State estimation method of electric drive system based on variational Bayesian adaptive particle filter

Through the variational Bayes adaptive particle filter method, the shortcomings of the electric vehicle electric drive system are solved, and the accurate state estimation of the rotor position, speed and control torque are achieved, which improves the estimation accuracy and robust performance , suitable for complex working conditions and non-Gaussian noise environments.

CN118410266BActive Publication Date: 2024-12-03SUZHOU VOCATIONAL UNIVERSITY (SUZHOU OPEN UNIVERSITY)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410460493.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-17
Publication Date
2024-12-03
Estimated Expiration
2044-04-17

AI Technical Summary

Technical Problem

The prior art cannot effectively estimate the rotor position, speed and control torque status of electric vehicle electric drive systems, especially in complex multi-operating conditions and non-Gaussian noise, resulting in inaccurate and poor robustness of state estimation.

Method used

The variational Bayesian adaptive particle filter method is adopted, combined with variational Bayesian estimation and adaptability, and the Belgian posterior probability distribution is approximate to the particle filter, and the system model state parameters are adjusted in real time, which is suitable for electric vehicle electric drives The system's strongly coupled nonlinear system realizes state estimation of rotor position, speed and control torque.

Benefits of technology

Improve the accuracy and robust performance of the electric drive system state estimation of electric vehicles, and can accurately estimate the system state under complex operating conditions, avoid applying too much control input energy to the electric vehicle, and overcome inaccurate system noise statistics. Effect of filtering accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118410266B_ABST
    Figure CN118410266B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for estimating the state of an electric drive system using a variational Bayesian adaptive particle filter. First, a rotor position-speed model and a torque-speed model of an electric drive system of an electric vehicle are given. The filter provides an effective way to estimate the state of the system for a strongly coupled nonlinear system and non-Gaussian noise situation such as an electric drive system of an electric vehicle. The main idea of ​​the variational Bayesian adaptive particle filter method is to use a variational inference method to approximate the Bayesian posterior probability distribution based on a variational Bayesian estimation algorithm, and adjust the system model state parameter estimation characteristics in real time according to the principle of adaptability. When dealing with non-Gaussian disturbances encountered in complex working conditions of an electric drive system of an electric vehicle, an accurate estimation of the system state is achieved, which can better cope with the complexity and uncertainty of the system and improve the accuracy and robustness of the estimation of the real-time state of the system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of electric drive control equipment, and in particular relates to an electric drive system state estimation method of a variational Bayesian adaptive particle filter. Background Art

[0002] The global carbon reduction policy has brought about the inevitable trend of comprehensive electrification of vehicles. With the rapid development of the new energy vehicle market, highly integrated integrated electric drive systems have emerged. This move can not only achieve weight reduction and cost reduction, but also effectively improve power density indicators. The electric drive system is an important system for realizing electric vehicles. If the online status can be discovered and determined in the first time during driving, it is of great significance to ensure the safe operation of electric vehicles. With the rapid development of electric vehicles, higher requirements are also placed on the performance and control accuracy of electric vehicles. State estimation is one of the important guarantees for the safe operation of electric drive systems of electric vehicles. It can provide accurate input signals by estimating the vehicle's motion state to achieve precise control of electric vehicles. In view of the complex and multi-operating conditions of motor-driven electric vehicles, the present invention proposes a basic scheme for the state estimation method of electric drive systems of electric vehicles using a variational Bayesian adaptive particle filter.

[0003] At present, after a large number of literature and patent searches, the main patents involved include, for example, a collaborative control method based on state estimation of electric drive system, a lateral control method and system based on collaborative estimation of multi-state parameters, and a hybrid order electromechanical system collaborative control method based on estimated state disturbance observation. None of the above methods realizes the state estimation of rotor position, speed and control torque of electric drive system of electric vehicles. Summary of the invention

[0004] Purpose of the invention: In view of the shortcomings of the prior art, the present invention provides a state estimation method of an electric drive system based on a variational Bayesian adaptive particle filter. The method combines the concepts of variational Bayesian estimation, adaptability and particle filtering to solve the online state estimation problem of the rotor position, speed and control torque of the electric drive system of an electric vehicle. It is not only suitable for estimating the steady-state state of the electric drive system of an electric vehicle, but also can better reflect the advantages of the method for the transient state of the motor. In the present invention, the rotor position-speed model and torque-speed model of the electric drive system of an electric vehicle are first given. This filter is specifically designed for a strongly coupled nonlinear system and non-Gaussian noise situation such as the electric drive system of an electric vehicle, and provides an effective way to estimate the state of the system. The main idea of ​​the variational Bayesian adaptive particle filter method is to use the variational inference method to approximate the Bayesian posterior probability distribution based on the variational Bayesian estimation algorithm, and adjust the system model state parameter estimation characteristics in real time according to the principle of adaptability. When dealing with non-Gaussian disturbances encountered in complex working conditions of the electric drive system of an electric vehicle, an accurate estimation of the system state is achieved, which can better cope with the complexity and uncertainty of the system and improve the estimation accuracy and robustness of the real-time state of the system.

[0005] Technical solution: A method for estimating the state of an electric drive system using a variational Bayesian adaptive particle filter of the present invention comprises the following steps:

[0006] Step 1: Based on the speed, rotor position and speed uncertainty interference of the electric drive system during the driving process of the target electric vehicle, a rotor position-speed model and a torque-speed model of the electric drive system of the electric vehicle are constructed;

[0007] Step 2: Based on the rotor position-speed model and torque-speed model of the electric drive system of the electric vehicle, a variational Bayesian adaptive particle filter estimation algorithm is introduced, and the variational inference method is used to approximate the Bayesian posterior probability distribution;

[0008] Step 3: Estimate the Bayesian posterior probability distribution through a set of weighted particle sample sets, and continuously update the particle sample sets and particle weights to obtain the actual values ​​of the target states corresponding to the position, speed and control torque of the target electric vehicle electric drive system.

[0009] Further, step 1 is specifically as follows: the electric vehicle electric drive system rotor position-speed model S k+1 and torque-speed model V k+1 The discrete equation is:

[0010]

[0011]

[0012] Where: w s,k and w v,kare the uncertain disturbances of rotor position and speed in the electric drive system of electric vehicles, T s is the sampling time of the electric drive system, S k is the rotor position of the electric drive system, V k is the speed of the electric drive system, T k is the control torque, β1 is the rotor position coefficient, β2 is the torque coefficient, is the driving torque, and They represent the estimated values ​​of the rotor position and speed respectively. The information transmission of the electric drive system will be affected by the operating conditions during driving.

[0013] Furthermore, step 2 specifically includes the following steps:

[0014] Step 2.1, the variational Bayesian adaptive particle filter estimation algorithm includes two steps: prediction and update;

[0015] Step 2.2, in the prediction step, the values ​​of the rotor position and torque parameters of the electric drive system depend on the previous measurement. After the time is updated, the prior distribution of the parameters is obtained, the state vectors of the rotor position, speed and control torque are initialized, and the time update of the prediction process is improved;

[0016] Step 2.3: In the updating step, the covariance matrix of the electric drive system is measured and updated based on variational Bayes.

[0017] Furthermore, in step 2.2, the state vectors of the rotor position, speed and control torque are initialized as follows:

[0018]

[0019]

[0020] Where: is the filter propagation matrix, is the estimated value of the rotor position of the electric drive system, is the estimated value of the rotor position at time k-1, is the estimated value of the rotor speed at time k-1, w s,k-1 and w v,k-1 are the uncertain disturbances of rotor position and speed of the electric vehicle electric drive system at time k-1, For S k-1 is the rotor position of the electric drive system at time k-1.

[0021] Furthermore, in order to make the electric drive system track the output of the actual model in real time under the disturbance state, it is necessary to make the control variable u converge to zero. are the three elements of the adaptive controller u parameter vector, which respectively represent the propagation equations of the electric drive system position, speed and control torque. By selecting the Lyapunov equation, the adaptive controller u is given as:

[0022]

[0023] In formula (5): ξ is the generalized error vector, ξ T is the transposed matrix of ξ, is the parameter error vector, A is a symmetric matrix, P is a positive definite matrix, Γ is the adaptive gain matrix, A T is the transposed matrix of A, yes The derivative of .

[0024] Furthermore, in step 2.3, the electric drive system covariance matrix measurement update based on variational Bayes is as follows:

[0025]

[0026] Where: ρ is the probability distribution coefficient, 0<ρ≤1, and its function is to make the electric drive system measurement noise covariance matrix change with a certain probability distribution, so that the probability density function form of the posterior and the prior is the same; represents the Gaussian density distribution of particles; are the normal coefficients of the position and speed of the Hurwitz stable polynomial electric drive system, x k-1 is the state vector at time k-1, For x k-1 The estimated value, P k-1 is the k-1 moment covariance matrix.

[0027] Furthermore, step 3 specifically includes the following steps:

[0028] Step 3.1, construct the target tracking model state equations about position, speed and control torque in the particle filter, and obtain the calculation results of particle weights;

[0029] Step 3.2: Minimize the rotor position, speed and control torque state vector at time k-1 The Frobenius norm of After the measurement is updated, the weights are normalized to obtain the actual values ​​of the target states corresponding to each particle of the electric drive system position, speed, and control torque.

[0030] Furthermore, step 3.1 is specifically as follows:

[0031] The target tracking model state equations for position, velocity and control torque in the particle filter are:

[0032]

[0033] Where: Represented as the target state at time k-1, ∈ k-1 is the control input at time k-1, α k-1 is the process disturbance at time k-1, and the driving process noise variation matrix Q k , particle importance weight It is expressed as:

[0034]

[0035] Where: is the weight value at time k-1, p(τ) is the prior probability distribution, y i () is the importance density function, which takes the i-th target feature value extracted at time k Substituting into equation (7), the sampled particles used to update the target feature value at the next moment can obtain the calculation result of the particle weight.

[0036] Furthermore, step 3.2 is specifically as follows:

[0037] To minimize the rotor position, speed and control torque at time k-1, the state vector The Frobenius norm of After the measurement update, it is rewritten as:

[0038]

[0039] Normalize the weights:

[0040]

[0041] Where: is the Kalman gain, is the j+1-step covariance matrix of the electric drive system, is the electric drive system measurement noise variance matrix, N L is the number of particles. If the result of formula (10) is lower than the particle importance weight Then, the target particle sampling operation of the sample space is performed, and the particles with large weights in the rotor position, speed and control torque are decomposed into multiple particles to replace the particles with small weights in the sample space. Otherwise, there is no need to perform particle resampling operation, and the above steps are repeated to obtain the actual value of the target state corresponding to each particle of the electric drive system position, speed and control torque.

[0042] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0043] 1. The variational Bayesian algorithm used in the present invention is a type of technology used for approximate calculation of complex integrals in the field of Bayesian estimation and machine learning. It is a variational Bayesian adaptive particle filter method formed by combining adaptive control and particle filter. Compared with artificial intelligence methods such as particle swarm algorithm and artificial neural network, it has been greatly improved, and further avoids applying too much control input energy to the electric drive system of electric vehicles. This control scheme can achieve state estimation of the rotor position, speed and control torque of the electric drive system within a limited time under certain disturbances.

[0044] 2. Compared with the particle filter, the variational Bayesian electric vehicle electric drive system based on the adaptive particle filter has a good real-time tracking effect of filtering accuracy, which can effectively overcome the influence of inaccurate system noise statistics on filtering accuracy, and then quickly obtain the state estimation of the electric vehicle electric drive system during driving.

[0045] 3. The adaptive particle filter relies too much on the mathematical model of the accurate system noise and measurement noise statistics of the electric drive system. When the electric drive system is online and in an interference environment such as electromagnetic noise, the statistical model cannot be accurately estimated, resulting in the system noise and measurement noise not satisfying the Gaussian distribution, which in turn affects the filtering performance and even causes it to diverge. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of the present invention;

[0047] Figure 2 It is the rotor position state estimation diagram of the electric drive system;

[0048] Figure 3 It is the rotor speed state estimation diagram of the electric drive system;

[0049] Figure 4 This is the torque state estimation diagram for the electric drive system control. DETAILED DESCRIPTION

[0050] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for estimating the state of an electric vehicle electric drive system based on a variational Bayesian adaptive particle filter, such as Figure 1 As shown, the following steps are included:

[0051] Considering the uncertain interferences encountered by various road conditions such as acceleration and deceleration during the driving of electric vehicles, the discrete equations of the rotor position-speed model (1) and torque-speed model (2) of the electric vehicle electric drive system can be obtained:

[0052]

[0053]

[0054] In formula (1-2): ws,k and w v,k are the uncertain disturbances of rotor position and speed in the electric drive system of electric vehicles, T s is the sampling time of the electric drive system, S k is the rotor position of the electric drive system, V k is the speed of the electric drive system, T k is the control torque, β1 is the rotor position coefficient, β2 is the torque coefficient, is the driving torque, and They represent the estimated values ​​of the rotor position and speed respectively. The information transmission of the electric drive system will be affected by the operating conditions during driving.

[0055] In the variational Bayesian adaptive particle filter method, each step consists of prediction and update. In the prediction step, the values ​​of the rotor position and torque parameters of the electric drive system depend on the previous measurement. After the time is updated, the prior distribution of the parameters is obtained, and the state vector x of the rotor position, speed and control torque is initialized. k , improving the time update of the prediction process.

[0056] Rotor position and torque parameter state vector prediction:

[0057]

[0058]

[0059] In formula (3-4): is the filter propagation matrix, is the estimated value of the rotor position of the electric drive system, is the estimated value of the rotor position at time k-1, is the estimated value of the rotor speed at time k-1, w s,k-1 and w v,k-1 are the uncertain disturbances of rotor position and speed of the electric vehicle electric drive system at time k-1, For S k-1 is the rotor position of the electric drive system at time k-1.

[0060] In order to make the electric drive system track the output of the actual model in real time under disturbance conditions, it is necessary to make the control variable u converge to zero. are the three elements of the adaptive controller u parameter vector, which respectively represent the propagation equations of the electric drive system position, speed and control torque. By selecting the Lyapunov equation, the adaptive controller u is given as:

[0061]

[0062] In formula (5): ξ is the generalized error vector, ξ T is the transposed matrix of ξ, is the parameter error vector, A is a symmetric matrix, P is a positive definite matrix, and Γ is the adaptive gain matrix. T is the transposed matrix of A. yes The derivative of .

[0063] Measurement update of the covariance matrix of the electric drive system based on variational Bayes:

[0064]

[0065] In formula (6), ρ is the probability distribution coefficient, 0<ρ≤1, and its function is to make the electric drive system measurement noise covariance matrix change with a certain probability distribution, so as to make the probability density function form of the posterior and the prior the same. Represents the Gaussian density distribution of particles. are the normal coefficients of the position and speed of the Hurwitz stable polynomial electric drive system, x k-1 is the state vector at time k-1, For x k-1 The estimated value, P k-1 is the k-1 moment covariance matrix.

[0066] Combined with the above variational Bayesian adaptive particle filter, the posterior probability density is estimated through a set of weighted particle sample sets, and the particle sample sets and particle weights are continuously updated, thereby achieving accurate estimation of the target state. The target tracking model state equations for position, speed and control torque in the particle filter are:

[0067]

[0068] In formula (7): Represented as the target state at time k-1, ∈ k-1 is the control input at time k-1, α k-1 is the process disturbance at time k-1, and the driving process noise variation matrix Q k , particle importance weight It is expressed as:

[0069]

[0070] In formula (8): is the weight value at time k-1, p(τ) is the prior probability distribution, y i () is the importance density function, which takes the i-th target feature value extracted at time k Substitute it into formula (7) to update the sampling particles of the target feature value at the next moment. Then the calculation result of the particle weight can be obtained.

[0071] In order to minimize the rotor position, speed and control torque, the state vector at time k-1 The Frobenius norm of After the measurement update, it is rewritten as:

[0072]

[0073] Normalize the weights:

[0074]

[0075] In formula (9-10): is the Kalman gain, is the j+1-step covariance matrix of the electric drive system, is the electric drive system measurement noise variance matrix, N L is the number of particles. If the result of formula (10) is lower than the particle importance weight Then, the target particle sampling operation of the sample space is performed, and the particles with large weights in the rotor position, speed and control torque are decomposed into multiple particles to replace the particles with small weights in the sample space. Otherwise, there is no need to perform particle resampling operation, and the above steps are repeated to obtain the actual values ​​of the target states corresponding to the particles of the rotor position, speed and control torque of the electric drive system.

[0076] like Figure 2 As shown, the actual value of the rotor position target state tracking is measured by the electric drive system state estimation method based on the variational Bayesian adaptive particle filter.

[0077] like Figure 3 As shown, the actual value of speed tracking is measured by the electric drive system state estimation method based on variational Bayesian adaptive particle filter.

[0078] like Figure 4 As shown, the control torque tracks the actual value measured by the electric drive system state estimation method based on the variational Bayesian adaptive particle filter.

Claims

1. A method for estimating the state of an electric drive system based on a variational Bayesian adaptive particle filter, characterized in that: The steps include: Step 1: Based on the speed, rotor position and speed uncertainty interference of the electric drive system during the driving process of the target electric vehicle, a rotor position-speed model and a torque-speed model of the electric drive system of the electric vehicle are constructed; Step 2: Based on the rotor position-speed model and torque-speed model of the electric drive system of the electric vehicle, a variational Bayesian adaptive particle filter estimation algorithm is introduced, and the variational inference method is used to approximate the Bayesian posterior probability distribution; Step 3: Estimate the Bayesian posterior probability distribution through a set of weighted particle sample sets, and continuously update the particle sample sets and particle weights to obtain the actual values ​​of the target states corresponding to the position, speed and control torque of the electric drive system of the target electric vehicle; Step 1 is as follows: Electric vehicle electric drive system rotor position-speed model S k+1 and torque-speed model V k+1 The discrete equation is: Where: w s,k and w v,k are the uncertain disturbances of rotor position and speed in the electric drive system of electric vehicles, T s is the sampling time of the electric drive system, S k is the rotor position of the electric drive system, V k is the speed of the electric drive system, T k is the control torque, β1 is the rotor position coefficient, β2 is the torque coefficient, is the driving torque, and They represent the estimated values ​​of the rotor position and speed respectively. When the vehicle is driving, the information transmission of the electric drive system will be disturbed by the working conditions during driving; Step 2 specifically includes the following steps: Step 2.1, the variational Bayesian adaptive particle filter estimation algorithm includes two steps: prediction and update; Step 2.2, in the prediction step, the values ​​of the rotor position and torque parameters of the electric drive system depend on the previous measurement. After the time is updated, the prior distribution of the parameters is obtained, the state vectors of the rotor position, speed and control torque are initialized, and the time update of the prediction process is improved; Step 2.3, in the updating step, the covariance matrix of the electric drive system is measured and updated based on variational Bayes; In step 2.2, the state vectors of the rotor position, speed and control torque are initialized as follows: Where: is the filter propagation matrix, is the estimated value of the rotor position of the electric drive system, is the estimated value of the rotor position from time k to k-1, is the estimated value of the rotor speed at time k-1, w s,k-1 and w v,k-1 are the uncertain disturbances of rotor position and speed of the electric vehicle electric drive system at time k-1, is the estimated speed of the electric drive system at time k-1, S k-1 is the rotor position of the electric drive system at time k-1; In order to make the electric drive system track the output of the actual model in real time under disturbance conditions, it is necessary to make the control variable u converge to zero. are the three elements of the adaptive controller u parameter vector, which represent the propagation equations of the rotor position estimate, speed and control torque of the electric drive system. By selecting the Lyapunov equation, the adaptive controller u is given as: In formula (5): ξ is the generalized error vector, ξ T is the transposed matrix of ξ, A is a symmetric matrix, P is a positive definite matrix, A T is the transposed matrix of A, yes The derivative of .

2. The electric drive system state estimation method based on a variational Bayesian adaptive particle filter according to claim 1 is characterized in that: In step 2.3, the electric drive system covariance matrix measurement update based on variational Bayes is as follows: Where: ρ is the probability distribution coefficient, 0<ρ≤1, and its function is to make the electric drive system measurement noise covariance matrix change with a certain probability distribution, so that the probability density function form of the posterior and the prior is the same; represents the Gaussian density distribution of particles; are the normal coefficients of the position and speed of the Hurwitz stable polynomial electric drive system, x k-1 is the state vector at time k-1, For x k-1 The estimated value, P k-1 is the k-1 moment covariance matrix.

3. The electric drive system state estimation method based on a variational Bayesian adaptive particle filter according to claim 1 is characterized in that: Step 3 specifically includes the following steps: Step 3.1, construct the target tracking model state equations about position, speed and control torque in the particle filter, and obtain the calculation results of particle weights; Step 3.2: Minimize the rotor position, speed and control torque state vector at time k-1 The Frobenius norm of After the measurement is updated, the weights are normalized to obtain the actual values ​​of the target states corresponding to each particle of the electric drive system position, speed, and control torque.

4. The electric drive system state estimation method of a variational Bayesian adaptive particle filter according to claim 3 is characterized in that: Step 3.1 is as follows: The target tracking model state equations for position, velocity and control torque in the particle filter are: Where: Represented as the target state at time k-1, ∈ k-1 is the control input at time k-1, α k-1 is the process disturbance at time k-1, and the driving process noise variation matrix Q k , particle importance weight It is expressed as: Where: is the weight value at time k-1, p(τ) is the prior probability distribution, y i () is the importance density function, which takes the i-th target feature value extracted at time k Substituting into equation (7), the sampled particles used to update the target feature value at the next moment can obtain the calculation result of the particle weight.

5. The electric drive system state estimation method of a variational Bayesian adaptive particle filter according to claim 3 is characterized in that: Step 3.2 is as follows: To minimize the rotor position, speed and control torque at time k-1, the state vector The Frobenius norm of After the measurement update, it is rewritten as: Normalize the weights: Where: is the Kalman gain, is the j+1-step covariance matrix of the electric drive system, is the electric drive system measurement noise variance matrix, N L is the number of particles. If the result of formula (10) is lower than the particle importance weight Then, the target particle sampling operation of the sample space is performed, and the particles with large weights in the rotor position, speed and control torque are decomposed into multiple particles to replace the particles with small weights in the sample space. Otherwise, there is no need to perform particle resampling operation, and the above steps are repeated to obtain the actual value of the target state corresponding to each particle of the electric drive system position, speed and control torque.