Permanent magnet synchronous motor parameter identification method based on NARX neural network

Abstract

The invention provides a permanent magnet synchronous motor parameter identification method based on an NARX neural network. According to the method, the parameter identification method is improved from two aspects. Firstly, a neural network is adopted to replace a traditional identification algorithm based on the nonlinear characteristic of the permanent magnet synchronous motor, and the neural network can well reflect a nonlinear system, so that the parameter identification precision is better than that of the traditional identification algorithm, and compared with the traditional identification algorithm such as MRAS, the method has higher universality. And secondly, first-order time delay is added on the basis of the neural network, so that the sequential relationship of online identification can be better reflected, the accuracy of the model is higher, the identification effect is better, and the robustness is higher.

Description

Technical Field

[0001] The invention relates to the field of parameter identification of surface-mounted permanent magnet synchronous motors, and in particular to a permanent magnet synchronous motor identification method based on a NARX neural network. Background Art

[0002] Permanent magnet synchronous motor is a high-efficiency, high-performance motor that is widely used in electric vehicles, industrial automation, and renewable energy. Accurate motor parameters are critical to control performance and system stability. However, due to factors such as manufacturing differences and environmental changes, motor parameters may change, and parameter identification is required to obtain accurate parameter information.

[0003] Traditional methods include mathematical models and experimental tests, but they have limitations. Therefore, it is necessary to explore new permanent magnet synchronous motor parameter identification methods to improve the accuracy and efficiency of parameter identification. Summary of the invention

[0004] In order to solve the above-mentioned deficiencies, the present invention provides a permanent magnet synchronous motor parameter identification method based on NARX neural network to improve the accuracy and efficiency of identification.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a permanent magnet synchronous motor parameter identification method based on NARX neural network, comprising the following steps:

[0006] According to the discrete current equation, a NARX neural network model of first-order parameters to be identified is constructed;

[0007] The historical state data of the motor is obtained and input into the NARX neural network model to obtain the predicted values ​​of the parameters to be identified through training.

[0008] The NARX neural network model is obtained by the following steps:

[0009] The initial NARX network expression is:

[0010] y(n+1)=f[y(n),...,y(nd y +1),u(nk),u(nk-1),...,u(nkd u +1)]

[0011] Among them, u(n)∈R and y(n)∈R represent the input and output of the NARX neural network model when the discrete time step is n, R is a real number, d u ≥1,d y ≥1 and d u ≤d yThey represent the order of input memory and output memory respectively. The parameter k is the delay term, which is used to represent the time delay between input and output considered in the NARX neural network model. k ≥ 0 means there is no delay or the delay is a non-negative value.

[0012] Let k = 0, the initial NARX network expression is simplified to:

[0013] y(n+1)=f[y(n),...,y(nd y +1),u(n),u(n-1),...,u(nd u +1)]

[0014] Selecting the first-order delay, the final NARX neural network form is:

[0015] y(n+1)=f[y(n), u(n), y(n-1), u(n-1)]

[0016] y(n+1) is the estimated value at time n+1.

[0017] The parameters to be identified are obtained by the following steps:

[0018] According to the discrete voltage equation combined with i d = 0 is converted into the q-axis discrete current equation, and the q-axis discrete current model is:

[0019] i q (k) = ai q (k-1)+b[u q (k)+u q (k-1)]+c[ω(k)i d (k)+ω(k-1)i d (k-1)]+d[ω(k)+ω(k-1)]

[0020] The parameters a, b, c, and d are expressed as:

[0021]

[0022] i q (k) is the q-axis current of the permanent magnet synchronous motor at time k, i q (k-1) is the q-axis current of the permanent magnet synchronous motor at time k-1, u q (k-1) is the q-axis voltage of the permanent magnet synchronous motor at time k-1, ω(k-1) is the speed of the permanent magnet synchronous motor at time k-1, where R s is the stator resistance of the permanent magnet synchronous motor, L is the inductance, ψ f is the permanent magnet flux linkage of the permanent magnet synchronous motor, T s is the sampling time;

[0023] The parameters to be identified are obtained through the relationship among a, b, c, and d:

[0024]

[0025] The NARX neural network model of the first-order parameters to be identified is as follows:

[0026]

[0027] f tansig represents the hyperbolic tangent Sigmoid activation function, m represents the number of neurons, W 1i , W 2i , W 3i , W 4i , W 5i represents the weight of the neural network model, b1 and b2 represent bias terms; u q (n) represents the q-axis voltage of the permanent magnet synchronous motor at time n, ω(n) represents the speed of the permanent magnet synchronous motor at time n, i d (n) is the d-axis current of the permanent magnet synchronous motor at time n, i q (n) is the q-axis current of the permanent magnet synchronous motor at time n.

[0028] The evaluation of the NARX neural network model includes the following steps:

[0029] The output of the NARX neural network model of the first-order parameters to be identified and the true value are compared with the mean square error loss function to evaluate the training error of the NARX neural network model:

[0030]

[0031] e(n) represents the error between the predicted value of the neural network model and the true value at time n; i q (n) is the q-axis current of the permanent magnet synchronous motor at time n, is the predicted value of the q-axis current of the permanent magnet synchronous motor at time n;

[0032] Compare J(n) with the threshold; if J(n) exceeds the threshold, adjust the neural network W according to the set accuracy. 1i , W 2i , W 3i , W 4i , W 5i The weights are adjusted until J(n) does not exceed the threshold range, and the identification value of the final parameter is obtained.

[0033] A permanent magnet synchronous motor parameter identification system based on NARX neural network, comprising:

[0034] A neural network model building module is used to build a NARX neural network model of first-order parameters to be identified based on a discrete current equation;

[0035] The motor parameter identification module is used to obtain the motor's historical state data and input the historical state data into the NARX neural network model to obtain the predicted value of the parameter to be identified through training.

[0036] The present invention has the following beneficial effects and advantages:

[0037] 1.NARX neural network can model complex nonlinear relationships, and the behavior of permanent magnet synchronous motors usually contains nonlinear characteristics. Compared with traditional linear model methods, NARX neural network can more accurately capture the nonlinear behavior of motor systems.

[0038] 2.NARX neural network can be adaptively adjusted according to the characteristics of input and output data. For permanent magnet synchronous motor parameter identification, the structure and parameters of the neural network can be flexibly adjusted according to the real-time collected current, voltage, speed and other data to adapt to the motor characteristics under different working conditions and environmental conditions.

[0039] 3. Compared with traditional mathematical model-based methods and experimental test methods, NARX neural network parameter identification can be performed more quickly. Once the neural network model is trained, the parameter identification process can be performed quickly in real time or offline, greatly shortening the identification time and improving the efficiency of engineering applications.

[0040] 4. The NARX neural network method is not only suitable for permanent magnet synchronous motor parameter identification, but can also be applied to parameter identification of other types of motors and complex systems. This versatility makes the NARX neural network a powerful and flexible tool for parameter identification tasks in a variety of application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Flow chart of the identification algorithm of the present invention.

[0042] Figure 2 This is the NARX neural network structure diagram. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0044] A method for identifying parameters of a permanent magnet synchronous motor based on a NARX neural network comprises the following steps:

[0045] Construct discrete voltage equations based on the mathematical model of permanent magnet synchronous motor;

[0046] Using the time delay characteristics of the NARX neural network, the first-order time delay is selected to construct the neural network;

[0047] Combining the discrete voltage equations d = 0 is converted into a discrete current equation and the estimated values ​​with identification parameters are obtained;

[0048] The objective function is obtained by combining the characteristics of discrete current with the established first-order NARX neural network;

[0049] The neural network is iterated cyclically according to the accuracy requirement to obtain the parameters to be determined, and the final motor parameter prediction values ​​are obtained according to the relationship between the parameters and the motor parameters to be identified.

[0050] The present invention provides a method for identifying parameters of a permanent magnet synchronous motor based on a NARX neural network. Figure 1 As shown in the figure, first, the mathematical model of permanent magnet synchronous motor is used to construct discrete voltage equations; then, the first-order delay is selected to construct the neural network using the delay characteristics of NARX neural network; then, the discrete voltage equation is combined with i d = 0 is converted into a discrete current equation; finally, the objective function is obtained by combining the characteristics of the discrete current with the established first-order NARX neural network, and the accuracy requirement is set to iterate the neural network to obtain the parameters to be determined, and the final motor parameter estimation value is obtained according to the relationship between the parameters and the motor parameters to be identified. The present invention improves the traditional identification method from two aspects.

[0051] Select the mathematical model under the synchronous rotating coordinate system dq, and its stator voltage equation can be expressed as:

[0052]

[0053] Among them, u d and u q are the direct and quadrature components of the voltage vector, i d and i q are the direct and quadrature components of the current vector, ψ d and ψ q are the direct-axis and quadrature-axis components of the magnetic flux, ω is the rotor angular frequency, L d , L q is the inductance on the dq axis, R s is the stator resistance.

[0054] The magnetic flux equation is:

[0055]

[0056] where ψ f is the permanent magnet flux, substituting (2) into (1), the voltage expression under the dq axis can be obtained as follows:

[0057]

[0058] Where p is the differential operator. In the dq rotating reference frame, the inductance of the surface mounted permanent magnet synchronous motor is the same, that is, L = L d =L q , uniformly recorded as L. When the sampling time is small enough, the discrete voltage equation is as follows:

[0059]

[0060]

[0061] Where T s is the sampling time.

[0062] Select the NARX neural network with first-order delay, the form of the neural network is as follows:

[0063] y(n+1)=f[u(n),y(n),u(n-1),y(n-1)] (6)

[0064] Then combine the discrete voltage equation with i d = 0 is converted into a discrete current equation, which is as follows:

[0065] i q (k) = ai q (k-1)+b[u q (k)+u q (k-1)]+c[ω(k)i d (k)+ω(k-1)i d (k-1)]+d[ω(k)+ω(k-1)] (7)

[0066] Where a, b, c, d can be expressed as:

[0067]

[0068] Where a, b, c, and d are constants related to the parameters to be identified.

[0069] In formula (7), i q (k) is the q-axis current of the permanent magnet synchronous motor at time k, i q (k-1) is the q-axis current of the permanent magnet synchronous motor at time k-1, u q (k-1) is the q-axis voltage of the permanent magnet synchronous motor at time k-1, ω(k-1) is the speed of the permanent magnet synchronous motor at time k-1, where R s is the stator resistance of the permanent magnet synchronous motor, L is the inductance, ψ f is the permanent magnet flux linkage of the permanent magnet synchronous motor, Ts is the sampling time.

[0070] Through the relationship between a, b, c, d, we can deduce:

[0071]

[0072] Combining the selected NARX neural network expression with first-order time delay and the discretized expression of q-axis current, the final NARX neural network form can be obtained as follows:

[0073]

[0074] The choice of activation function will directly affect the convergence, generalization ability and complexity of the training process of the neural network, so we choose hyperbolic tangent Sigmoid as the transfer function.

[0075] The NARX neural network expression for parameter identification is as follows: Figure 2 :

[0076]

[0077] where f tansig represents the hyperbolic tangent Sigmoid activation function, m represents the number of neurons, W 1i , W 2i , W 3i , W 4i , W 5i represents the weight of the neural network model, b1 and b2 represent bias terms; u q (n) represents the q-axis voltage of the permanent magnet synchronous motor at time n, ω(n) represents the speed of the permanent magnet synchronous motor at time n, i d (n) is the d-axis current of the permanent magnet synchronous motor at time n, i q (n) is the q-axis current of the permanent magnet synchronous motor at time n.

[0078] Finally, the mean square error (MSE) is selected as the loss function, and the performance indicators are expressed as follows:

[0079]

[0080] where i q (n) is the true value of the q-axis current at time n.

[0081] In the identification process, the weights of the neural network are continuously adjusted by minimizing the loss function to obtain the estimated values ​​of parameters a, b, c, and d, and then the R of the motor parameters is calculated. s , L, ψ f The estimated value of .

Claims

1. A permanent magnet synchronous motor parameter identification method based on NARX neural network, characterized in that: The following steps are involved: According to the discrete current equation, a NARX neural network model of first-order parameters to be identified is constructed; The historical state data of the motor is obtained and input into the NARX neural network model to obtain the predicted values ​​of the parameters to be identified through training.

2. According to claim 1, a permanent magnet synchronous motor parameter identification method based on NARX neural network is characterized in that: The NARX neural network model is obtained by the following steps: The initial NARX network expression is: y(n+1)=f[y(n),...,y(n-d y +1),u(n-k),u(n-k-1),...,u(n-k-d u +1)] Among them, u(n)∈R and y(n)∈R represent the input and output of the NARX neural network model when the discrete time step is n, R is a real number, d u ≥1,d y ≥1 and d u ≤d y They represent the order of input memory and output memory respectively. The parameter k is the delay term, which is used to represent the time delay between input and output considered in the NARX neural network model. k ≥ 0 means there is no delay or the delay is a non-negative value. Let k = 0, the initial NARX network expression is simplified to: y(n+1)=f[y(n),...,y(n-d y +1),u(n),u(n-1),...,u(n-d u +1)] Selecting the first-order delay, the final NARX neural network form is: y(n+1)=f[y(n),u(n),y(n-1),u(n-1)] y(n+1) is the predicted value at time n+1.

3. The method for identifying parameters of a permanent magnet synchronous motor based on a NARX neural network according to claim 1 is characterized in that: The parameters to be identified are obtained by the following steps: According to the discrete voltage equation combined with i d = 0 is converted into the q-axis discrete current equation, and the q-axis discrete current model is: i q (k)=ai q (k-1)+b[u q (k)+u q (k-1)] +c[ω(k)i d (k)+ω(k-1)i d (k-1)]+d[ω(k)+ω(k-1)] Among them, parameters a, b, c, d are expressed as: i q (k) is the q-axis current of the permanent magnet synchronous motor at time k, i q (k-1) is the q-axis current of the permanent magnet synchronous motor at time k-1, u q (k-1) is the q-axis voltage of the permanent magnet synchronous motor at time k-1, ω(k-1) is the speed of the permanent magnet synchronous motor at time k-1, where R s is the stator resistance of the permanent magnet synchronous motor, L is the inductance, ψ f is the permanent magnet flux linkage of the permanent magnet synchronous motor, T s is the sampling time; i d (k) is the d-axis current of the permanent magnet synchronous motor at time k; The parameters to be identified are obtained through the relationship among a, b, c, and d:

4. The method for identifying parameters of a permanent magnet synchronous motor based on a NARX neural network according to claim 1 is characterized in that: The NARX neural network model of the first-order parameters to be identified is as follows: f tansig represents the hyperbolic tangent Sigmoid activation function, m represents the number of neurons, W 1i , W 2i , W 3i , W 4i , W 5i represents the weight of the neural network model, b1 and b2 represent bias terms; u q (n) represents the q-axis voltage of the permanent magnet synchronous motor at time n, ω(n) represents the speed of the permanent magnet synchronous motor at time n, i d (n) is the d-axis current of the permanent magnet synchronous motor at time n, i q (n) is the q-axis current of the permanent magnet synchronous motor at time n.

5. The method for identifying permanent magnet synchronous motor numbers based on NARX neural network according to claim 1 is characterized in that: The evaluation of the NARX neural network model includes the following steps: The output of the NARX neural network model of the first-order parameters to be identified and the true value are compared with the mean square error loss function to evaluate the training error of the NARX neural network model: e(n) represents the error between the predicted value of the neural network model and the true value at time n; i q (n) is the q-axis current of the permanent magnet synchronous motor at time n, is the predicted value of the q-axis current of the permanent magnet synchronous motor at time n; Compare J(n) with the threshold; if J(n) exceeds the threshold, adjust the neural network W according to the set accuracy. 1i , W 2i , W 3i , W 4i , W 5i The weights are adjusted until J(n) does not exceed the threshold range, and the identification value of the final parameter is obtained; otherwise, the identification value of the parameter is directly obtained.

6. A permanent magnet synchronous motor parameter identification system based on NARX neural network, characterized in that: include: A neural network model building module is used to build a NARX neural network model of first-order parameters to be identified based on a discrete current equation; The motor parameter identification module is used to obtain the motor's historical state data and input the historical state data into the NARX neural network model to obtain the predicted value of the parameter to be identified through training.

Images