CN116203432B - Method for predicting battery state of charge based on CSO optimized unscented Kalman filter - Google Patents

Abstract

The invention discloses a CSO optimization-based unscented Kalman filtering method for predicting the state of charge of a battery, which comprises the following steps: s1: calculating relevant parameters affecting the prediction accuracy when predicting the state of charge (SOC) of the battery by adopting Unscented Kalman Filtering (UKF) algorithm; s2: testing the predicted battery to obtain UUUDS data of the urban road circulation condition; s3: adopting a criss-cross algorithm CSO to optimize an unscented Kalman filter UKF algorithm; s4: the optimized unscented Kalman filter UKF algorithm is applied to the prediction of the state of charge (SOC) of the battery to obtain a more accurate battery working state, so that the service life and the service efficiency of the battery are improved. The invention utilizes the advantages of fast convergence speed and strong global searching capability of the crisscross algorithm, can rapidly and accurately obtain the optimal UKF fitting parameters under different battery working condition data, avoids the parameters from sinking into local optimum, ensures the prediction precision of the battery state of charge to be as high as possible, and realizes the accurate tracking of the battery working state.

Description

Method for predicting battery state of charge based on CSO optimized unscented Kalman filter

Technical field

The present invention relates to the technical field of electric vehicle battery status monitoring, and in particular to a method for predicting battery state of charge based on CSO-optimized unscented Kalman filtering.

Background technique

In order to achieve accurate prediction of battery state of charge (State of Charge, SOC), domestic and foreign scholars have proposed a prediction scheme based on the Kalman filter algorithm (Kalman filter, KF). Since the battery's state of charge is a nonlinear system, it is necessary to use unscented transformation to convert the nonlinear system into a linear system and then perform Kalman filtering, that is, use the Unscented Kalman Filter algorithm (Unscented Kalman Filter, UKF). The unscented Kalman filter algorithm requires parameter tuning and optimization before it can make predictions. Traditional optimization algorithms include genetic algorithms, particle swarm algorithms, etc. However, when solving complex optimization problems, these algorithms have slow convergence speeds and are prone to falling into local optima, which has limitations. This further improves the accuracy of battery state-of-charge prediction. To this end, the present invention proposes a prediction method based on CrisscrossOptimized Unscented Kalman Filter (CUKF) optimized by the cross-cross algorithm.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the existing technology and propose a method for predicting the battery state of charge based on CSO-optimized unscented Kalman filtering. By introducing the idea of full cross-over of inter-individual variables using the Crisscross Optimization Algorithm (CSO), the parameter setting method of the unscented Kalman filter UKF algorithm is optimized to achieve accurate prediction of battery state of charge and accurate tracking of battery working status.

In order to achieve the above objects, the technical solutions provided by the present invention are:

The method of predicting battery state of charge based on CSO optimized unscented Kalman filter is characterized by including the following steps:

S1: When using the unscented Kalman filter UKF algorithm to predict the battery state of charge SOC, calculate the relevant parameters that affect the prediction accuracy;

S2: Test the predicted battery and obtain its urban road cycle condition UUDS data;

S3: Use the vertical and horizontal crossover algorithm CSO to optimize the unscented Kalman filter UKF algorithm;

S4: Apply the optimized unscented Kalman filter UKF algorithm to predict the battery state of charge SOC to obtain a more accurate battery working state, thereby improving battery life and efficiency;

In step S1, the specific steps for calculating relevant parameters that affect prediction accuracy are as follows:

S1-1: In the unscented transformation UT parameter initialization stage of the unscented Kalman filter UKF algorithm, it is necessary to carry out the symmetry parameter λ of the Sigma point set, the approximate mean Sigma point weight matrix W _m and the approximate covariance Sigma point weight matrix W _c The calculations are:

(1) In the formula, i is the dimension of the unscented transformation, α is the sampling parameter of the Sigma point, β is a characteristic parameter of the Gaussian distribution, L is the dimension of the state vector, and k is the characteristic parameter that controls the distance of the Sigma point from the mean. ;

S1-2: In the prediction stage of the unscented Kalman filter UKF algorithm, the filter uses the posterior estimate of the previous state to calculate the prior estimate of the current state. In this process, the prior estimate covariance needs to be calculated.

(2) In the formula, is the Sigma point sampling result of the prior probability distribution of i dimension at time t,/> is the a priori estimated mean at time t, T is the inverse matrix symbol, and Q is the covariance matrix of the process noise;

S1-3: In the update stage of the unscented Kalman filter UKF algorithm, the filter uses the observation value of the current state to optimize the prediction value obtained in the prediction stage to obtain a more accurate new estimate value. In this process, calculation is required Covariance of observation z _t at time t

(3) In the formula, is the i-dimensional observation vector at time t, μ _zt is the mean value of observation z _t at time t, and R is the covariance matrix of measurement noise;

S1-4: Perform optimization parameter analysis of the unscented Kalman filter UKF algorithm:

According to the unscented Kalman filter UKF algorithm, there are the following parameters that need to be determined before the algorithm starts running:

The covariance matrix Q of the process noise is used to estimate the error and uncertainty of the system model, which is usually a diagonal matrix. In the process of predicting the battery state of charge SOC by the unscented Kalman filter UKF algorithm, the covariance matrix of the process noise Q is:

Then in equation (4), there are three parameters that need to be tuned, namely the covariances Q ₁ , Q ₂ , and Q ₃ of the process noise;

The covariance matrix R of the measurement noise is used to estimate the error and uncertainty of the measurement model. It is usually a diagonal matrix. In the process of predicting the battery state of charge SOC by the unscented Kalman filter UKF algorithm, the covariance matrix of the measurement noise R is:

R＝[R ₁ ] (5)

Then in equation (5), there is one parameter that needs to be tuned, which is the covariance R ₁ of the measurement noise;

The sampling parameter α is used to control the distribution density of Sigma points. Appropriate values should be selected to ensure sufficient coverage of UT points. The value range of the sampling parameter α is between 0 and 1;

The step S2 tests the predicted battery and obtains its UUDS data of urban road cycle conditions. The specific steps are as follows:

S2-1: Determine the rated capacity and rated voltage of the battery to be measured, and determine the parameters of the test device, including load resistance and load reactance;

S2-2: Charge the battery to be measured to 100% capacity and leave it alone for 1-2 hours to stabilize the battery state;

S2-3: Remove the battery from the charger and let it sit at room temperature for at least 30 minutes so that the internal temperature of the battery reaches room temperature;

S2-4: Place the battery on the UUDS test device under urban road cycle conditions and ensure that both ends of the battery are in correct contact with the electrodes of the test device;

S2-5: Set test conditions according to the requirements of UUDS under urban road cycle conditions, including current range, voltage range, temperature and time, to ensure that the test conditions meet the battery usage conditions;

S2-6: Start the test and record the test data, including current, voltage and time, until the battery power drops to a certain level or the test time reaches the preset value;

S2-7: Analyze the test data, evaluate the validity of the data, and obtain the current data and voltage data for the calculation of step S3;

The step S3 uses the vertical and horizontal crossover algorithm CSO to optimize the unscented Kalman filter UKF algorithm. The specific steps are as follows:

S3-1: The upper and lower limits of the five parameters in the initialization step S1, including the covariance Q ₁ , Q ₂ , Q ₃ of the process noise, the covariance R ₁ of the measurement noise and the sampling parameter α, randomly generate N within this interval Data pairs are called populations;

S3-2: Use the N data pairs generated in step S3-1 to perform the first population fitness calculation. The greater the population fitness, the smaller the error in the fitting process of the group of parameters. Specific calculation of population fitness. Proceed as follows:

S3-2-1: Obtain the observation matrix Z according to the test conditions determined in step S2-1;

S3-2-2: Initialize the unscented transformation UT parameters based on the input parameter data pairs to be tuned;

S3-2-3: Add a certain amount of normal noise to the UUDS current data of urban road cycle conditions obtained in step S2 to simulate the actual working environment;

S3-2-4: Generate Sigma points based on the noise-added current data obtained in step S3-2-3;

S3-2-5: Predict the state of Kalman filter KF based on the Sigma points obtained in step S3-2-4;

S3-2-6: Perform Sigma point reconstruction based on the state prediction results obtained in step S3-2-5;

S3-2-7: Based on the voltage data of the urban road cycle condition UUDS obtained in step S2 and the reconstructed point of state prediction obtained in step S3-2-6, perform the observation update of the Kalman filter KF;

S3-2-8: Repeat steps S3-2-4 to S3-2-7 until the predicted value of the battery state of charge for the entire test time period is obtained;

S3-2-9: Use the ampere-hour integration method to calculate the battery state of charge and use it as a true reference value for the battery state of charge;

S3-2-10: Calculate the mean error based on the prediction results of S3-2-8 and the results of the ampere-hour integration method of S3-2-9 as the standard;

S3-2-11: Repeat steps S3-2-4 to S3-2-10 5 times to eliminate occasional over- or under-predictions caused by current noise, and average the errors of the 5 times. Take the reciprocal as the population fitness of the population, and output it to step S3-3 for use;

S3-3: Perform the vertical crossover operation of the vertical and horizontal crossover method CSO. The vertical crossover process is an arithmetic operation that crosses the optimization parameters of different dimensions. Its expression is:

M _vc (m,d ₁ )＝r×X(m,d ₁ )+(1-r)×X(m,d ₂ )+c(X(m,d ₁ )-X(m,d ₂ ) ) (6)

In the formula (6), r and c are random numbers between 0 and 1, X(m,d ₁ ) and X(m,d ₂ ) are parent optimization parameters of different dimensions, M _vc (m, d ₁ ) is the optimization parameter of the offspring generated by vertical cross of different dimensions of the parent, the parameter m=1,2,...,N, N is the population size; d ₁ ,d ₂ =1,2,...,C , C is the number of dimensions of the population;

S3-4: Calculate the population fitness of the offspring generated in step S3-3, compare it with the population fitness of the parent, and retain the population with greater population fitness so that the overall population evolves in a better direction;

S3-5: Perform the horizontal crossover operation of the vertical and horizontal crossover method CSO. The horizontal crossover process is an arithmetic operation that crosses all the optimization parameters of different solutions. Its expression is:

In the formula (7), the parameters n=1,2,…,N, N is the population size;

S3-6: Calculate the population fitness of the offspring generated in step S3-5, compare it with the population fitness of the parent, and retain the population with greater population fitness, so that the overall population evolves in a better direction;

S3-7: Use the following steps to regenerate the population:

S3-7-1: If the unscented Kalman filter UKF algorithm prediction process terminates, the population fitness of the population will be marked as 0;

S3-7-2: Before the end of each iteration, regenerate the population with a statistical population fitness of 0;

S3-7-3: Obtain each dimension i, and respectively obtain the maximum value max _i and minimum value min _i of the population whose fitness is not 0 in this dimension;

S3-7-4: When a new population is generated, the parameters of each dimension i are regenerated from (1+k _s )max _i to (1-k _s )min _i respectively, and k _s is the regeneration diffusion coefficient, so that This population is not easy to converge, causing the population fitness to fall into a local optimum;

S3-8: Repeat steps S3-3 to S3-7 until the number of iterations reaches the maximum number of iterations set by the system;

S3-9: Obtain the population with the best population fitness at this time as the optimal population, and use it as the model application parameter in step S4;

The step S4 applies the optimized unscented Kalman filter UKF algorithm to predict the battery state of charge SOC to obtain a more accurate battery working state, thereby improving battery life and usage efficiency.

It can be seen from the above technical solutions that the embodiments of the present invention have the following beneficial effects:

1. Optimize battery performance: Accurate battery state-of-charge SOC prediction can help optimize battery use and maintenance, thereby extending battery life and performance;

2. Improve battery efficiency: By accurately predicting the battery's state of charge SOC, the charging and discharging process can be better managed to improve the battery's energy utilization and efficiency;

3. Increase driving range: For electric vehicles, accurate battery state-of-charge SOC prediction can help determine the remaining driving range, thereby increasing driving safety and reliability;

4. Avoid battery overcharging or under-discharging: Accurately predicting battery state-of-charge (SOC) can also help avoid battery overcharging or under-discharging situations that may harm battery performance or cause safety issues.

Description of the drawings

Figure 1 is a flow chart of the method of predicting battery state of charge based on CSO optimized unscented Kalman filter according to the present invention;

Figure 2 shows the errors at each moment of prediction by the three methods of artificial experience tuning, particle swarm algorithm PSO tuning and cross-horizontal algorithm CSO tuning;

Figure 3 shows the curves of battery state-of-charge SOC prediction results using three methods: artificial experience tuning, particle swarm algorithm PSO tuning and cross-over algorithm CSO tuning.

Figure 4 shows the convergence speed of particle swarm algorithm PSO tuning and cross-horizontal algorithm CSO tuning.

Detailed ways

The present invention will be further described below in conjunction with specific examples:

As shown in Figure 1, the method of predicting the battery state of charge based on CSO optimized unscented Kalman filter described in this embodiment includes the following steps:

S1: When using the unscented Kalman filter UKF algorithm to predict the battery state of charge SOC, calculate the relevant parameters that affect the prediction accuracy;

S2: Test the predicted battery and obtain its urban road cycle condition UUDS data;

S3: Use the vertical and horizontal crossover algorithm CSO to optimize the unscented Kalman filter UKF algorithm;

S4: Apply the optimized unscented Kalman filter UKF algorithm to predict the battery state of charge SOC to obtain a more accurate battery working state, thereby improving battery life and efficiency;

In step S1, the specific steps for calculating relevant parameters that affect prediction accuracy are as follows:

S1-1: In the unscented transformation UT parameter initialization stage of the unscented Kalman filter UKF algorithm, it is necessary to carry out the symmetry parameter λ of the Sigma point set, the approximate mean Sigma point weight matrix W _m and the approximate covariance Sigma point weight matrix W _c The calculations are:

(1) In the formula, i is the dimension of the unscented transformation, α is the sampling parameter of the Sigma point, β is a characteristic parameter of the Gaussian distribution, L is the dimension of the state vector, and k is the characteristic parameter that controls the distance of the Sigma point from the mean. ;

S1-2: In the prediction stage of the unscented Kalman filter UKF algorithm, the filter uses the posterior estimate of the previous state to calculate the prior estimate of the current state. In this process, the prior estimate covariance needs to be calculated.

(2) In the formula, is the Sigma point sampling result of the prior probability distribution of i dimension at time t,/> is the a priori estimated mean at time t, T is the inverse matrix symbol, and Q is the covariance matrix of the process noise;

S1-3: In the update stage of the unscented Kalman filter UKF algorithm, the filter uses the observation value of the current state to optimize the prediction value obtained in the prediction stage to obtain a more accurate new estimate value. In this process, calculation is required Covariance P _zt of observation quantity z _t at time t:

(3) In the formula, is the observation vector of dimension i at time t,/> is the mean value of the observation quantity z _t at time t, and R is the covariance matrix of the measurement noise;

S1-4: Perform optimization parameter analysis of the unscented Kalman filter UKF algorithm:

According to the unscented Kalman filter UKF algorithm, there are the following parameters that need to be determined before the algorithm starts running:

The covariance matrix Q of the process noise is used to estimate the error and uncertainty of the system model, which is usually a diagonal matrix. In the process of predicting the battery state of charge SOC by the unscented Kalman filter UKF algorithm, the covariance matrix of the process noise Q is:

Then in equation (4), there are three parameters that need to be tuned, namely the covariances Q ₁ , Q ₂ , and Q ₃ of the process noise;

The covariance matrix R of the measurement noise is used to estimate the error and uncertainty of the measurement model. It is usually a diagonal matrix. In the process of predicting the battery state of charge SOC by the unscented Kalman filter UKF algorithm, the covariance matrix of the measurement noise R is:

R＝[R ₁ ] (5)

Then in equation (5), there is one parameter that needs to be tuned, which is the covariance R ₁ of the measurement noise;

The sampling parameter α is used to control the distribution density of Sigma points. Appropriate values should be selected to ensure sufficient coverage of UT points. The value range of the sampling parameter α is between 0 and 1;

The step S2 tests the predicted battery and obtains its UUDS data of urban road cycle conditions. The specific steps are as follows:

S2-1: Determine the rated capacity and rated voltage of the battery to be measured, and determine the parameters of the test device, including load resistance and load reactance;

S2-2: Charge the battery to be measured to 100% capacity and leave it alone for 1-2 hours to stabilize the battery state;

S2-3: Remove the battery from the charger and let it sit at room temperature for at least 30 minutes so that the internal temperature of the battery reaches room temperature;

S2-4: Place the battery on the UUDS test device under urban road cycle conditions and ensure that both ends of the battery are in correct contact with the electrodes of the test device;

S2-5: Set test conditions according to the requirements of UUDS under urban road cycle conditions, including current range, voltage range, temperature and time, to ensure that the test conditions meet the battery usage conditions;

S2-6: Start the test and record the test data, including current, voltage and time, until the battery power drops to a certain level or the test time reaches the preset value;

S2-7: Analyze the test data, evaluate the validity of the data, and obtain the current data and voltage data for the calculation of step S3;

The step S3 uses the vertical and horizontal crossover algorithm CSO to optimize the unscented Kalman filter UKF algorithm. The specific steps are as follows:

S3-1: The upper and lower limits of the five parameters in the initialization step S1, including the covariance Q ₁ , Q ₂ , Q ₃ of the process noise, the covariance R ₁ of the measurement noise and the sampling parameter α, randomly generate N within this interval Data pairs are called populations;

S3-2: Use the N data pairs generated in step S3-1 to perform the first population fitness calculation. The greater the population fitness, the smaller the error in the fitting process of the group of parameters. Specific calculation of population fitness. Proceed as follows:

S3-2-1: Obtain the observation matrix Z according to the test conditions determined in step S2-1;

S3-2-2: Initialize the unscented transformation UT parameters based on the input parameter data pairs to be tuned;

S3-2-3: Add a certain amount of normal noise to the UUDS current data of urban road cycle conditions obtained in step S2 to simulate the actual working environment;

S3-2-4: Generate Sigma points based on the noise-added current data obtained in step S3-2-3;

S3-2-5: Predict the state of Kalman filter KF based on the Sigma points obtained in step S3-2-4;

S3-2-6: Perform Sigma point reconstruction based on the state prediction results obtained in step S3-2-5;

S3-2-7: Based on the voltage data of the urban road cycle condition UUDS obtained in step S2 and the reconstructed point of state prediction obtained in step S3-2-6, perform the observation update of the Kalman filter KF;

S3-2-8: Repeat steps S3-2-4 to S3-2-7 until the predicted value of the battery state of charge for the entire test time period is obtained;

S3-2-9: Use the ampere-hour integration method to calculate the battery state of charge and use it as a true reference value for the battery state of charge;

S3-2-10: Calculate the mean error based on the prediction results of S3-2-8 and the results of the ampere-hour integration method of S3-2-9 as the standard;

S3-2-11: Repeat steps S3-2-4 to S3-2-10 5 times to eliminate occasional over- or under-predictions caused by current noise, and average the errors of the 5 times. Take the reciprocal as the population fitness of the population, and output it to step S3-3 for use;

S3-3: Perform the vertical crossover operation of the vertical and horizontal crossover method CSO. The vertical crossover process is an arithmetic operation that crosses the optimization parameters of different dimensions. Its expression is:

M _vc (m,d ₁ )＝r×X(m,d ₁ )+(1-r)×X(m,d ₂ )+c(X(m,d ₁ )-X(m,d ₂ ) ) (6)

In the formula (6), r and c are random numbers between 0 and 1, X(m,d ₁ ) and X(m,d ₂ ) are parent optimization parameters of different dimensions, M _vc (m, d ₁ ) is the optimization parameter of the offspring generated by vertical cross of different dimensions of the parent, the parameter m=1,2,...,N, N is the population size; d ₁ ,d ₂ =1,2,...,C , C is the number of dimensions of the population;

S3-4: Calculate the population fitness of the offspring generated in step S3-3, compare it with the population fitness of the parent, and retain the population with greater population fitness so that the overall population evolves in a better direction;

S3-5: Perform the horizontal crossover operation of the vertical and horizontal crossover method CSO. The horizontal crossover process is an arithmetic operation that crosses all the optimization parameters of different solutions. Its expression is:

In the formula (7), the parameters n=1,2,...,N, N is the population size;

S3-6: Calculate the population fitness of the offspring generated in step S3-5, compare it with the population fitness of the parent, and retain the population with greater population fitness, so that the overall population evolves in a better direction;

S3-7: Use the following steps to regenerate the population:

S3-7-1: If the unscented Kalman filter UKF algorithm prediction process terminates, the population fitness of the population will be marked as 0;

S3-7-2: Before the end of each iteration, regenerate the population with a statistical population fitness of 0;

S3-7-3: Obtain each dimension i, and respectively obtain the maximum value max _i and minimum value min _i of the population whose fitness is not 0 in this dimension;

S3-7-4: When a new population is generated, the parameters of each dimension i are regenerated from (1+k _s )max _i to (1-k _s )min _i respectively, and k _s is the regeneration diffusion coefficient, so that This population is not easy to converge, causing the population fitness to fall into a local optimum;

S3-8: Repeat steps S3-3 to S3-7 until the number of iterations reaches the maximum number of iterations set by the system;

S3-9: Obtain the population with the best population fitness at this time as the optimal population, and use it as the model application parameter in step S4;

The step S4 applies the optimized unscented Kalman filter UKF algorithm to predict the battery state of charge SOC to obtain a more accurate battery working state, thereby improving battery life and usage efficiency.

In order to verify the effectiveness of the unscented Kalman filter method for predicting the battery state of charge based on CSO optimization, the unscented Kalman filter was used to predict the battery state of charge in the Matlab environment. The parameters of the unscented Kalman filter were determined using artificial experience. tuning, particle swarm algorithm PSO tuning and cross-horizontal algorithm CSO tuning. At the same time, the ampere-hour integration method Ah is used to calculate the battery's state of charge as a true reference value for the battery's state of charge. PSO and CSO perform the same 200 iterations. The evaluation method is to use the same parameter group to repeat 5 predictions. The average error between the 5 prediction results and the true value is averaged as the error average of the parameter group. The smaller the error average. The better. The optimal parameter group and its corresponding error mean obtained by the three methods of artificial experience tuning, particle swarm algorithm PSO tuning and cross-horizontal algorithm CSO tuning are shown in Table 1. The errors of the three methods of artificial experience tuning, particle swarm algorithm PSO tuning and cross-horizontal algorithm CSO tuning at each moment of prediction are shown in Figure 2. The three methods of artificial experience tuning, particle swarm algorithm PSO tuning and cross-horizontal algorithm CSO tuning are substituted into the unscented Kalman filter to predict the battery state of charge SOC. The result curve is shown in Figure 3. The convergence speeds of particle swarm algorithm PSO tuning and vertical-horizontal crossover algorithm CSO tuning are shown in Figure 4.

Table 1 The optimal parameter groups of the three tuning methods and their corresponding error averages

Combining Table 1 and Figures 2 to 4, it can be seen that compared with the parameter set set by manual experience and the parameter set set by the particle swarm algorithm PSO, the parameter set set by the cross-horizontal algorithm CSO is used for unscented Kalman filtering to predict the battery state of charge, and the prediction is accurate. The degrees are increased by 26520 times and 7.63 times respectively. The convergence speed of the CSO algorithm is 91.7% higher than that of the PSO algorithm. This shows that the global convergence speed of the CSO algorithm is faster, and the unscented Kalman filter parameters can effectively escape from the local optimum, and the global search The superior ability is stronger.

The implementation examples described above are only preferred embodiments of the present invention and do not limit the scope of the present invention. Therefore, any changes made based on the shape and principle of the present invention should be covered by the protection scope of the present invention.

Claims (1)

1. The method for predicting the battery state of charge based on CSO optimization by unscented Kalman filtering is characterized by comprising the following steps:

step S1: calculating relevant parameters affecting the prediction accuracy when predicting the state of charge (SOC) of the battery by adopting Unscented Kalman Filtering (UKF) algorithm;

step S2: testing the predicted battery to obtain UUUDS data of the urban road circulation condition;

step S3: adopting a criss-cross algorithm CSO to optimize an unscented Kalman filter UKF algorithm;

step S4: the optimized unscented Kalman filter UKF algorithm is applied to the prediction of the battery state of charge (SOC) to obtain a more accurate battery working state, so that the service life and the service efficiency of the battery are improved;

in the step S1, the specific steps of calculating the relevant parameters affecting the prediction accuracy are as follows:

s1-1: in the unscented transformation UT parameter initialization stage of unscented Kalman filtering UKF algorithm, symmetry parameter lambda of Sigma point set is needed to be performed, and approximate mean Sigma point weight matrix W is approximated _m And approximate covariance Sigma Point weight matrix W _c The calculation of (1) is as follows:

(1) Wherein i is the dimension of unscented transformation, alpha is the sampling parameter of Sigma points, beta is one characteristic parameter of Gaussian distribution, L is the dimension of state vectors, and k is the characteristic parameter for controlling the distance from Sigma points to the mean value;

s1-2: in the prediction stage of unscented Kalman filtering UKF algorithm, the filter adopts the posterior estimation of the last state to calculate the prior estimation of the current state, and the prior estimation covariance needs to be calculated in the process

(2) In the method, in the process of the invention,sigma point sampling result for a priori probability distribution of dimension i at time t,>the prior estimated mean value at the moment T is represented by an inverse matrix symbol, and the covariance matrix of the process noise is represented by Q;

s1-3: in the updating stage of unscented Kalman filtering UKF algorithm, the filter optimizes the predicted value obtained in the predicting stage by using the observed value of the current state to obtain a more accurate new estimated value, and in the process, the observed value z at the moment t needs to be calculated _t Covariance of (2)

(3) In the method, in the process of the invention,for the observation vector of dimension i at time t, +.>For the observation of the quantity z at time t _t R is the covariance matrix of the measured noise;

s1-4: and (3) carrying out optimization parameter analysis of unscented Kalman filtering UKF algorithm:

according to unscented kalman filter UKF algorithm, there are the following parameters that need to be determined before the algorithm starts to run:

the covariance matrix Q of the process noise is used for estimating the error and uncertainty of the system model, and is usually a diagonal matrix, and in the process of predicting the battery state of charge SOC by using the unscented kalman filter UKF algorithm, the covariance matrix Q of the process noise is:

in equation (4), the parameters to be set are 3, the covariance Q of the process noise ₁ ，Q ₂ ，Q ₃ ；

The covariance matrix R of the measurement noise is used for estimating the error and uncertainty of a measurement model, and is usually a diagonal matrix, and in the process of predicting the battery charge state SOC by using the unscented Kalman filter UKF algorithm, the covariance matrix R of the measurement noise is as follows:

R＝[R ₁ ] (5)

then in equation (5), the parameters to be set are 1, which is the covariance R of the measurement noise ₁ ；

The sampling parameter alpha is used for controlling the distribution density of Sigma points, and proper values are selected to ensure the full coverage of UT points, and the value range of the sampling parameter alpha is between 0 and 1;

step S2 tests the predicted battery to obtain UUDS data of the urban road circulation working condition, and the specific steps are as follows:

s2-1: determining rated capacity and rated voltage of a battery to be measured, and determining parameters of a testing device, including load resistance and load reactance;

s2-2: charging the battery to be measured to 100% of electric quantity, and standing for 1-2 hours to stabilize the battery state;

s2-3: removing the battery from the charger and allowing it to stand at room temperature for at least 30 minutes so that the internal temperature of the battery reaches room temperature;

s2-4: placing the battery on a UUDS testing device under the urban road circulation working condition, and ensuring that two ends of the battery are correctly contacted with electrodes of the testing device;

s2-5: setting testing conditions including a current range, a voltage range, a temperature and a time according to the UUUDS requirement of the urban road circulation working condition, and ensuring that the testing conditions accord with the service condition of the battery;

s2-6: starting a test, and recording test data including current, voltage and time until the electric quantity of the battery is reduced to a certain degree or the test time reaches a preset value;

s2-7: analyzing the test data, evaluating the validity of the data, and taking the current data and the voltage data in the test data for calculating in the step S3;

the step S3 adopts a criss-cross algorithm CSO to optimize an unscented Kalman filter UKF algorithm, and comprises the following specific steps:

s3-1: initializing upper and lower limits of five parameters in step S1, including covariance Q of process noise ₁ ，Q ₂ ，Q ₃ Covariance R of measurement noise ₁ And sampling parameter alpha, randomly generating N data pairs, called population, in the interval;

s3-2: and (3) carrying out first population fitness calculation by adopting the N data pairs generated in the step (S3-1), wherein the larger the population fitness is, the smaller the error of the group of parameters in the fitting process is represented, and the specific calculation steps of the population fitness are as follows:

s3-2-1: obtaining an observation matrix Z according to the test conditions determined in the step S2-1;

s3-2-2: initializing unscented transformation UT parameters according to the input parameter data pair to be set;

s3-2-3: adding certain normal noise to the current data of the UUDS of the urban road circulation working condition obtained in the step S2 to simulate the actual working environment;

s3-2-4: generating Sigma points according to the current data added with noise obtained in the step S3-2-3;

s3-2-5: carrying out state prediction of Kalman filtering KF according to the Sigma points obtained in the step S3-2-4;

s3-2-6: performing Sigma point reconstruction according to the state prediction result obtained in the step S3-2-5;

s3-2-7: according to the voltage data of the UUZDS of the urban road circulation working condition obtained in the step S2 and the point after the state prediction reconstruction obtained in the step S3-2-6, carrying out the observation updating of the Kalman filtering KF;

s3-2-8: repeating steps S3-2-4 to S3-2-7 until a predicted value of the battery state of charge is obtained for the whole test time period;

s3-2-9: calculating the state of charge of the battery by adopting an ampere-hour integration method, and taking the state of charge as a true reference value of the state of charge of the battery;

s3-2-10: calculating an error mean value by taking the prediction result of the S3-2-8 and the result of the ampere-hour integration method of the S3-2-9 as a standard;

s3-2-11: repeating the steps S3-2-4 to S3-2-10 for 5 times to eliminate accidental prediction too good or too bad caused by current noise, taking the average value of error means of 5 times, taking the reciprocal value as the population fitness of the population, and outputting the population fitness to the step S3-3 for use;

s3-3: the longitudinal crossover operation of the longitudinal crossover method CSO is carried out, the longitudinal crossover process is an arithmetic operation of crossing the preference parameters with different dimensions, and the expression is as follows:

M _vc (m,d ₁ )＝r×X(m,d ₁ )+(1-r)×X(m,d ₂ )+c(X(m,d ₁ )-X(m,d ₂ )) (6)

(6) Wherein r and c are random numbers ranging from 0 to 1, X (m, d ₁ ) And X (m, d) ₂ ) For parent preference parameters of different dimensions, M _vc (m,d ₁ ) The offspring trending parameters generated by longitudinally crossing parent different dimensions are represented by parameters m=1, 2, …, N and N being population sizes; d, d ₁ ,d ₂ =1, 2,..c, C is the number of dimensions of the population;

s3-4: calculating the population fitness of the offspring generated in the step S3-3, and comparing the population fitness with the population fitness of the father, and reserving the population with larger population fitness to enable the whole population to evolve towards a better direction;

s3-5: the transverse crossover operation of the crisscross CSO is carried out, the transverse crossover process is an arithmetic operation of crossing all the optimal parameters of different solutions, and the expression is as follows:

(7) Wherein, the parameter n=1, 2, …, N is population size;

s3-6: calculating the population fitness of the offspring generated in the step S3-5, and comparing the population fitness with the population fitness of the father, and reserving the population with larger population fitness to enable the whole population to evolve towards a better direction;

s3-7: the population is regenerated by the following steps:

s3-7-1: if the unscented Kalman filtering UKF algorithm prediction process is terminated, marking the population fitness of the population as 0;

s3-7-2: regenerating the population with the statistical population fitness of 0 before each round of iteration is finished;

s3-7-3: obtaining each dimension i, and respectively obtaining the maximum value max of the population with the population fitness not being 0 in the dimension _i And minimum value min _i ；

S3-7-4: when generating new population, the parameters of each dimension i are respectively in (1+k) _s )max _i To (1-k) _s )min _i Is regenerated, k _s For generating the diffusion coefficient, the population is not easy to converge to cause the population fitness to fall into local optimum;

s3-8: repeating the steps S3-3 to S3-7 until the iteration number reaches the maximum iteration number set by the system;

s3-9: acquiring a population with optimal population fitness at the moment as an optimal population, and taking the optimal population as a model application parameter of the step S4;

and step S4, applying the optimized unscented Kalman filtering UKF algorithm to the predicted battery state of charge SOC to obtain a more accurate battery working state, thereby improving the service life and the service efficiency of the battery.