Battery charge state estimation method based on strong tracking adaptive unscented Kalman filtering

Abstract

The invention provides a battery state-of-charge estimation method based on strong tracking adaptive unscented Kalman filtering. The method comprises the following steps: generating a state vector Sigma point set and a corresponding weight; updating the state matrix and the covariance matrix; the second Sigma point set corresponds to the state matrix; updating the measurement matrix; updating calculation; if an emergency occurs, calculating a strong tracking coefficient, and returning to update the state matrix and the covariance matrix; if no emergency situation occurs, calculating a Kalman filtering gain coefficient, updating and outputting the state matrix and the covariance matrix, and then judging whether the circulation is ended or not; if the circulation is finished, the whole process is finished; and if the circulation is not finished, updating the noise matrix, and returning to generate the state vector Sigma point set and the corresponding weight. The method is composed of the ST algorithm and the A-UKF algorithm, the ST-A-UKF algorithm can improve the real battery SOC value tracking capacity under the condition that the environment changes, and the method has good robustness, high adaptability and good convergence.

Description

Technical Field

[0001] The invention belongs to the technical field of marine battery management and control, and in particular relates to a battery state of charge estimation method based on strong tracking adaptive unscented Kalman filtering. Background Art

[0002] Battery state of charge (SOC) estimation is an important function of the battery management system. Accurate estimation of battery SOC determines the reliability and safety of the diesel-electric hybrid system and is crucial to the battery management system. Although various online battery SOC estimation methods try to improve the estimation accuracy as much as possible within a limited range, few patents discuss the sources of errors in these battery SOC estimation methods. Accurate SOC estimation is of great significance to the safe operation of hybrid ships and to improving the service life of batteries. Among the existing battery SOC estimation methods, the Unscented Kalman Filter (UKF) algorithm is widely used in SOC estimation due to its lossless transformation and high estimation accuracy. However, the traditional UKF algorithm is greatly affected by system noise and observation noise during the battery SOC estimation process.

[0003] As the battery ages, the accuracy of battery SOC estimation becomes lower and lower, shortening the battery life, and the fuel economy of the ship during the entire battery life cycle is poor. Under complex navigation conditions, in order to fully improve battery performance, it is necessary to estimate the battery SOC online. However, existing patents rarely consider the temperature changes during the actual ship's navigation, battery aging conditions, and the requirements for real-time and noise during the navigation process, which leads to large errors in the battery SOC estimation in actual ships. As the battery ages, the accuracy of battery SOC estimation will become lower and lower, shortening the service life of the power battery, and the fuel economy of the ship will become worse and worse during the entire battery life cycle.

[0004] During the voyage of a ship, the navigation conditions may change at any time, and the working state of the battery will also change accordingly. The noise characteristics collected are generally more complex Gaussian white noise, rather than simple Gaussian white noise. This requires that the battery SOC estimation method used can not only update the battery SOC in real time, but also must meet the following two requirements: First, when the external environment changes, the SOC estimation method has a fast response speed, can track the battery status in a very short time, and feed back the real-time battery status to the battery management system of the centralized control center; second, in addition to changes in the external environment, during the battery charging and discharging process, the battery SOC estimation accuracy must be high, and accurate battery SOC estimation can be achieved in a non-Gaussian white noise environment. Summary of the invention

[0005] In order to solve the technical problems in the prior art, the present invention provides a battery state of charge estimation method based on strong tracking adaptive unscented Kalman filtering, comprising the following steps:

[0006] start;

[0007] Production state vector Sigma point set and corresponding weights;

[0008] Update the state matrix and covariance matrix;

[0009] The second Sigma point set corresponds to the state matrix;

[0010] Update the measurement matrix;

[0011] Update calculations;

[0012] If an emergency / burst occurs, the strong tracking coefficients are calculated and the updated state matrix and covariance matrix are returned;

[0013] If no emergency / burst occurs, the Kalman filter gain coefficient is calculated, and the state matrix and covariance matrix are updated and output, and then it is determined whether the loop is completed;

[0014] If the loop ends, the entire process ends; if the loop does not end, the noise matrix is ​​updated, and the generated state vector Sigma point set and corresponding weights are returned.

[0015] Further, updating the state matrix and the covariance matrix includes:

[0016] When initializing the UKF algorithm, the covariance matrix P is first set to a large value so that the model has the ability to adapt to wide noise areas to simulate emergencies. The P value of the UKF algorithm is adjusted to improve the ability to cope with emergencies.

[0017] Furthermore, the P value of the UKF algorithm is adjusted to improve the ability to cope with emergencies, including:

[0018] Solve from the following formula and P k

[0019]

[0020]

[0021] The sudden change of the external environment can be regarded as a transient process, which will have a certain impact on the system in a short period of time. After a period of time, the system state will automatically return to its original state; where: when the strong tracking coefficient γ is less than or equal to 1, no fading factor is added, and the normal UKF algorithm is used. When the strong tracking coefficient γ is greater than 1, the variable fading factor γ is introduced 0kWhen the covariance matrix is ​​calculated according to γ 0k to correct;

[0022]

[0023] In formula (4), for The trace of the matrix; P tk is the new information covariance, as shown in formula (5); P tk is the statistical width, which is obtained from the actual new information of m; the value of m can be pre-set according to the actual situation. The larger the value of m, the higher the estimation accuracy, and the corresponding estimation response speed is slower, as shown in formula (6):

[0024]

[0025] Further, updating the noise matrix includes:

[0026] The Sage-Husa adaptive filtering algorithm is added, and the Sage-Husa noise estimator is used to calculate the unknown noise system, and it is updated in real time through iterative cycles. The forgetting factor is introduced to adjust the noise estimator in real time according to the noise changes, thereby improving the tracking effect of the A-UKF algorithm on noise changes, eliminating historical data, and avoiding the influence of disturbance variables, thereby improving the robustness of the algorithm. The adaptive filtering algorithm corrects the measurement noise R and system noise Q in the UKF algorithm in real time, thereby achieving the goal of improving the accuracy of system state estimation by the improved UKF algorithm in a time-varying noise environment.

[0027] Furthermore, the implementation process of the adaptive filtering algorithm is as follows:

[0028] (1) Noise q in the state equation k+1 The average value of:

[0029]

[0030] Where: d k It can be expressed as:

[0031]

[0032] In the formula: c is the genetic factor;

[0033] (2) Covariance of noise in the state equation:

[0034]

[0035] Where: e k It can be expressed as:

[0036]

[0037] (3) Noise r in the measurement equation k+1 The average value calculation formula is:

[0038]

[0039] (4) Noise r in the measurement equation k+1 The covariance of is:

[0040]

[0041] At the end of each iteration, the noise matrix in the UKF algorithm is Instead, the next iteration is performed, allowing the A-UKF algorithm to update the noise online.

[0042] Furthermore, the algorithm implementation includes:

[0043] Construct equations and determine parameters;

[0044] Algorithm initialization.

[0045] Further, the construction equations and parameter determination include:

[0046] The measurement equations and state equations required to establish the ST-A-UKF algorithm are:

[0047]

[0048] The coefficient α is 0.02, the coefficient β is 2, and k is an undetermined parameter whose value is 3-n.

[0049] 8. The method according to claim 6, characterized in that the algorithm initialization comprises:

[0050] The state vector is:

[0051]

[0052] The covariance matrix is:

[0053]

[0054] The system noise matrix is:

[0055]

[0056] The measurement noise matrix is:

[0057] r0=[0.002],

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] The present invention is composed of an ST algorithm and an A-UKF algorithm, which is suitable for estimating the battery SOC in emergencies and non-Gaussian white noise environments. The ST algorithm, the adaptive algorithm and the UKF algorithm are combined together to give full play to their respective advantages and expand the battery estimation range. When the environment changes, the ST-A-UKF algorithm can improve the ability to track the actual battery SOC value, and has good robustness, strong adaptability and good convergence. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a specific flow chart of the ST-A-UKF algorithm of the present invention;

[0061] Figure 2 It is a block diagram of the algorithm combination of the battery model of the present invention and the ST-A-UKF algorithm. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described below in conjunction with the accompanying drawings and specific implementation methods.

[0063] 1 Unscented Kalman Filter

[0064] 1.1 Traceless transformation

[0065] The UKF algorithm analyzes the probability density of the function relationship in detail, and then obtains the sgma point corresponding to the estimated value after a series of traceless transformations, approximating the probability density of the state, reducing the error, greatly improving the accuracy of the estimated value, and reducing the difficulty in obtaining the Jacobian matrix for the state observation matrix and the transfer matrix respectively. In the calculation process, the influence of high-order terms on the matrix is ​​fully considered, which improves the accuracy of the estimated value.

[0066] If there is a nonlinear system function y = f(x), its state variable x has n variables, the covariance matrix is ​​P, and the mean is There are 2n+1 Sigma points in total, among which:

[0067]

[0068] Where: Represents the i-th column of the square root of the matrix P.

[0069] The weights corresponding to each sampling point are:

[0070]

[0071] Where: w c is the weight of the covariance; w m Represents the weight of the mean square error; λ = α 2(n+κ)-n, α represents the distribution state of the sampling points, and its value range is between 0 and 1; k is an undetermined parameter; β≥0, which is a non-negative weight coefficient to combine the moments of high-order terms in the equation.

[0072] 1.2 Unscented Kalman Filter

[0073] Substituting x into formulas (1) and (2) k Perform untraceable transformation in the system to obtain the x-column vector Sigma point matrix and its weight at time k.

[0074] (1) The x column vector sigma point matrix and weight calculation formula at time k are as follows:

[0075]

[0076] (2) Weighted state matrix It can be obtained from equations (4) and (5) and the corresponding covariance matrix P k+1|k get.

[0077]

[0078] (3) In equation (5), State Matrix is substituted into equation (1) and then an unscented transformation is used to generate a series of new sigma points, where i = 0, 2, ... 2n.

[0079]

[0080] (4) Substitute into the observation equation, through the corresponding mapping relationship Get the predicted observation matrix Y, where i = 0, 2, … 2n.

[0081]

[0082] (5) The mean and covariance matrix of the predicted observation matrix are calculated from the corresponding weights in (2).

[0083]

[0084] (6) In formula (9), According to equations (10) and (11), the corresponding Kalman gain coefficient K is obtained k+1 .

[0085]

[0086] (7) Then the gain coefficient Update and covariance matrix Pk+1∣k=1 and the state matrix P k+1∣k=1 , where K k+1 represents the measurement matrix, represents the prediction observation matrix.

[0087]

[0088] Updated state matrix Output and P k+1∣k+1 It is used as the initial value of the next loop and the next iteration is executed until the loop ends. In fact, the biggest difference between the UKF algorithm and the EKF algorithm is that the EKF algorithm solves the Jacobian matrix of the state equation and the Jacobian matrix of the measurement equation separately in each loop, which increases the computational difficulty. The UKF algorithm uses the solution of a certain number of sigma points instead of the step of linearizing the nonlinear equation, which greatly reduces the algorithm difficulty, reduces the number of iterations, improves the computational efficiency, and reduces the estimation time.

[0089] 2 Strong Tracking Adaptive Unscented Kalman Filter Algorithm

[0090] During the voyage of a ship, the navigation conditions may change at any time, and the working state of the battery will also change accordingly. The noise characteristics collected are generally more complex Gaussian white noise, rather than simple Gaussian white noise. This requires that the battery SOC estimation method used can not only update the battery SOC in real time, but also must meet the following two requirements: First, when the external environment changes, the SOC estimation method has a fast response speed, can track the battery status in a very short time, and feed back the real-time battery status to the battery management system on the bridge; second, in addition to changes in the external environment, during the battery charging and discharging process, the battery SOC estimation accuracy must be high, and accurate battery SOC estimation can be achieved in a non-Gaussian white noise environment.

[0091] In view of the low accuracy of battery SOC estimation under sudden changes, which causes short-term distortion, this design combines the strong tracking algorithm (ST) and the adaptive unscented Kalman filter algorithm (A-UKF) into a strong tracking adaptive unscented Kalman filter algorithm (ST-AUKF). When initializing the UKF algorithm, the covariance matrix P is first set to a large value so that the model has the ability to adapt to wide noise areas to simulate emergencies, and the P value of the UKF algorithm is adjusted to improve the ability to cope with emergencies. It can be solved from equation (14) that the covariance matrix P k It can be solved from equation (15).

[0092]

[0093] The sudden change of the external environment can be regarded as a transient process, which will have a certain impact on the system in a short period of time. After a period of time, the system state will automatically return to its original state. In the formula: when the strong tracking coefficient γ is less than or equal to 1, no fading factor is added, and the normal UKF algorithm is used. When the strong tracking coefficient γ is greater than 1, the variable fading factor γ is introduced 0k When the covariance matrix is ​​calculated according to γ 0k To correct.

[0094]

[0095] In formula (17), for The trace of the matrix; P tk is the new information covariance, as shown in formula (18); P tk is the statistical width, which is obtained from the actual new information of m; the value of m can be pre-set according to the actual situation. The larger the value of m, the higher the estimation accuracy and the corresponding slower the estimation response speed, as shown in formula (19).

[0096]

[0097] This design uses the idea of ​​algorithm combination, adds the Sage-Husa adaptive filtering algorithm, and proposes an adaptive unscented Kalman filtering algorithm. The basic idea of ​​the Sage-Husa adaptive filtering algorithm is to calculate the unknown noise system by using the Sage-Husa noise estimator, and perform iterative cycles to update it in real time. By introducing the forgetting factor, the algorithm can adjust the noise estimator in real time according to the noise changes, improve the tracking effect of the A-UKF algorithm on noise changes, eliminate historical data, and avoid the influence of disturbance variables, thereby improving the robustness of the algorithm. The adaptive filtering algorithm corrects the measurement noise R and system noise Q in the UKF algorithm in real time, so that the improved UKF algorithm can improve the accuracy of system state estimation in a time-varying noise environment. The implementation process of the adaptive filtering algorithm is as follows:

[0098] (1) Noise q in the state equation k+1 The average value of:

[0099]

[0100] Where: d k It can be expressed as:

[0101]

[0102] Where: c is the genetic factor.

[0103] (2) Covariance of noise in the state equation:

[0104]

[0105] Where: e k It can be expressed as:

[0106]

[0107] (3) Noise r in the measurement equation k+1 The average value calculation formula is:

[0108]

[0109] (4) Noise r in the measurement equation k+1 The covariance of is:

[0110]

[0111] At the end of each iteration, the noise matrix in the UKF algorithm is Instead, the next iteration is performed, allowing the A-UKF algorithm to update the noise online.

[0112] The ST-A-UKF algorithm consists of the ST algorithm and the A-UKF algorithm. The algorithm is suitable for estimating the battery SOC in emergencies and non-Gaussian white noise environments. The specific process is as follows: Figure 1 shown.

[0113] Although the traditional UKF algorithm has high estimation accuracy, it is not very adaptable when the working environment of the battery changes. Combining the STF algorithm, adaptive algorithm and UKF algorithm, and giving full play to their respective advantages, can expand the battery estimation range. It can improve the ability to track the actual battery SOC value when the environment changes, and has good robustness, strong adaptability and good convergence.

[0114] 3ST-A-UKF algorithm implementation

[0115] 3.1 Construction equation and parameter determination

[0116] Since the battery model is not linear but nonlinear, it needs to be linearized using an untraceable transformation. The untraceable transformation is used to obtain the sigma point set of the same probability density function as the state at the previous moment, and the mapping of each sigma point in the state equation is obtained. Because the measurement equation itself is not linear, the state vector estimated at the next moment must be transformed again using an untraceable transformation to obtain the sigma point set. Substitute multiple groups of sigma points into the measurement equation to obtain the mapping of the predicted state vector in the measurement equation, obtain the Kalman filter gain coefficient, and finally obtain the state vector for the next step. The combined block diagram of the battery model and the ST-A-UKF algorithm is shown in the figure. Figure 2The upper part is the battery model, which is mainly composed of the input state equation, battery parameter matrix and measurement matrix; the lower part is the algorithm model, which is mainly composed of the state equation and observation equation covariance matrix.

[0117] The measurement equations and state equations required to establish the ST-A-UKF algorithm are:

[0118]

[0119] The coefficient α is 0.02, the coefficient β is 2, and k is an undetermined parameter whose value is 3-n.

[0120] 3.2 Algorithm Initialization

[0121] The state vector is:

[0122]

[0123] The covariance matrix is:

[0124]

[0125] The system noise matrix is:

[0126]

[0127] The measurement noise matrix is:

[0128] r0=[0.002],

[0129] In summary, the above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereto. All equivalent changes and modifications made according to the patent scope of the present invention and the contents of the specification are within the scope covered by the patent of the present invention.

Claims

1. A battery state of charge estimation method based on strong tracking adaptive unscented Kalman filtering, characterized in that: The following steps are involved: start; Production state vector Sigma point set and corresponding weights; Update the state matrix and covariance matrix; The second Sigma point set corresponds to the state matrix; Update the measurement matrix; Update calculations; If an emergency / burst occurs, the strong tracking coefficients are calculated and the updated state matrix and covariance matrix are returned; If no emergency / burst occurs, the Kalman filter gain coefficient is calculated, and the state matrix and covariance matrix are updated and output, and then it is determined whether the loop is completed; If the loop ends, the whole process ends; If the loop is not ended, the noise matrix is ​​updated and the generated state vector Sigma point set and the corresponding weights are returned.

2. The method according to claim 1, characterized in that Updating the state matrix and covariance matrix involves: When initializing the UKF algorithm, the covariance matrix P is first set to a large value so that the model has the ability to adapt to wide noise areas to simulate emergencies. The P value of the UKF algorithm is adjusted to improve the ability to cope with emergencies.

3. The method according to claim 2, characterized in that The P value of the UKF algorithm is adjusted to improve the ability to cope with emergencies, including: Solve from the following formula and P k The sudden change of the external environment can be regarded as a transient process, which will have a certain impact on the system in a short period of time. After a period of time, the system state will automatically return to its original state; where: when the strong tracking coefficient γ is less than or equal to 1, no fading factor is added, and the normal UKF algorithm is used. When the strong tracking coefficient γ is greater than 1, the variable fading factor γ is introduced 0k When the covariance matrix is ​​calculated according to γ 0k to correct; In formula (4), for The trace of the matrix; P tk is the new information covariance, as shown in formula (5); P tk is the statistical width, which is obtained from the actual new information of m; the value of m can be pre-set according to the actual situation. The larger the value of m, the higher the estimation accuracy, and the corresponding estimation response speed is slower, as shown in formula (6):

4. The method according to claim 1, characterized in that Update the noise matrix include:. The Sage-Husa adaptive filtering algorithm is added, and the Sage-Husa noise estimator is used to calculate the unknown noise system, and it is updated in real time through iterative cycles. The forgetting factor is introduced to adjust the noise estimator in real time according to the noise changes, thereby improving the tracking effect of the A-UKF algorithm on noise changes, eliminating historical data, and avoiding the influence of disturbance variables, thereby improving the robustness of the algorithm. The adaptive filtering algorithm corrects the measurement noise R and system noise Q in the UKF algorithm in real time, thereby achieving the goal of improving the accuracy of system state estimation by the improved UKF algorithm in a time-varying noise environment.

5. The method according to claim 4, characterized in that The implementation process of the adaptive filtering algorithm is as follows: (1) Noise q in the state equation k+1 The average value of: Where: d k It can be expressed as: In the formula: c is the genetic factor; (2) Covariance of noise in the state equation: Where: e k It can be expressed as: (3) Noise r in the measurement equation k+1 The average value calculation formula is: (4) Noise r in the measurement equation k+1 The covariance of is: At the end of each iteration, the noise matrix in the UKF algorithm is Instead, the next iteration is performed, allowing the A-UKF algorithm to update the noise online.

6. The method according to claim 1, characterized in that The algorithm implementation includes: Construct equations and determine parameters; Algorithm initialization.

7. The method according to claim 6, characterized in that The construction equations and parameter determination include: The measurement equations and state equations required to establish the ST-A-UKF algorithm are: The coefficient αα is taken as 0.02, the coefficient β is taken as 2, and k is an undetermined parameter, and its value is 3-n.

8. The method according to claim 6, characterized in that the algorithm initialization comprises: The state vector is: The covariance matrix is: The system noise matrix is: The measurement noise matrix is: r0＝[0.002]，

Images