SOC estimation method based on adaptive fuzzy neural network error compensation

Abstract

The invention provides an SOC estimation method based on adaptive fuzzy neural network error compensation, and belongs to the technical field of lithium ion batteries. The problem that the AEKF prediction SOC error is too large is solved. According to the technical scheme, the method comprises the following steps: 1) carrying out an intermittent constant-current discharge experiment on the lithium ion battery; 2) establishing a second-order RC equivalent circuit model; 3) performing lithium battery SOC rough estimation by using AEKF; 4) proposing an error compensation network based on an adaptive fuzzy neural network; and 5) compensating a rough prediction result by using a prediction error obtained by the adaptive fuzzy neural network. The method has the beneficial effects that the SOC is predicted and corrected, and the precision is very high.

Description

Technical Field

[0001] The invention relates to the technical field of lithium ion batteries, and in particular to a SOC estimation method based on adaptive fuzzy neural network error compensation. Background Art

[0002] Lithium batteries have the characteristics of high energy density, high voltage output, low self-discharge rate, no memory effect, long service life and fast charging capability. These characteristics make them widely used in fields such as electric vehicles and engineering equipment.

[0003] As people's technical requirements for battery management systems continue to increase, the prediction of key battery parameters has received more and more attention. For example, the estimation of battery state of charge (SOC) and battery health status (SOH). The prediction of these key parameters affects the energy utilization rate of the battery and the safe operation of the battery. In "Lithium-ion Battery SOC Estimation Method for Wide Temperature Range", a multi-time scale joint estimation method for lithium-ion battery SOC under a wide temperature range based on the multi-information adaptive robust unscented Kalman filter (MIARUKF) algorithm is proposed. In "Lithium Battery SOC Estimation Based on Dynamic Forgetting Factor Recursive Least Squares Method and Improved Particle Filter Algorithm", the dynamic genetic factor is introduced to improve the SOC prediction accuracy in order to solve the problem that the fixed forgetting factor recursive least squares method is difficult to maintain fast convergence and stability in battery parameter identification. The second-order RC equivalent model has relatively small computational complexity, simple structure and high accuracy. The core idea of ​​the adaptive extended Kalman filter is to predict the new system state through the system model and use actual observations to correct these predictions, so as to accurately estimate the true state of the system. This filtering method is particularly suitable for noisy data and environmental uncertainties, and can effectively improve the accuracy and reliability of data. However, since both the state function and the measurement function are nonlinear, errors caused by Taylor expansion and first-order approximation will occur. How to solve the above technical problems is a challenge faced by the present invention. Summary of the invention

[0004] The purpose of the present invention is to provide a SOC estimation method based on adaptive fuzzy neural network error compensation. In actual work, lithium-ion batteries are affected by various factors such as temperature and operating conditions, and the lithium battery itself has certain nonlinearity. The use of an adaptive extended Kalman filter based on a second-order RC circuit model to estimate SOC has the characteristics of small calculation amount and good convergence. However, the adaptive extended Kalman filter itself will cause errors due to Taylor expansion. Considering the above problems, the present invention introduces an error compensation module based on an adaptive fuzzy neural network to improve the accuracy of SOC estimation using an adaptive extended Kalman filter without increasing the amount of calculation. The present invention uses an SOC estimation method based on adaptive fuzzy neural network error compensation to accurately predict the SOC of lithium batteries, and verifies the accuracy using a variety of operating conditions. The results show that the SOC estimation method based on adaptive fuzzy neural network error compensation has high accuracy, and the predicted SOC error is significantly lower than other methods.

[0005] The present invention is achieved by the following measures: A SOC estimation method based on adaptive fuzzy neural network error compensation comprises the following steps:

[0006] Step 1) performing an intermittent constant current discharge experiment on the lithium-ion battery, measuring its terminal voltage and load current data, and determining the functional relationship of OCV-SOC by polynomial fitting, so as to calculate the open circuit voltage of the lithium battery at each moment;

[0007] Step 2) establishing a second-order RC equivalent circuit model, and using a recursive Bayesian algorithm to perform parameter identification on the established model;

[0008] Step 3) using an adaptive extended Kalman filter to perform a rough prediction of the lithium battery SOC;

[0009] Step 4) an error compensation network based on an adaptive fuzzy neural network is proposed, the noise covariance matrix and Kalman gain generated in the adaptive extended Kalman filter estimation process are used as the input of the trained error compensation network, and the prediction error of SOC is output;

[0010] Step 5) Use the prediction error obtained by the adaptive fuzzy neural network to compensate the rough prediction result.

[0011] As a SOC estimation method based on adaptive fuzzy neural network error compensation of the present invention, the specific steps of step 2) are as follows:

[0012] Establish a second-order RC equivalent circuit model. This model uses R 1 and C 1 The parallel structure represents the electrochemical polarization effect of the battery, using R 2 and C 2The parallel structure represents the concentration difference polarization effect of the battery. The open circuit voltage and terminal voltage are U oc and U. U 1 , U 2 Represents the voltage of two RC parallel links, R omc Indicates the ohmic internal resistance of the battery;

[0013] The SOC of a lithium-ion battery is defined as the ratio of its actual remaining capacity to its rated capacity, expressed as:

[0014]

[0015] Among them, Q n is the rated capacity of the battery, t 0 is the initial time, t 1 is the current moment, and I is the battery load current.

[0016] Based on Kirchhoff's law, the transfer function of the system can be obtained as:

[0017]

[0018] Mapping the above formula from the s plane to the z plane, we can get:

[0019]

[0020] Where T is the sampling period, τ 1 =R 1 C 1 , τ 2 =R 2 C 2 , a=R omc , b = τ 1 τ 2 , c = τ 1 +τ 2 , d = R omc +R 1 +R 2 , e=R omc (τ 1 +τ 2 )+R 1 τ 1 +R 2 τ 2 .

[0021] The difference equation obtained after discretization of the transfer function is:

[0022]

[0023] Let y(t) = U(t) - U oc(t), the autoregressive (ARX) model with external input that is suitable for parameter identification of lithium-ion batteries can be obtained as follows:

[0024]

[0025] in, θ=[a 1 ,a 2 ,b 0 ,b 1 ,b 2 ] T Through the recursive Bayesian algorithm, the elements in θ can be identified. According to the functional relationship between a, b, c, d, e and the elements in θ, a, b, c, d, e can be derived, and finally the model parameter R can be obtained. omc , R 1 , R 2 , C 1 , C 2 .

[0026] The main formula of the recursive Bayesian algorithm is:

[0027]

[0028] The steps of implementing the recursive Bayesian identification algorithm to identify the battery system parameters are as follows:

[0029] (1) Initialization parameters: set the sampling period T, P(0) and variance σ v Initial value;

[0030] (2) SOC(k) is calculated using formula (1), and the open circuit voltage U is calculated based on the functional relationship between OCV and SOC. oc (k);

[0031] (3) According to formula (5), we get the information vector and output vector y(k);

[0032] (4) According to formula (7), the intermediate vector L(k) at time k is calculated in sequence, and the parameter estimation Covariance matrix P(k);

[0033] (5) Separate the parameters and obtain the parameters of the equivalent circuit model;

[0034] Repeat steps (2) to (5) until the end time is reached and then stop the identification.

[0035] Furthermore, the step 3) comprises the following steps:

[0036] According to the battery SOC defined in step 2), its discrete form can be defined as:

[0037]

[0038] Where ζ represents the charge and discharge efficiency of the battery. The state equation and measurement equation of the second-order RC equivalent circuit model are:

[0039]

[0040] Where X k and k are the state vector and measurement vector at time k respectively. k is the system input at time k. f(·) and h(·) are nonlinear functions. w k and v k are process noise and measurement noise, respectively, both are white Gaussian noise. Q and R are the corresponding process noise covariance matrix and measurement noise covariance matrix.

[0041] Establish the state space equation for lithium battery SOC estimation:

[0042]

[0043] The prediction process and update process of the adaptive extended Kalman filter can be expressed as:

[0044]

[0045] Based on the traditional Kalman filter, the adaptive Kalman filter can correct the noise matrix in real time:

[0046]

[0047] Where L is the window size for covariance matching.

[0048] The adaptive extended Kalman filter is used to update the state matrix to achieve a rough estimate of the lithium battery SOC and obtain the first diagonal element of the filter parameter covariance matrix at time k. The first element K of the Kalman filter gain matrix k1 .

[0049] Furthermore, the step 4) comprises the following steps:

[0050] Step 4-1) In order to obtain the linearization loss in the filtering process from the training data through the compensation network, the coarse filtering result SOC 0The difference between the SOC obtained by the ampere-hour integration method is defined as ΔSOC. In the SOC estimation process of the adaptive extended Kalman filter, the error of the nonlinear system is mainly reflected in the prediction process of the state covariance, and the resulting error is further introduced into the SOC estimation through the residual and gain of the system. Therefore, the first element of the diagonal element of the covariance matrix First element as network input.

[0051] Network input:

[0052] Network output: Outout j ={[ΔSOC] 1j ,…,[ΔSOC] nj}

[0053] Step 4-2) The training process is:

[0054] (1) Input parameters Perform fuzzy processing.

[0055] (2) The membership degree of each variable is multiplied to obtain the trigger strength of each rule.

[0056] (3) Normalize the penalty intensity of each layer.

[0057] (4) Make a linear output of the normalized trigger intensity.

[0058] (5) Perform weighted summation and defuzzification on the results of each rule.

[0059] (6) Use the output of the adaptive fuzzy neural network to obtain SOC compensation, calculate the error and integrate it into the result. If the conditions are met, the trained network structure will be saved. Otherwise, the network structure will be changed and step 4-2) will be repeated until the conditions are met.

[0060] Further, the step 5) includes the following steps: repeating step 2) and step 3) to filter the parameters generated in the process As the network input, the compensation value ΔSOC is output, and the compensation value is used to correct the rough estimate. The specific steps are as follows:

[0061] (1) Model parameter identification using recursive Bayesian algorithm

[0062] (2) Use adaptive extended Kalman filter to perform rough estimation of lithium battery SOC.

[0063] (3) Using the adaptive fuzzy neural network model determined by training in step 4), the filter parameters are provided as input data to the compensation network, and the compensation value ΔSOC is output.

[0064] (4) Compensate the filtering results of the adaptive extended Kalman filter and evaluate the performance.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. The present invention takes into account the inevitable errors generated by the extended Kalman filter in the Taylor expansion process, takes the filter parameters of the adaptive extended Kalman filter as input, and constructs a compensation network to compensate the prediction results.

[0067] 2. An error compensation network is constructed based on an adaptive fuzzy neural network, which is composed of an adaptive network and a fuzzy inference system. The adaptive fuzzy neural network combines the learning ability of the adaptive network and the interpretability of the fuzzy inference system, while facilitating the change of system parameters to obtain better output results. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0069] Figure 1 It is the second-order RC circuit equivalent model used in the present invention.

[0070] Figure 2 This is the basic structure of the adaptive fuzzy neural network provided by the present invention.

[0071] Figure 3 It is a flow chart of the SOC estimation method based on adaptive fuzzy neural network error compensation provided by the present invention.

[0072] Figure 4 In Example 1 provided by the present invention, under the FUDS working condition, the SOC is predicted by the SOC estimation method based on the adaptive fuzzy neural network error compensation, the SOC predicted by only using the AEKF algorithm, and the actual SOC comparison curve.

[0073] Figure 5 This is a graph showing the error comparison between the SOC estimation method based on the adaptive fuzzy neural network error compensation and the SOC prediction method using the adaptive extended Kalman filter alone under the FUDS condition in Example 1 provided by the present invention.

[0074] Figure 6 In Example 2 provided by the present invention, under the DST condition, the SOC is predicted by the SOC estimation method based on the adaptive fuzzy neural network error compensation, the SOC predicted by only using the AEKF algorithm, and the actual SOC comparison curve.

[0075] Figure 7This is a graph showing the error comparison between the SOC estimation method based on the adaptive fuzzy neural network error compensation and the SOC prediction method using the adaptive extended Kalman filter alone under the DST condition in Example 2 provided by the present invention. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0077] Example 1

[0078] The research object of this embodiment is Panasonic lithium-ion battery NCR-18650B, with a rated voltage of 3.7V and a capacity of 3400mAh. First, the battery is charged in a constant current charging mode (0.5C) until the cut-off voltage is reached, and the battery is allowed to stand for a period of time to reach a fully charged state. In Example 1, the lithium battery operates and collects data under the FUDS condition. The FUDS (Federal Urban Driving Schedule) condition is a test method for simulating battery performance under urban driving conditions, and is often used to evaluate the performance of lithium batteries during actual driving. The FUDS condition is widely used in lithium battery testing because it can more realistically reflect the use of the battery in actual urban driving, which is particularly important for battery performance evaluation of electric vehicles and hybrid vehicles.

[0079] See also Figures 1 to 5 This embodiment provides a SOC estimation method based on adaptive fuzzy neural network error compensation for a lithium battery working under FUDS conditions, comprising the following steps:

[0080] Step 1) performing an intermittent constant current discharge experiment on the lithium-ion battery, measuring its terminal voltage and load current data, and determining the functional relationship of OCV-SOC by polynomial fitting, so as to calculate the open circuit voltage of the lithium battery at each moment;

[0081] Step 2) establishing a second-order RC equivalent circuit model, and using a recursive Bayesian algorithm to perform parameter identification on the established model;

[0082] Step 3) using an adaptive extended Kalman filter to perform a rough prediction of the lithium battery SOC;

[0083] Step 4) an error compensation network based on an adaptive fuzzy neural network is proposed, the noise covariance matrix and Kalman gain generated in the adaptive extended Kalman filter estimation process are used as the input of the trained error compensation network, and the prediction error of SOC is output;

[0084] Step 5) Use the prediction error obtained by the adaptive fuzzy neural network to compensate the rough prediction result.

[0085] As a SOC estimation method based on adaptive fuzzy neural network error compensation of the present invention, the specific steps of step 2) are as follows:

[0086] Establish a second-order RC equivalent circuit model. This model uses R 1 and C 1 The parallel structure represents the electrochemical polarization effect of the battery, using R 2 and C 2 The parallel structure represents the concentration difference polarization effect of the battery. The open circuit voltage and terminal voltage are U oc and U. U 1 , U 2 Represents the voltage of two RC parallel links, R omc Indicates the ohmic internal resistance of the battery;

[0087] The SOC of a lithium-ion battery is defined as the ratio of its actual remaining capacity to its rated capacity, expressed as:

[0088]

[0089] Among them, Q n is the rated capacity of the battery, t 0 is the initial time, t 1 is the current moment, and I is the battery load current.

[0090] Based on Kirchhoff's law, the transfer function of the system can be obtained as:

[0091]

[0092] Mapping the above formula from the s plane to the z plane, we can get:

[0093]

[0094] Where T is the sampling period, τ 1 =R 1 C 1 , τ 2 =R 2 C 2 , a=R omc , b = τ 1 τ 2 , c = τ 1 +τ 2 , d = R omc +R 1 +R 2 , e=R omc (τ 1 +τ2 )+R 1 τ 1 +R 2 τ 2 .

[0095] The difference equation obtained after discretization of the transfer function is:

[0096]

[0097] Let y(t) = U(t) - U oc (t), the autoregressive (ARX) model with external input that is suitable for parameter identification of lithium-ion batteries can be obtained as follows:

[0098]

[0099] in, θ=[a 1 ,a 2 ,b 0 ,b 1 ,b 2 ] T Through the recursive Bayesian algorithm, the elements in θ can be identified. According to the functional relationship between a, b, c, d, e and the elements in θ, a, b, c, d, e can be derived, and finally the model parameter R can be obtained. omc , R 1 , R 2 , C 1 , C 2 .

[0100] The main formula of the recursive Bayesian algorithm is:

[0101]

[0102] The steps of implementing the recursive Bayesian identification algorithm to identify the battery system parameters are as follows:

[0103] (1) Initialization parameters: set the sampling period T, P(0) and variance σ v Initial value;

[0104] (2) SOC(k) is calculated using formula (1), and the open circuit voltage U is calculated based on the functional relationship between OCV and SOC. oc (k);

[0105] (3) According to formula (5), we get the information vector and output vector y(k);

[0106] (4) According to formula (7), the intermediate vector L(k) at time k is calculated in sequence, and the parameter estimation Covariance matrix P(k);

[0107] (5) Separate the parameters and obtain the parameters of the equivalent circuit model;

[0108] Repeat steps (2) to (5) until the end time is reached and then stop the identification.

[0109] Specifically, the step 3) comprises the following steps:

[0110] According to the battery SOC defined in step 2), its discrete form can be defined as:

[0111]

[0112] Where ζ represents the charge and discharge efficiency of the battery. The state equation and measurement equation of the second-order RC equivalent circuit model are:

[0113]

[0114] Where X k and k are the state vector and measurement vector at time k respectively. k is the system input at time k. f(·) and h(·) are nonlinear functions. w k and v k are process noise and measurement noise, respectively, both are white Gaussian noise. Q and R are the corresponding process noise covariance matrix and measurement noise covariance matrix.

[0115] Establish the state space equation for lithium battery SOC estimation:

[0116]

[0117] The prediction process and update process of the adaptive extended Kalman filter can be expressed as:

[0118]

[0119] Based on the traditional Kalman filter, the adaptive Kalman filter can correct the noise matrix in real time:

[0120]

[0121] Where L is the window size for covariance matching.

[0122] The adaptive extended Kalman filter is used to update the state matrix to achieve a rough estimate of the lithium battery SOC and obtain the first diagonal element of the filter parameter covariance matrix at time k. First element

[0123] Specifically, the step 4) comprises the following steps:

[0124] Step 4-1) In order to obtain the linearization loss in the filtering process from the training data through the compensation network, the coarse filtering result SOC 0 The difference between the SOC obtained by the ampere-hour integration method is defined as ΔSOC. In the SOC estimation process of the adaptive extended Kalman filter, the error of the nonlinear system is mainly reflected in the prediction process of the state covariance, and the resulting error is further introduced into the SOC estimation through the residual and gain of the system. Therefore, the first element of the diagonal element of the covariance matrix Kalman filter gain K k11 as network input.

[0125] Network input:

[0126] Network output: Outout j ={[ΔSOC] 1j ,…,[ΔSOC] nj}

[0127] Step 4-2) The training process is:

[0128] (7) Input parameters Perform fuzzy processing.

[0129] (8) The membership degree of each variable is multiplied to obtain the trigger strength of each rule.

[0130] (9) Normalize the penalty intensity of each layer.

[0131] (10) The normalized trigger intensity is linearly output.

[0132] (11) Perform weighted summation and defuzzification on the results of each rule.

[0133] (12) The SOC compensation is obtained by using the output of the adaptive fuzzy neural network, the error is calculated and integrated into the result, and if the conditions are met, the trained network structure is saved. Otherwise, the network structure is changed and the above steps are repeated until the conditions are met.

[0134] Specifically, the step 5) includes the following steps: repeating the steps 2) and 3) to obtain the parameters generated in the filtering process. As the network input, the compensation value ΔSOC is output, and the compensation value is used to correct the rough estimate. The specific steps are as follows:

[0135] (1) Model parameter identification using recursive Bayesian algorithm

[0136] (2) Use adaptive extended Kalman filter to perform rough estimation of lithium battery SOC.

[0137] (3) Using the adaptive fuzzy neural network model determined by training in step 4), the filter parameters are provided as input data to the compensation network, and the compensation value ΔSOC is output.

[0138] (4) Compensate the filtering results of the adaptive extended Kalman filter and evaluate the performance.

[0139] The flowchart of the SOC estimation method based on adaptive fuzzy neural network error compensation of the present invention is as follows: Figure 3 The present invention uses a hierarchical high-efficiency lithium battery SOC determination framework based on an adaptive fuzzy neural network with an adaptive error compensation module to predict the SOC result and error as shown in FIG. Figure 4 , Figure 5 As shown above, it can be seen that the adaptive extended Kalman filter combined with the compensation network based on adaptive fuzzy neural network effectively reduces the battery SOC prediction error. Compared with other methods, the prediction accuracy is higher and has certain engineering value.

[0140] Example 2

[0141] The research object of this embodiment is Panasonic lithium-ion battery NCR-18650B, with a rated voltage of 3.7V and a capacity of 3400mAh. First, the battery is charged in a constant current charging mode (0.5C) until the cut-off voltage is reached, and the battery is allowed to stand for a period of time to reach a fully charged state. In Example 2, the lithium battery is operated under DST conditions and data is collected. The full name of DST is Dynamic Stress Test, which is a dynamic stress test method used to evaluate the comprehensive performance of batteries under real conditions of use. It simulates various load changes that may occur in the vehicle during driving, such as acceleration, deceleration and idling conditions.

[0142] See also Figures 1 to 3 , Figure 6 to Figure 7 This embodiment provides a SOC estimation method based on adaptive fuzzy neural network error compensation when a lithium battery works under DST conditions, comprising the following steps:

[0143] Step 1) performing an intermittent constant current discharge experiment on the lithium-ion battery, measuring its terminal voltage and load current data, and determining the functional relationship of OCV-SOC by polynomial fitting, so as to calculate the open circuit voltage of the lithium battery at each moment;

[0144] Step 2) establishing a second-order RC equivalent circuit model, and using a recursive Bayesian algorithm to perform parameter identification on the established model;

[0145] Step 3) using an adaptive extended Kalman filter to perform a rough prediction of the lithium battery SOC;

[0146] Step 4) an error compensation network based on an adaptive fuzzy neural network is proposed, the noise covariance matrix and Kalman gain generated in the adaptive extended Kalman filter estimation process are used as the input of the trained error compensation network, and the prediction error of SOC is output;

[0147] Step 5) Use the prediction error obtained by the adaptive fuzzy neural network to compensate the rough prediction result.

[0148] As a SOC estimation method based on adaptive fuzzy neural network error compensation of the present invention, the specific steps of step 2) are as follows:

[0149] A second-order RC equivalent circuit model is established, and the recursive Bayesian algorithm is used to identify the model parameters.

[0150] The specific steps of parameter identification of the second-order RC equivalent circuit model are as follows:

[0151] (1) Initialization parameters: set the sampling period T, P(0) and variance σ v Initial value;

[0152] (2) SOC(k) is calculated using formula (1), and the open circuit voltage U is calculated based on the functional relationship between OCV and SOC. oc (k);

[0153] (3) According to formula (5), we get the information vector and output vector y(k);

[0154] (4) According to formula (7), the intermediate vector L(k) at time k is calculated in sequence, and the parameter estimation Covariance matrix P(k);

[0155] (5) Separate the parameters and obtain the parameters of the equivalent circuit model;

[0156] Repeat steps (2) to (5) until the end time is reached and then stop the identification.

[0157] Specifically, the step 3) comprises the following steps:

[0158] The adaptive extended Kalman filter is used to update the state matrix to achieve a rough estimate of the lithium battery SOC and obtain the first diagonal element of the filter parameter covariance matrix at time k. The second element

[0159] Specifically, the step 4) comprises the following steps:

[0160] Step 4-1) In order to obtain the linearization loss in the filtering process from the training data through the compensation network, the coarse filtering result SOC 0 The difference between the SOC obtained by the ampere-hour integration method is defined as ΔSOC. In the SOC estimation process of the adaptive extended Kalman filter, the error of the nonlinear system is mainly reflected in the prediction process of the state covariance, and the resulting error is further introduced into the SOC estimation through the residual and gain of the system. Therefore, the first element of the diagonal element of the covariance matrix First element as network input.

[0161] Network input:

[0162] Network output: Outout j ={[ΔSOC] 1j ,…,[ΔSOC] nj}

[0163] Step 4-2) The training process is:

[0164] (1) Input parameters Perform fuzzy processing.

[0165] (2) The membership degree of each variable is multiplied to obtain the trigger strength of each rule.

[0166] (3) Normalize the penalty intensity of each layer.

[0167] (4) Make a linear output of the normalized trigger intensity.

[0168] (5) Perform weighted summation and defuzzification on the results of each rule.

[0169] (6) The SOC compensation is obtained by using the output of the adaptive fuzzy neural network, the error is calculated and integrated into the result, and if the conditions are met, the trained network structure is saved. Otherwise, the network structure is changed and the above steps are repeated until the conditions are met.

[0170] Specifically, the step 5) includes the following steps: repeating the steps 2) and 3) to obtain the parameters generated in the filtering process. As the network input, the compensation value ΔSOC is output, and the compensation value is used to correct the rough estimate. The specific steps are as follows:

[0171] (1) Model parameter identification using recursive Bayesian algorithm

[0172] (2) Use adaptive extended Kalman filter to perform rough estimation of lithium battery SOC.

[0173] (3) Using the adaptive fuzzy neural network model determined by training in step 4), the filter parameters are provided as input data to the compensation network, and the compensation value ΔSOC is output.

[0174] (4) Compensate the filtering results of the adaptive extended Kalman filter and evaluate the performance.

[0175] Under DST conditions, the present invention uses the lithium battery SOC estimation results and errors based on the adaptive fuzzy neural network with an adaptive error compensation module. Figure 6 , Figure 7 As shown above, it can be seen that the adaptive extended Kalman filter combined with the compensation network based on adaptive fuzzy neural network effectively reduces the battery SOC prediction error. Compared with other methods, the prediction accuracy is higher and has certain engineering value.

[0176] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A SOC estimation method based on adaptive fuzzy neural network error compensation, characterized in that: The following steps are involved: Step 1) performing an intermittent constant current discharge experiment on the lithium-ion battery, measuring its terminal voltage and load current data, and determining the functional relationship of OCV-SOC by polynomial fitting, so as to calculate the open circuit voltage of the lithium battery at each moment; Step 2) establishing a second-order RC equivalent circuit model, and using a recursive Bayesian algorithm to perform parameter identification on the established model; Step 3) using an adaptive extended Kalman filter to perform a rough prediction of the lithium battery SOC; Step 4) an error compensation network based on an adaptive fuzzy neural network is proposed, the noise covariance matrix and Kalman gain generated in the adaptive extended Kalman filter estimation process are used as the input of the trained error compensation network, and the prediction error of SOC is output; Step 5) Use the prediction error obtained by the adaptive fuzzy neural network to compensate the rough prediction result.

2. The SOC estimation method based on adaptive fuzzy neural network error compensation according to claim 1, characterized in that: The step 2) comprises the following steps: A second-order RC equivalent circuit model is established. The model represents the electrochemical polarization effect of the battery through the parallel structure of R1 and C1, and the concentration difference polarization effect of the battery through the parallel structure of R2 and C2. The open circuit voltage and terminal voltage are U oc and U; U1, U2 represent the voltage of two RC parallel links, R omc Indicates the ohmic internal resistance of the battery; The SOC of a lithium-ion battery is defined as the ratio of its actual remaining capacity to its rated capacity, expressed as: Among them, Q n is the rated capacity of the battery, t0 is the initial time, t1 is the current time, and I is the battery load current; Based on Kirchhoff's law, the transfer function of the system is: Mapping the above formula from the s plane to the z plane, we get: Where T is the sampling period, τ1=R1C1, τ2=R2C2, a=R omc , b=τ1τ2, c=τ1+τ2, d=R omc +R1+R2, e=R omc (τ1+τ2)+R1τ1+R2τ2; The difference equation obtained after discretization of the transfer function is: Let y(t) = U(t) - U oc (t), the autoregressive ARX model with external input that is suitable for parameter identification of lithium-ion batteries is obtained as follows: in, θ=[a1,a2,b0,b1,b2] T , the elements in θ are identified by recursive Bayesian algorithm, and a, b, c, d, e are derived according to the functional relationship between a, b, c, d, e and the elements in θ, and finally the model parameter R is obtained omc , R1, R2, C1, C2; The recursive Bayesian algorithm formula is: The steps of implementing the recursive Bayesian identification algorithm to identify the battery system parameters are as follows: (1) Initialization parameters: set the sampling period T, P(0) and variance σ v Initial value; (2) SOC(k) is calculated using formula (1), and the open circuit voltage U is calculated based on the functional relationship between OCV and SOC. oc (k); (3) According to formula (5), we get the information vector and output vector y(k); (4) According to formula (7), the intermediate vector L(k) at time k is calculated in sequence, and the parameter estimation Covariance matrix P(k); (5) Separate the parameters and obtain the parameters of the equivalent circuit model; Repeat steps (2) to (5) until the end time is reached and then stop the identification.

3. The SOC estimation method based on adaptive fuzzy neural network error compensation according to claim 1, characterized in that: The step 3) comprises the following steps: According to the battery SOC defined in step 2), its discrete form is defined as: Where ζ represents the charge and discharge efficiency of the battery, and the state equation and measurement equation of the second-order RC equivalent circuit model are: Where X k and k are the state vector and measurement vector at time k, u k is the system input at time k, f(·) and h(·) are nonlinear functions, and w k and v k are process noise and measurement noise, respectively, both are white Gaussian noise, Q and R are the corresponding process noise covariance matrix and measurement noise covariance matrix; Establish the state space equation for lithium battery SOC estimation: The prediction process and update process of the adaptive extended Kalman filter are expressed as: Based on the traditional Kalman filter, the adaptive Kalman filter corrects the noise matrix in real time: Where L is the window size for covariance matching; The adaptive extended Kalman filter is used to update the state matrix to achieve a rough estimate of the lithium battery SOC and obtain the first diagonal element of the filter parameter covariance matrix at time k. The first element K of the Kalman filter gain matrix k1 .

4. The SOC estimation method based on adaptive fuzzy neural network error compensation according to claim 1, characterized in that: The step 4) comprises the following steps: Step 4-1) In order to obtain the linearization loss in the filtering process from the training data through the compensation network, the difference between the coarse filtering result SOC0 and the SOC obtained by the ampere-hour integration method is defined as ΔSOC. In the SOC estimation process of the adaptive extended Kalman filter, the error of the nonlinear system is mainly reflected in the prediction process of the state covariance. The resulting error is further introduced into the SOC estimation through the residual and gain of the system. Therefore, the first element of the diagonal element of the covariance matrix The second element As network input; Network input: Network output: Outout j ={[ΔSOC] 1j ,…,[ΔSOC] nj } Step 4-2) The training process is: (1) Input parameters Perform fuzzy processing; (2) The membership degree of each variable is multiplied to obtain the trigger strength of each rule; (3) Normalize the penalty intensity of each layer; (4) linearly output the normalized trigger intensity; (5) Perform weighted summation and defuzzification on the results of each rule; (6) Use the output of the adaptive fuzzy neural network to obtain SOC compensation, calculate the error and integrate it into the result. If the conditions are met, the trained network structure will be saved, otherwise the network structure will be changed and step 4-2) will be repeated until the conditions are met.

5. The SOC estimation method based on adaptive fuzzy neural network error compensation according to claim 1, characterized in that: The step 5) includes the following steps: repeating the steps 2) and 3) with the parameters generated in the filtering process As the network input, the compensation value ΔSOC is output, and the rough estimate is corrected using the compensation value. The specific steps are as follows: (1) Use the recursive Bayesian algorithm to identify model parameters; (2) Using adaptive extended Kalman filter to roughly estimate the SOC of lithium battery; (3) using the adaptive fuzzy neural network model determined by training in step 4), providing filter parameters as input data to the compensation network, and outputting a compensation value ΔSOC; (4) Compensate the filtering results of the adaptive extended Kalman filter and evaluate the performance.

Images