A design method for reversible state-space observer based on flow model
Through the reversible state space observer method based on the flow model, a reversible flow model of the state subspace is constructed and a full-connected neural network is used for prediction, which solves the problem that nonlinear systems in the prior art is difficult to achieve reversible state space, and improves the identification Accuracy and reliability.
Patent Information
- Application Number
- CN202411392453.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-08
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-10-08
AI Technical Summary
The existing nonlinear system identification methods are difficult to realize the reversible state space, resulting in high computational complexity, low accuracy and unstable calculations, and cannot meet the requirements of low computational complexity, high accuracy, robust and robust state estimation.
Design a reversible state space observer method based on the flow model. By collecting the current input and position angle output of the permanent magnet synchronous motor, constructing a coupling layer and transforming the coupling sequence, constructing a reversible flow model of state subspace, and using a fully connected neural network Perform predicted state data processing, and realize reversible state space observation through the state transformation matrix and its inverse matrix.
It improves the reliability and accuracy of nonlinear system identification, alleviates state observation errors, and realizes state estimation with low computational complexity and high precision.
Smart Images

Figure CN118897473B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of nonlinear system identification, and in particular to a reversible state space observer design method based on a flow model. Background Art
[0002] At present, in order to achieve the feature fitting of complex nonlinear systems, deep learning methods are widely used in the research of system identification. It trains complex neural network models through a large amount of data, automatically learns the input-output relationship of the system, and widely used methods such as convolutional neural networks or recursive neural networks can effectively capture the dynamic characteristics and timing relationships of the system to achieve the construction of data-driven models. At the same time, in order to improve the accuracy of nonlinear system identification, deep feedforward neural networks, in particular, have strong nonlinear fitting capabilities and can handle complex nonlinear relationships. By training deep neural networks, the nonlinear dynamic behavior of the system can be identified and a more accurate model can be obtained.
[0003] However, deep networks are inherently opaque and may learn unstable solution spaces, which may lead to large observation errors in some untrained dynamic modes. Therefore, most nonlinear system identification methods cannot achieve reversible state spaces and cannot meet the requirements of low computational complexity, high precision, and robust state estimation. Summary of the invention
[0004] Based on this, it is necessary to provide a reversible state space observer design method based on flow model, which includes:
[0005] S1: collecting the current input and the rotor position angle output of the permanent magnet synchronous motor vector control system; taking the current input as input data and the position angle output as response data;
[0006] S2: Define the optimization goal, construct the coupling layer, change the coupling order, and build a reversible flow model in the state subspace;
[0007] S3: inputting the response data into the flow model to obtain intermediate latent variables; multiplying the intermediate latent variables with the state transformation matrix to obtain state data;
[0008] S4: combining the state data with the input data and inputting them into a fully connected neural network to obtain predicted state data of the system;
[0009] S5: multiplying the predicted state data by the inverse matrix of the state transformation matrix to obtain latent variables; the flow model maps the latent variables to a latent space to obtain a preliminary identification result;
[0010] S6: Based on the preliminary identification result and the hidden variable, the output equation of the permanent magnet synchronous motor vector control system is changed to obtain a reversible state space observer; the reversible state space observer outputs the identification result according to the input data to be measured and the response data.
[0011] Beneficial effects: This method is applied to control system design to identify input and output data in control system design and predict the response of the system. The constructed reversible state space observer improves the reliability and accuracy of nonlinear system identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0013] Figure 1 This is a flow chart of a reversible state space observer design method based on a flow model according to an embodiment of the present application.
[0014] Figure 2 This is a structural diagram of a permanent magnet synchronous motor vector control system according to an embodiment of the present application.
[0015] Figure 3 This is a graph showing loss experiment results for an embodiment of the present application.
[0016] Figure 4 The figure is a comparison chart of the actual and predicted experimental results of the embodiments of the present application.
[0017] Figure 5 4 is a structural diagram of a reversible state space observer according to an embodiment of the present application. DETAILED DESCRIPTION
[0018] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.
[0019] like Figure 1 As shown, this embodiment provides a reversible state space observer design method based on a flow model, the method comprising:
[0020] S1: Collecting the current input and the rotor position angle output of the permanent magnet synchronous motor vector control system; using the current input as input data and the position angle output as response data.
[0021] Furthermore, the mathematical description of the permanent magnet synchronous motor vector control system is expressed as:
[0022] ;
[0023] in, x k Represents the state data of the permanent magnet synchronous motor vector control system at the kth moment, ; Dimensions representing state data; u k represents the input data of the permanent magnet synchronous motor vector control system at the kth moment, ; Indicates the dimension of the input data; x k+1 Represents the state data of the permanent magnet synchronous motor vector control system at the k+1th moment; y k represents the output data of the permanent magnet synchronous motor vector control system at the kth moment, ; Indicates the dimension of the output data; f (·) represents the state transition mapping, ; h (·) represents the mapping relationship between state data and output data in the permanent magnet synchronous motor vector control system. .
[0024] This embodiment uses a permanent magnet synchronous motor (PMSM) for simulation. Through the system identification method, the output information of the system is collected, and the relationship between input and output is fitted using data processing technology to obtain an equivalent model between the two. This method is widely used in the field of servo control; the structure of the permanent magnet synchronous motor vector control system is as follows Figure 2 As shown. The system mainly includes the following modules: PMSM module, current loop module, coordinate transformation module, space vector pulse width modulation (SVPWM) module, and measurement module. Given a reference current , , after coordinate transformation , The three-phase stator current is converted into a three-phase stationary coordinate system, and then the space vector pulse width modulation module outputs six pulse width modulation waves to drive the inverter circuit, generating a three-phase sinusoidal voltage with variable amplitude and frequency to control the motor operation. Vector control uses a dq rotating coordinate system and a direct axis current Therefore, the permanent magnet synchronous motor has only the quadrature axis current component during analysis, and the motor torque has only the permanent magnet torque component. The size of the rotor can control the position angle The main parameters used are shown in Table 1.
[0025] Table 1 is the main parameter explanation table of permanent magnet synchronous motor vector control system;
[0026] ;
[0027] based on The control structure system takes into account the position angle of the rotor of the PMSM system It is regulated by current, therefore, the present invention expresses the input variable u and the response variable y as:
[0028] ;
[0029] ;
[0030] The superscript i represents time, represents the q-axis current under Park variation, is the rotor position angle.
[0031] In the process of collecting samples, the PMSM system is connected to the oscilloscope and the position sensor, and the sampling interval is set to continuously collect sample data during normal operation. In this way, the present invention collects 20,000 and 10,000 samples for training and testing respectively. According to the model of this article, the hyperparameters are selected as follows: the number of iterations Nepoch = 100; the batch size Nb = 64; the dropout rate Dropout = 0.2; the learning rate η = 0.0004; the training test ratio is 2:1.
[0032] In order to evaluate and verify the performance of the method provided in this embodiment, the specific data of the actual PMSM system is used to test the state subspace observer based on the flow model proposed in the embodiment of the present invention. The experimental results are as follows: Figure 3 , Figure 4 The motor angle estimated by the state subspace reversible observer based on the flow model proposed in the present invention is shown as follows. Motor angle with real PMSM system The loss function values of the two data are plotted. It can be seen that after training iterations, the loss function trend continues to decline, and after the fourth iteration, both training and testing tend to converge, and the test loss is very small and close to 0. At the same time, Figure 4The comparison of the estimated and true values shows that the model has good fitting ability. Then it is proved that the reversible state space observer design method based on flow model proposed in this invention shows the position angle of the rotor of PMSM system under time series under the mutual assistance of neural network model and flow model. Good identification ability greatly alleviates the state observation error of the nonlinear system state observer.
[0033] S2: Define the optimization objective, construct the coupling layer, and change the coupling order to build a reversible flow model in the state subspace.
[0034] Specifically, in order to initially build a generative model, this project first builds and defines the reversible characteristics of the flow model to achieve the mapping of the state subspace. .first, As a learnable network model, it constructs coupling layers by block reorganization. 1 , I 2 is the part of [1,D], I 1 =[1,d],I 2 =[d,D], m is an arbitrary function. By definition have:
[0035] ;
[0036] ;
[0037] The reversible transformation is:
[0038] ;
[0039] ;
[0040] It can be seen that function m does not affect the reversible transformation. Therefore, m can be a neural network of any complexity. This project sets it as a fully connected neural network. The Jacobian matrix in this transformation process is:
[0041] ;
[0042] if Additive coupling layer , then we can get
[0043] ;
[0044] ;
[0045] at this time, , the determinant of the entire Jacobian matrix is 1, and its logarithm is 0, so its calculation does not need to be considered.
[0046] It can be seen that the first transformation of the designed coupling layer is an identity transformation, and the identity transformation property will be maintained no matter how many layers it passes through. Therefore, in order to increase the nonlinear ability, it is considered to introduce a combination of multiple simple transformations to enhance the fitting ability, that is:
[0047] ;
[0048] ;
[0049] in, f represents the probability distribution transformation function; f 1 represents the probability distribution transformation function of the first layer; f 2 represents the probability distribution transformation function of the second layer; f K represents the probability distribution transformation function of the Kth layer; h 1 Represents a simple distribution sample after the first layer transformation; h 2 Represents a simple distribution sample after the second layer transformation; Indicates a mapping relationship; Represents the composition of functions;
[0050] The combination of each of the above transformations is additively coupled, forming a flow, which is obtained by the chain rule:
[0051] ;
[0052] For each additive coupling transformation sequence, before each additive coupling, the order of each dimension is disrupted or flipped, and each feature dimension finally encoded is scaled to obtain the flow model. Based on the optimization target, the flow model is trained using a gradient descent optimization algorithm. During training, the negative log-likelihood of the data is minimized to train the flow model through a gradient descent optimization algorithm, so that the flow model can accurately capture the probability distribution of the generated data and learn an accurate mapping relationship to achieve a reversible effect of the flow model.
[0053] The expression of the optimization objective is:
[0054] ;
[0055] in, px ( x ) represents the logarithmic probability density; px (z) means z Logarithmic probability density under standard normal distribution; D represents the sample size; i represents the current sample size; det(·) represents the determinant;J represents the Jacobian matrix; represents the inverse function of the transformation function; z represents the original simple distribution sample; x Represents the original sample.
[0056] After constructing the optimization target, the model is trained by minimizing the negative logarithmic probability density of the data, thereby completing the construction of the flow model.
[0057] Then, according to the built flow generation model, the hidden variables are generated under the state space transformation. Given two sets of data y and z, where y obeys a known simple prior distribution π(y) (usually a Gaussian distribution) and x obeys a complex distribution p(z) (i.e. the distribution represented by the training data), now find a transformation function , which can establish a mapping from y to z , so that for each sample point y' in π(y), there is a (new) sample point z' corresponding to it in p(z). In this process, the function In this process, the generator G to be solved is defined by optimizing the objective formula, from which the mapping relationship from z to x can be obtained, which is in the following form:
[0058] ;
[0059] In order to facilitate the subsequent system identification work, it is advisable to make the following assumptions:
[0060] A1: To ensure must exist and be calculable, which means The input and output dimensions must be consistent and the determinant of G cannot be 0, that is, Must be of the same dimension.
[0061] A2: W 1 The inverse matrix W 2 exists, and W 1 W 2 =I.
[0062] Input data for learning With intermediate hidden variables The nonlinear mapping relationship between them is obtained by utilizing the reversible characteristics of the state subspace of the constructed flow model. , is the subsequent state subspace The forward mapping provides model support. Then, based on assumption A2, a reversible state transformation matrix is defined. W 1, and the intermediate latent variable Multiply them together to get the state variable .
[0063] S3: Input the response data into the flow model to obtain intermediate latent variables; multiply the intermediate latent variables by the state transformation matrix to obtain state data.
[0064] Specifically, the response data is input into the flow model to obtain the intermediate latent variables, and the calculation formula is:
[0065] ;
[0066] in, z k represents the intermediate latent variable; Representation flow model; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment;
[0067] The intermediate latent variable is multiplied by the state transformation matrix to obtain the state data. The calculation formula is:
[0068] ;
[0069] in, x k Represents the state data of the permanent magnet synchronous motor vector control system at the kth moment; W 1 Represents the state transformation matrix.
[0070] S4: Combine the state data with the input data and input them into a fully connected neural network to obtain predicted state data of the system.
[0071] Furthermore, this step also includes: transforming the mathematical description of the permanent magnet synchronous motor vector control system according to the predicted state data, and the transformation formula is:
[0072] ;
[0073] in, Represents the predicted state data at the k+1th moment; W 1 represents the state transformation matrix; represents the hidden variable at the k+1th moment; f (·) represents a fully connected neural network; z k represents the intermediate latent variable at the kth moment; u k Represents the input data of the permanent magnet synchronous motor vector control system at the kth moment; yk Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment; h (·) represents the mapping relationship between state data and output data in the permanent magnet synchronous motor vector control system; x k Represents the state data of the permanent magnet synchronous motor vector control system at the kth moment.
[0074] S5: multiplying the predicted state data by the inverse matrix of the state transformation matrix to obtain latent variables; the flow model maps the latent variables to the latent space to obtain a preliminary identification result.
[0075] Specifically, the predicted state data is multiplied by the inverse matrix of the state transformation matrix to obtain the latent variable, and the calculation formula is:
[0076] ;
[0077] in, Represents the predicted state data at the k+1th moment; W 2 represents the inverse matrix of the state transformation matrix; represents the hidden variable at the k+1th moment;
[0078] The flow model maps the latent variables to the latent space to obtain a preliminary identification result, and the calculation formula is:
[0079] ;
[0080] in, represents the preliminary identification result at the k+1th moment; Represents a flow model.
[0081] S6: Based on the preliminary identification result and the hidden variable, the output equation of the permanent magnet synchronous motor vector control system is changed to obtain a reversible state space observer; the reversible state space observer outputs the identification result according to the input data to be measured and the response data.
[0082] In this embodiment, the identification result is the next moment response of the permanent magnet synchronous motor vector control system, that is, the position angle of the rotor at the next moment.
[0083] Specifically, the mathematical description of the transformed permanent magnet synchronous motor vector control system is transformed again according to the hidden variables and the flow model, and the mathematical description of the transformed permanent magnet synchronous motor vector control system is expressed as:
[0084] ;
[0085] in, represents the hidden variable at the k+1th moment; W 2 represents the inverse matrix of the state transformation matrix; f (·) represents a fully connected neural network; W 1 represents the state transformation matrix; z k represents the intermediate latent variable at the kth moment; u k Represents the input data of the permanent magnet synchronous motor vector control system at the kth moment; Representation flow model; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment.
[0086] Furthermore, based on the mathematical description of the transformed permanent magnet synchronous motor vector control system, the preliminary identification results and the hidden variables, the output equation of the permanent magnet synchronous motor vector control system is changed to obtain a reversible state space observer. The structure of the reversible state space observer is as follows: Figure 5 As shown; the expression of the reversible state space observer is:
[0087] ;
[0088] in, represents the identification result at the k+1th moment; Representation flow model; represents the hidden variable at the k+1th moment; W 2 represents the inverse matrix of the state transformation matrix; f (·) represents a fully connected neural network; W 1 represents the state transformation matrix; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment; u k Represents the input data of the permanent magnet synchronous motor vector control system at the kth moment.
[0089] In this embodiment, a first loss function is constructed by a state transformation matrix and its inverse matrix, which is used to adjust the state transformation matrix and its inverse matrix; a second loss function is constructed by an identification result and its corresponding true value, which is used to adjust the model parameters of a reversible state space observer; a third loss function is constructed by the optimization objective, which is used to adjust and optimize the parameters of the flow model;
[0090] A total loss function is constructed according to the first loss function, the second loss function and the third loss function, and the overall framework is back-propagated based on the total loss function until convergence.
[0091] Specifically, the expression of the total loss function is:
[0092] ;
[0093] ;
[0094] ;
[0095] ;
[0096] in, L total represents the total loss function; L 1 represents the first loss function; L 2 represents the second loss function; L 3 represents the third loss function; W 1 represents the state transformation matrix; W 2 represents the inverse matrix of the state transformation matrix; I represents the identity matrix; represents the least squares norm form; represents the identification result at the kth moment; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment.
[0097] The purpose of the reversible state space observer design method based on the flow model provided in this embodiment is to realize the reversible transformation of the observed state subspace through the inverse mapping and state transformation matrix represented by the flow model, and to design a hybrid identification system of data and mechanism of the nonlinear system, and to use the input and output data of the nonlinear system to train the deep FCNN model and the flow model, and continuously approximate the nonlinear behavior of the nonlinear system. System identification based on the reversible state space observer is mainly used in control system design, to identify the input and output data in the control system design, and to predict the response of the system. This invention aims to improve the reliability and accuracy of nonlinear system identification.
[0098] Compared with the prior art, the reversible state space observer design method based on the flow model provided in this embodiment has the following beneficial effects:
[0099] 1. The state subspace reversible observer design method based on the flow model described in this embodiment is a data-driven intelligent method that can be used for general nonlinear system identification. It provides a new solution to the problem that some existing system identification methods perform poorly because they cannot achieve reversible state space, so as to meet the requirements of low computational complexity, high precision, and robust state estimation.
[0100] 2. The state subspace observer design method based on the flow model described in this embodiment can better realize the observation of the reversible transformation of the state subspace through the inverse mapping and state transformation matrix represented by the flow model.
[0101] 3. The state subspace observer design method based on the flow model described in this embodiment designs an adaptive loss learning strategy with a reversible state transformation matrix. W 1 The matrix parameters of are fixed, and the reconstruction loss is calculated. W 2 The matrix is updated to pull the reconstructed latent variables back to the normal domain, and finally the true value of the output is estimated.
[0102] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0103] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.
Claims
1. A reversible state space observer design method based on flow model, characterized in that: include: S1: Collect the current input of the permanent magnet synchronous motor vector control system and the rotor position angle output; using the current input as input data and the position angle output as response data; S2: Define the optimization goal, construct the coupling layer, change the coupling order, and build a reversible flow model in the state subspace; The expression of the optimization objective is: ; in, px ( x ) represents the logarithmic probability density; px (z) means z Logarithmic probability density under standard normal distribution; D represents the sample size; i represents the current sample size; det(·) represents the determinant; J represents the Jacobian matrix; represents the inverse function of the transformation function; z represents the original simple distribution sample; x represents the original sample; The fully connected neural network is reorganized into blocks to construct a coupling layer; Introducing the combination of multiple transformations in the fully connected neural network, the calculation formula is: ; ; in, f represents the probability distribution transformation function; f 1 represents the probability distribution transformation function of the first layer; f 2 represents the probability distribution transformation function of the second layer; f K represents the probability distribution transformation function of the Kth layer; h 1 represents a simple distribution sample after the first layer transformation; h 2 represents the simple distribution sample after the second layer transformation; Indicates a mapping relationship; Represents the composition of functions; The combination of each of the transformations is additively coupled, forming a flow; For each additive coupling transformation sequence, before each additive coupling, the order of each dimension is disrupted or flipped, and each feature dimension encoded at the end is scaled to obtain a flow model; S3: inputting the response data into the flow model to obtain intermediate latent variables; multiplying the intermediate latent variables with the state transformation matrix to obtain state data; S4: combining the state data with the input data and inputting them into a fully connected neural network to obtain predicted state data of the system; S5: multiplying the predicted state data by the inverse matrix of the state transformation matrix to obtain latent variables; the flow model maps the latent variables to a latent space to obtain a preliminary identification result; S6: Based on the preliminary identification result and the hidden variable, the output equation of the permanent magnet synchronous motor vector control system is changed to obtain a reversible state space observer; the reversible state space observer outputs the identification result according to the input data to be measured and the response data.
2. The reversible state space observer design method based on flow model according to claim 1 is characterized in that: The mathematical description of the permanent magnet synchronous motor vector control system is expressed as: ; in, x k Represents the state data of the permanent magnet synchronous motor vector control system at the kth moment, ; Dimensions representing state data; u k represents the input data of the permanent magnet synchronous motor vector control system at the kth moment, ; Indicates the dimension of the input data; x k+1 Represents the state data of the permanent magnet synchronous motor vector control system at the k+1th moment; y k represents the output data of the permanent magnet synchronous motor vector control system at the kth moment, ; Indicates the dimension of the output data; f (·) represents the state transition mapping, ; h (·) represents the mapping relationship between state data and output data in the permanent magnet synchronous motor vector control system. .
3. The reversible state space observer design method based on flow model according to claim 2 is characterized in that: Based on the optimization objective, the flow model is trained using a gradient descent optimization algorithm.
4. The reversible state space observer design method based on flow model according to claim 2 is characterized in that: The response data is input into the flow model to obtain the intermediate latent variables, and the calculation formula is: ; in, z k represents the intermediate latent variable; Representation flow model; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment; The intermediate latent variable is multiplied by the state transformation matrix to obtain the state data. The calculation formula is: ; in, x k Represents the state data of the permanent magnet synchronous motor vector control system at the kth moment; W 1 represents the state transformation matrix.
5. The reversible state space observer design method based on flow model according to claim 4 is characterized in that: S4 also includes: transforming the mathematical description of the permanent magnet synchronous motor vector control system according to the predicted state data, and the transformation formula is: ; in, Represents the predicted state data at the k+1th moment; W 1 represents the state transformation matrix; represents the hidden variable at the k+1th moment; f (·) represents a fully connected neural network; z k represents the intermediate latent variable at the kth moment; u k Represents the input data of the permanent magnet synchronous motor vector control system at the kth moment; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment; h (·) represents the mapping relationship between state data and output data in the permanent magnet synchronous motor vector control system; x k Represents the state data of the permanent magnet synchronous motor vector control system at the kth moment.
6. The method for designing a reversible state space observer based on a flow model according to claim 5, characterized in that: The predicted state data is multiplied by the inverse matrix of the state transformation matrix to obtain the latent variable, and the calculation formula is: ; in, Represents the predicted state data at the k+1th moment; W 2 represents the inverse matrix of the state transformation matrix; represents the hidden variable at the k+1th moment; The flow model maps the latent variables to the latent space to obtain a preliminary identification result, and the calculation formula is: ; in, represents the preliminary identification result at the k+1th moment; Represents a flow model.
7. The method for designing a reversible state space observer based on a flow model according to claim 6, characterized in that: The mathematical description of the transformed permanent magnet synchronous motor vector control system is transformed again according to the hidden variables and the flow model. The mathematical description of the transformed permanent magnet synchronous motor vector control system is expressed as: ; in, represents the hidden variable at the k+1th moment; W 2 represents the inverse matrix of the state transformation matrix; f (·) represents a fully connected neural network; W 1 represents the state transformation matrix; z k represents the intermediate latent variable at the kth moment; u k Represents the input data of the permanent magnet synchronous motor vector control system at the kth moment; Representation flow model; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment.
8. The method for designing a reversible state space observer based on a flow model according to claim 7, characterized in that: Based on the transformed mathematical description of the permanent magnet synchronous motor vector control system, the preliminary identification results and the hidden variables, the output equation of the permanent magnet synchronous motor vector control system is changed to obtain a reversible state space observer; the expression of the reversible state space observer is: ; in, represents the identification result at the k+1th moment; Representation flow model; represents the hidden variable at the k+1th moment; W 2 represents the inverse matrix of the state transformation matrix; f (·) represents a fully connected neural network; W 1 represents the state transformation matrix; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment; u k Represents the input data of the permanent magnet synchronous motor vector control system at the kth moment.
9. The reversible state space observer design method based on flow model according to claim 3 is characterized in that: A first loss function is constructed by using a state transformation matrix and its inverse matrix to adjust the state transformation matrix and its inverse matrix; a second loss function is constructed by using the identification result and its corresponding true value to adjust the model parameters of the reversible state space observer; a third loss function is constructed by using the optimization target to adjust and optimize the parameters of the flow model; A total loss function is constructed according to the first loss function, the second loss function and the third loss function, and the overall framework is back-propagated based on the total loss function until convergence.
10. The method for designing a reversible state space observer based on a flow model according to claim 9, characterized in that: The expression of the total loss function is: ; ; ; ; in, L total represents the total loss function; L 1 represents the first loss function; L 2 represents the second loss function; L 3 represents the third loss function; W 1 represents the state transformation matrix; W 2 represents the inverse matrix of the state transformation matrix; I represents the identity matrix; represents the least squares norm form; represents the identification result at the kth moment; y k Represents the output data of the permanent magnet synchronous motor vector control system at the kth moment.
Citation Information
Patent Citations
Permanent magnet synchronous motor parameter identification method based on model predictive control
CN112054731A
Parameter identification method for permanent magnet synchronous motor
CN114204864A
Permanent magnet synchronous motor parameter identification method, device and system
CN114244216A
Method for parameter identification of a synchronous motor based on a neural network
DE112021007423T5
