of the synchronous motor, ω1 is the electrical angular velocity of the synchronous motor, L q0 is the linear characteristic part of the q-axis inductance of the synchronous motor, ΔL q0 is the non-linear characteristic part of the q-axis inductance of the synchronous motor, i sq is the q-axis current of the synchronous motor, |φ f | is the amplitude of the permanent magnet flux linkage of the synchronous motor, θ is the zero angle of the synchronous motor, U sq is the q-axis voltage, i sd is the d-axis current, L d0 is the linear characteristic part of the d-axis inductance, and ΔL d0 is the non-linear characteristic part of the d-axis inductance.

[0094] During the training of the second neural network model 80, with respect to the first neural network model 60, the adjustment starts from the weight values obtained at the end of the training of the first neural network model 60. To facilitate compensation through the non-linear characteristic portion, the linear characteristic portions L d0 and L q0 of the d-axis and q-axis inductances are fixed. As a result, the weights represented by these two characterizations are assumed not to participate in the second iterative training. Therefore, in the adjustment process, only the values of R and φ f are changed.

[0095] Regarding the compensation module 70 (i.e., the third neural network model), a predetermined value of the internal weights of this network is changed to change the output of ΔL d0 and ΔL q0 .

[0096] This iterative process of updating the weights continues until the following loss function converges.

Equation

[0097] where N is the number of known observation data used, U sd is the d-axis voltage of the synchronous motor in the known observation data,

Equation

Equation

[0098] When the training of the second neural network model 80 is completed, the parameters R and φ to be identifiedf The final result can be determined based on the internal weights of the corresponding first neural network model 60. Further, upon completion of training, with the assistance of the compensation module 70, the linear characteristic parts L d0 and L q0 combined with the adapted L d and L q of the non-linear characteristic parts ΔL d0 and ΔL q0 are used to determine the functional relationships between L d and L q and the d-axis and q-axis currents.

[0099] Figures 10a and 10b show schematic diagrams of the non-linear characteristic parts of the d-axis inductance and q-axis inductance of a synchronous motor determined by the method according to the present invention that varies with current.

[0100] From Figures 10a and 10b, it can be observed that with the assistance of the compensation module, the non-linear characteristic parts ΔL d0 and ΔL q0 of the d-axis and q-axis inductances can be adapted to the variations in the d-axis and q-axis currents i sd and i sq to form a surface plot. Based on the neural network technology used and based on limited observed data points, intelligent inferences and judgments can be applied to unknown parts of the surface, enabling automatic interpolation and reconstruction of missing data to restore the complete functional relationship between ΔL d0 , ΔL q0 and the current. From the fitting results, it can be seen that the non-linear characteristic parts of the d-axis and q-axis inductances are not constant but vary due to the influence of flux non-linearity caused by phenomena such as magnetic saturation of the synchronous motor. The non-linear characteristics are more prominent especially when the current values i sd and i sq are small.

[0101] Certain embodiments of the present invention have been described in detail herein, but they are provided for illustrative purposes only and should not be construed as limiting the scope of the present invention. Various substitutions, changes, and modifications can be devised without departing from the spirit and scope of the present invention.

Claims

1. A method for identifying parameters of a synchronous motor based on a neural network, comprising: a computer S1) establishing and training a first neural network model (60), and using the trained first neural network model (60) to determine an intermediate result of a parameter to be identified of the synchronous motor; S2) forming a second neural network model (80) by adding a compensation module (70) to the trained first neural network model (60); S3) training the second neural network model (80) based on the intermediate result of the parameter to be identified, and using the trained second neural network model (80) to determine a final result of the parameter to be identified. A method comprising the above steps.

2. The intermediate result includes a linear approximation value of a first parameter among the parameters to be identified, and a first estimated value of a second parameter among the parameters to be identified that is less than the linear approximation value of the first parameter. The method according to claim 1.

3. The final result includes the linear approximation value of the first parameter among the parameters to be identified after compensating for non-linearity, and a second estimated value of the second parameter among the parameters to be identified after compensating for the non-linearity of the linear approximation value of the first parameter. The method according to claim 2.

4. The parameters to be identified of the synchronous motor include the stator resistance, d-axis inductance, q-axis inductance, and permanent magnet flux of the synchronous motor. The method according to claim 1.

5. Step S1 includes establishing an input-output relationship of the first neural network model (60) based on the static voltage state equation of the synchronous motor in the d-q coordinate system, wherein the input of the first neural network model (60) includes the q-axis current, d-axis current, and electrical angular velocity of the synchronous motor, the output of the first neural network model (60) includes the q-axis voltage and d-axis voltage of the synchronous motor, and the weights of the first neural network model (60) are at least partially represented by the parameters to be identified. The method according to claim 2.

6. Step S1 is to obtain known observation data of the synchronous motor, where the known observation data includes the d-axis current, q-axis current, d-axis voltage, q-axis voltage, and electrical angular velocity of the synchronous motor measured at different operating points, to train the first neural network model (60) using the known observation data, and adjust the weights of the first neural network model (60) until the loss function satisfies a first predetermined condition, when the loss function satisfies the first predetermined condition, based on the weights of the first neural network model (60), determine an intermediate result of the parameter to be identified of the synchronous motor, The method according to claim 5, comprising the above steps.

7. Adjusting the weights of the first neural network model (60) includes repeatedly updating the weights of the first neural network model (60) using a gradient descent algorithm, particularly the backpropagation algorithm. The method according to claim 6.

8. Step S2 is using the q-axis current and d-axis current as inputs to a third neural network model, and using the non-linear characteristic part of the first parameter among the parameters to be identified as the output of the third neural network model, to establish the third neural network model as the compensation module (70). The method according to claim 2.

9. Step S2 is including adding the compensation module (70) to the first neural network model (60) by sharing a part of the input layer of the first neural network model (60) with the input layer of the compensation module (70), and connecting the output layer of the compensation module (70) to the output layer of the first neural network model (60). The method according to claim 1.

10. Step S3 is to train the second neural network model (80) using the known observation data, and adjust the weights of the trained first neural network model (60) and the weights of the compensation module (70) until the loss function satisfies a second predetermined condition. When the loss function satisfies the second predetermined condition, based on the weights of the trained second neural network model (80) and the weights of the compensation module (70), determining the final result of the parameter to be identified of the synchronous motor; The method according to claim 6, comprising:

11. While adjusting the weights of the trained first neural network model (60), the intermediate result of the first parameter among the parameters to be identified is retained without change, and the intermediate result of the second parameter among the parameters to be identified is changed. The method according to claim 10.

12. The loss function is represented by the following formula, 【Number 1】 where N is the number of known observation data used, and U ｓｄ is the d-axis voltage of the synchronous motor in the known observation data, 【Number 2】 is the d-axis voltage output by the first neural network model (60) or the second neural network model (80), U ｓｑ is the q-axis voltage of the synchronous motor in the known observation data, [Number 3] is the q-axis voltage output by the first neural network model (60) or the second neural network model (80). The method according to claim 6 or 11.

13. The method includes Transmitting the final result of the parameter to be identified to the controller of the synchronous motor to control the operation of the synchronous motor based on the final result; And / or When applying the weights of the second neural network model (80), enabling the internal neural network of the controller of the synchronous motor to determine the online final result of the parameter to be identified in real time based on the operating parameters of the synchronous motor, and transmitting the weights of the trained second neural network model (80), which controls the operation of the synchronous motor based on the online final result, to the internal neural network. The method according to claim 10, further comprising:

14. A synchronous motor parameter identification device (1) based on a neural network, wherein the device (1) is configured to perform the method according to claim 1, and the device (1) A first identification module (10) configured to establish and train a first neural network model (60) and use the trained first neural network model (60) to determine an intermediate result of a parameter to be identified of the synchronous motor; A modification module (20) configured to form a second neural network model (80) by adding a compensation module (70) to the trained first neural network model (60); A second identification module (30) configured to train the second neural network model (80) based on the intermediate result of the parameter to be identified and to determine a final result of the parameter to be identified using the trained second neural network model (80); A synchronous motor parameter identification device (1) comprising the above.

15. A computer program configured to cause a computer to execute the method according to claim 1 when the computer program is executed by the computer.

Images