A multi-algorithm fusion SOC estimation method for power batteries based on Bayesian probability

Abstract

The present invention specifically relates to a multi-algorithm fusion SOC estimation method for a power battery based on Bayesian probability, including: constructing a corresponding equivalent circuit model, and performing parameter identification on the model parameters corresponding to the equivalent circuit model; constructing multiple observers for predicting the terminal voltage and SOC of the power battery based on multiple estimation algorithms; under dynamic stress test conditions, collecting the associated parameters of the power battery, inputting each observer to output the corresponding terminal voltage prediction value and SOC estimation value; assigning corresponding weighted values ​​based on the residual between the terminal voltage prediction value and the corresponding measured value at different times in combination with Bayesian probability; and performing weighted accumulation of the SOC estimation values ​​output by each observer based on the corresponding weighted values ​​to generate the corresponding fusion SOC estimation value. The present invention can effectively integrate the SOC estimation results of multiple estimation algorithms and realize the complementarity of multiple estimation algorithms, so as to maintain the global optimal SOC estimation in the entire charge and discharge interval of the power battery.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle power battery management, and in particular to a power battery multi-algorithm fusion SOC estimation method based on Bayesian probability. Background Art

[0002] Power batteries are currently widely used in the field of electric vehicles. They are a key factor affecting the performance of electric vehicles, and have a direct impact on driving mileage, acceleration capability, and maximum climbing grade. Battery state of charge (SOC) estimation is the core and difficulty of battery management system research. The nonlinear characteristics of power batteries make it difficult for many filtering methods to obtain accurate estimation results. In the prior art, equivalent circuit models are often used in the management and analysis of power battery status. A circuit network with dynamic characteristics composed of resistors, capacitors, and voltage sources based on the dynamic characteristics and working principles of the battery is used to describe the open circuit voltage, ohmic internal resistance, and polarization and diffusion phenomena of the battery's internal reaction process, and then the power battery SOC is estimated through the corresponding filtering algorithm.

[0003] At present, the main methods for estimating the SOC of power batteries include the ampere-hour measurement method, the open-circuit voltage method, and the neural network method. The ampere-hour measurement method focuses on the application occasion, and is online, convenient, and accurate, but requires high-precision measurement equipment; the open-circuit voltage method is only applicable to estimation after the battery has been stationary for a long enough time, and cannot be estimated in real time; the neural network can be estimated online, but the disadvantage is that a large amount of training data of similar batteries is required. To this end, the Chinese patent with publication number CN111098755A discloses "A method for estimating the SOC of power batteries for electric vehicles", including: measuring the current and voltage of the power battery by the intermittent constant current discharge method, and determining the functional relationship of the OCV-SOC of the power battery by the polynomial fitting method; establishing a second-order RC equivalent circuit model of the power battery and deriving its identification model; constructing the process of the improved chaotic gravitational search algorithm to complete the identification; establishing a state space model of the battery for the nonlinear system of lithium-ion batteries, constructing the process of the Gauss-Hermitian filtering algorithm, and forming a joint estimation algorithm with the improved chaotic gravitational search algorithm.

[0004] The power battery SOC estimation method in the above-mentioned existing scheme improves the real-time performance of SOC estimation by jointly estimating model parameter identification and SOC estimation. However, the applicant has found that the existing power battery SOC estimation method generally relies on a single estimation algorithm to complete SOC estimation, and a single estimation algorithm cannot guarantee that the estimation of the power battery state is always the most accurate throughout the entire charge and discharge interval, that is, it is impossible to always keep the SOC estimation optimal throughout the entire charge and discharge interval of the power battery. For this reason, the applicant has thought of using multiple estimation algorithms to achieve global SOC estimation of the power battery, so that the SOC estimation can be kept globally optimal throughout the entire charge and discharge interval of the power battery. However, how to design a framework that can effectively integrate the SOC estimation results of multiple estimation algorithms and realize the complementarity of multiple estimation algorithms is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a power battery multi-algorithm fusion SOC estimation method based on Bayesian probability, which can effectively integrate the SOC estimation results of multiple estimation algorithms and realize the complementarity of multiple estimation algorithms, so as to maintain the global optimal SOC estimation in the entire charging and discharging range of the power battery and ensure the accuracy of the power battery SOC estimation.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] The multi-algorithm fusion SOC estimation method for power batteries based on Bayesian probability includes the following steps:

[0008] S1: constructing a corresponding equivalent circuit model based on the corresponding power battery, and then performing parameter identification on model parameters corresponding to the equivalent circuit model through associated parameters for characterizing the power battery terminal voltage and SOC;

[0009] S2: Multiple observers are constructed based on multiple estimation algorithms to predict the terminal voltage and SOC of the power battery;

[0010] S3: Under the dynamic stress test condition, the associated parameters of the power battery are collected based on the equivalent circuit model; then the associated parameters of the power battery are respectively input into each observer to output the corresponding terminal voltage prediction value and SOC estimation value;

[0011] S4: Based on the residual between the terminal voltage prediction value output by each observer at different times and the corresponding measured value, a corresponding weighted value is assigned to the SOC estimation value output by each observer in combination with Bayesian probability;

[0012] S5: Perform weighted accumulation on the SOC estimation values ​​output by each observer based on the corresponding weighted value to generate a corresponding fused SOC estimation value.

[0013] Preferably, in step S1, the equivalent circuit model includes but is not limited to a Rint model, a Thevenin model and a DP model.

[0014] Preferably, in step S1, the equivalent circuit model is a Thevenin model, and its circuit equation is as follows:

[0015]

[0016]

[0017] Where: U t Indicates the terminal voltage of the power battery; U D Indicates the voltage drop of the RC parallel link of the power battery; U oc Represents the ideal voltage source of the power battery; R i Indicates the ohmic internal resistance of the power battery; R D Indicates the polarization resistance of the power battery; C D Represents the polarization capacitance of the power battery; i L Indicates the current flowing through the power battery.

[0018] Preferably, in step S1, the associated parameters used to characterize the terminal voltage and SOC of the power battery include but are not limited to the terminal voltage, charge and discharge current, and surface temperature data of the power battery.

[0019] Preferably, in step S1, the terminal voltage, charge and discharge current and surface temperature data of the power battery are collected at intervals of sampling time Δt, and then the model parameters corresponding to the equivalent circuit model are identified by the terminal voltage, charge and discharge current and surface temperature data of the power battery.

[0020] Preferably, in step S1, the parameter identification method includes but is not limited to Kalman filtering, H∞ Kalman filtering, genetic algorithm, particle swarm algorithm, recursive least squares algorithm and least mean square algorithm.

[0021] Preferably, in step S2, firstly, a corresponding state equation and a measurement equation are constructed based on the equivalent circuit model, and then an observer for predicting the power battery terminal voltage and SOC is constructed based on the state equation and the measurement equation in combination with a corresponding estimation algorithm.

[0022] Preferably, in step S2, the multiple estimation algorithms include but are not limited to EKF algorithm, AEKF algorithm and H∞ algorithm.

[0023] Preferably, step S4 specifically includes the following steps:

[0024] S401: Calculate the deviation between the predicted value and the measured value of the terminal voltage output by the i-th observer at time k;

[0025] in, Where: r k,i Indicates the deviation between the predicted value and the measured value of the terminal voltage; U k,i represents the measured value of the terminal voltage at time k; is the terminal voltage prediction value at time k output by the i-th observer;

[0026] S402: Calculate the terminal voltage residual mean and the terminal voltage residual mean square error of each observer at time k based on the deviation between the terminal voltage prediction value and the measured value;

[0027] in, Where: Represents the mean value of the terminal voltage residual; s k,i represents the residual mean square error of the terminal voltage; M represents the set moving window length;

[0028] S403: Calculate the conditional probability density function of each observer at time k based on the terminal voltage residual mean and the terminal voltage residual mean square error;

[0029] in, Where: represents the conditional probability density function; α i represents the i-th observer;

[0030] S404: Calculating the weighted value of each observer at time k based on the conditional probability density function of each observer at time k;

[0031] in, Where: k,i represents the weighted value of the i-th observer at time k; n represents the number of observers.

[0032] Preferably, in step S5, the fused SOC estimation value is calculated by the following formula:

[0033]

[0034] Where: Z F,k Indicates the fusion SOC value; Z k,i represents the SOC estimation value output by the i-th observer at time k.

[0035] Compared with the prior art, the power battery multi-algorithm fusion SOC estimation method of the present invention has the following beneficial effects:

[0036] The present invention constructs multiple observers for predicting the terminal voltage and SOC of the power battery through multiple estimation algorithms, so that the SOC estimation of the power battery can be realized through multiple estimation algorithms, which is beneficial to achieving the most accurate estimation of the power battery state in the entire charge and discharge interval, that is, the SOC estimation can be kept globally optimal in the entire charge and discharge interval of the power battery, thereby ensuring the accuracy of the SOC estimation of the power battery.

[0037] Secondly, the present invention assigns corresponding weighted values ​​to the SOC estimation values ​​output by each observer through the residual between the terminal voltage prediction value and the measured value combined with the Bayesian probability, so that the SOC estimation results of multiple estimation algorithms can be effectively integrated and the complementarity of multiple estimation algorithms can be achieved. Compared with a single estimation algorithm, it has a smoother and more accurate SOC estimation result, thereby ensuring the accuracy of the power battery SOC estimation.

[0038] Finally, the present invention collects relevant parameters of the power battery under dynamic stress test conditions, so that it can simulate the operating condition changes of the power battery under the actual driving state of the electric vehicle, thereby further improving the accuracy of the power battery SOC estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to make the purpose, technical solution and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0040] Figure 1 The logical block diagram of the multi-algorithm fusion SOC estimation method for power batteries based on Bayesian probability;

[0041] Figure 2 It is a flow chart of the multi-algorithm fusion SOC estimation method for power batteries based on Bayesian probability;

[0042] Figure 3 (a) is a general schematic diagram showing the comparison between the estimated SOC values ​​and the experimental values ​​before and after the fusion of each observer (estimation algorithm);

[0043] Figure 3 (b) and Figure 3 (c) Yes Figure 3 (a) Schematic diagram of two different locations.

[0044] Figure 4 It is the SOC error analysis result before and after the fusion of each observer (estimation algorithm) under dynamic stress test conditions. DETAILED DESCRIPTION

[0045] The following is a further detailed description through specific implementation methods:

[0046] Example:

[0047] This embodiment discloses a power battery multi-algorithm fusion SOC estimation method based on Bayesian probability.

[0048] like Figure 1 and Figure 2 As shown in the figure, the multi-algorithm fusion SOC estimation method of power battery based on Bayesian probability includes:

[0049] S1: construct a corresponding equivalent circuit model based on the corresponding power battery, and then perform parameter identification on the model parameters corresponding to the equivalent circuit model through the associated parameters used to characterize the power battery terminal voltage and SOC; in this embodiment, the parameter identification method includes but is not limited to Kalman filtering, H∞ Kalman filtering, genetic algorithm, particle swarm algorithm, recursive least squares algorithm and least mean square algorithm.

[0050] S2: Multiple observers are constructed based on multiple estimation algorithms to predict the terminal voltage and SOC of the power battery;

[0051] S3: Under the dynamic stress test condition, the associated parameters of the power battery are collected based on the equivalent circuit model; then the associated parameters of the power battery are respectively input into each observer to output the corresponding terminal voltage prediction value and SOC estimation value;

[0052] S4: Based on the residual between the terminal voltage prediction value output by each observer at different times and the corresponding measured value, a corresponding weighted value is assigned to the SOC estimation value output by each observer in combination with Bayesian probability;

[0053] S5: Perform weighted accumulation on the SOC estimation values ​​output by each observer based on the corresponding weighted value to generate a corresponding fused SOC estimation value.

[0054] The present invention constructs multiple observers for predicting the terminal voltage and SOC of the power battery through multiple estimation algorithms, so that the SOC estimation of the power battery can be realized through multiple estimation algorithms, which is conducive to achieving the most accurate estimation of the power battery state in the entire charging and discharging interval, that is, it can maintain the global optimal SOC estimation in the entire charging and discharging interval of the power battery, thereby ensuring the accuracy of the power battery SOC estimation. Secondly, the present invention assigns corresponding weighted values ​​to the SOC estimation values ​​output by each observer through the residual between the terminal voltage prediction value and the measured value combined with the Bayesian probability, so that the SOC estimation results of multiple estimation algorithms can be effectively integrated and the complementarity of multiple estimation algorithms can be realized, and compared with a single estimation algorithm, it has a smoother and more accurate SOC estimation result, thereby ensuring the accuracy of the power battery SOC estimation. Finally, the present invention collects the associated parameters of the power battery under the dynamic stress test condition, so that it can simulate the working condition changes of the power battery under the actual driving state of the electric vehicle, thereby further improving the accuracy of the power battery SOC estimation.

[0055] During the specific implementation process, the equivalent circuit model includes but is not limited to the Rint model, the Thevenin model and the DP model.

[0056] In this embodiment, the equivalent circuit model is the Thevenin model, and its circuit equation is as follows:

[0057]

[0058]

[0059] Where: U t Indicates the terminal voltage of the power battery; U D Indicates the voltage drop of the RC parallel link of the power battery; U oc Represents the ideal voltage source of the power battery; R i Indicates the ohmic internal resistance of the power battery; R D Indicates the polarization resistance of the power battery; C D Represents the polarization capacitance of the power battery; i L Indicates the current flowing through the power battery.

[0060] It should be noted that the Rint model and the DP model are both existing equivalent circuit models and will not be described in detail here.

[0061] The present invention can effectively collect the relevant parameters of the power battery and assist in building an observer through the above-mentioned Thevenin model, thereby assisting in realizing the SOC estimation of the power battery.

[0062] In the specific implementation process, the associated parameters used to characterize the terminal voltage and SOC of the power battery include but are not limited to the terminal voltage, charge and discharge current and surface temperature data of the power battery. Specifically, the terminal voltage, charge and discharge current and surface temperature data of the power battery are collected at intervals of sampling time Δt, and then the model parameters corresponding to the equivalent circuit model are identified through the terminal voltage, charge and discharge current and surface temperature data of the power battery.

[0063] The present invention uses the terminal voltage, charge and discharge current, and surface temperature data of the power battery as associated parameters to characterize the terminal voltage and SOC of the power battery, so that each observer can effectively realize the prediction and estimation of the terminal voltage and SOC of the power battery based on the associated parameters, and then can better assign corresponding weighted values ​​to the SOC estimation values ​​output by each observer, so as to effectively integrate the SOC estimation results of multiple estimation algorithms and realize the complementarity of multiple estimation algorithms.

[0064] In the specific implementation process, the corresponding state equation and measurement equation are first constructed based on the equivalent circuit model, and then an observer for predicting the terminal voltage and SOC of the power battery is constructed based on the state equation and the measurement equation in combination with the corresponding estimation algorithm. Specifically, the state equation is an equation obtained based on the relationship between battery parameters. For example, in the Thevenin model, the state equation is the parameter relationship obtained based on the circuit diagram of the Thevenin model. The observation equation is an equation obtained based on measurement, such as fitting the OCV-SOC relationship, which is the observation equation. Both equations can estimate the required quantity.

[0065] During the specific implementation process, multiple estimation algorithms include but are not limited to EKF algorithm, AEKF algorithm and H∞ algorithm.

[0066] Specifically, the specific processes of the EKF algorithm, AEKF algorithm and H∞ algorithm are as follows:

[0067] 1) EKF algorithm process:

[0068]

[0069]

[0070] 2) AEKF algorithm process:

[0071]

[0072] 3) H∞ algorithm process:

[0073]

[0074]

[0075] This embodiment uses the EKF algorithm, AEKF algorithm and H∞ algorithm as an example to illustrate:

[0076] Step S4 specifically includes the following steps:

[0077] S401: Calculate the deviation between the predicted value and the measured value of the terminal voltage output by the i-th observer at time k, where i=1, 2, 3;

[0078] in, Where: r k,i Indicates the deviation between the predicted value and the measured value of the terminal voltage; U k,i represents the measured value of the terminal voltage at time k; is the terminal voltage prediction value at time k output by the i-th observer;

[0079] S402: Calculate the terminal voltage residual mean and the terminal voltage residual mean square error of each observer at time k based on the deviation between the terminal voltage prediction value and the measured value;

[0080] in, Where: Represents the mean value of the terminal voltage residual; s k,i represents the residual mean square error of the terminal voltage; M represents the set moving window length;

[0081] S403: Calculate the conditional probability density function of each observer at time k based on the terminal voltage residual mean and the terminal voltage residual mean square error;

[0082] in, Where: represents the conditional probability density function; α i represents the i-th observer;

[0083] S404: Calculating the weighted value of each observer at time k based on the conditional probability density function of each observer at time k;

[0084] in, Where: k,i represents the weighted value of the i-th observer at time k; n represents the number of observers.

[0085] Through the above steps, the present invention can effectively realize the allocation of corresponding weighted values ​​to the SOC estimation values ​​output by each observer through the residual between the terminal voltage prediction value and the measured value combined with the Bayesian probability, so that the SOC estimation results of multiple estimation algorithms can be effectively integrated and the complementarity of multiple estimation algorithms can be achieved. Compared with a single estimation algorithm, it has a smoother and more accurate SOC estimation result, thereby ensuring the accuracy of the power battery SOC estimation.

[0086] Specifically, the fused SOC estimation value is calculated by the following formula:

[0087]

[0088] Where: Z F,k Indicates the fusion SOC value; Z k,i represents the SOC estimation value output by the i-th observer at time k.

[0089] In order to better illustrate the advantages of the multi-algorithm fusion SOC estimation method of the present invention, the applicant conducted corresponding simulation experiments, in which:

[0090] Figure 3 (a) shows the comparison between the SOC estimation value and the experimental value before and after the fusion of the three observers (estimation algorithms), where FUSE represents the fusion result and Reference represents the experimental value. Figure 3 (b) and Figure 3 (c) It can be seen that the multi-algorithm fusion SOC estimation method of the present invention has a smoother SOC estimation result than each single estimation algorithm, and has better followability in the entire process of estimating SOC.

[0091] Figure 4 This is the SOC error analysis before and after the fusion of the multi-algorithm fusion SOC estimation method of the present invention under dynamic stress test conditions. Figure 4 (a) is the statistics of the maximum error of battery SOC estimation before and after the fusion of each observer (estimation algorithm). The maximum error of SOC estimation after fusion (FUSE) is 0.0046, which is higher than the AEKF algorithm and lower than the EKF algorithm and H∞ algorithm; Figure 4 (b) is the average error statistics of SOC estimation before and after the fusion of each observer (estimation algorithm). The average error of SOC estimation after fusion (FUSE) is 0.0023, which is lower than that of EKF algorithm and H∞ algorithm. Figure 4 (c) is the mean square error statistics of SOC estimation before and after fusion of each algorithm. The root mean square error of SOC estimation after fusion (FUSE) is 0.0024, which is only higher than the AEKF algorithm.

[0092] The above simulation results show that under the dynamic stress test condition, the fused SOC estimation value obtained by the multi-algorithm fusion SOC estimation method of the present invention is better than the single estimation algorithm before fusion in terms of prediction accuracy. Moreover, compared with the three single estimation algorithms, the multi-algorithm fusion SOC estimation method of the present invention can reliably and accurately predict the SOC of the power battery.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.

Claims

1. A multi-algorithm fusion SOC estimation method for power batteries based on Bayesian probability, characterized in that: The following steps are involved: S1: constructing a corresponding equivalent circuit model based on the corresponding power battery, and then performing parameter identification on model parameters corresponding to the equivalent circuit model through associated parameters for characterizing the power battery terminal voltage and SOC; S2: Multiple observers are constructed based on multiple estimation algorithms to predict the terminal voltage and SOC of the power battery; S3: Under the dynamic stress test condition, the associated parameters of the power battery are collected based on the equivalent circuit model; then the associated parameters of the power battery are respectively input into each observer to output the corresponding terminal voltage prediction value and SOC estimation value; S4: Based on the residual between the terminal voltage prediction value output by each observer at different times and the corresponding measured value, a corresponding weighted value is assigned to the SOC estimation value output by each observer in combination with Bayesian probability; S5: Perform weighted accumulation on the SOC estimation values ​​output by each observer based on the corresponding weighted value to generate a corresponding fused SOC estimation value.

2. The method for estimating SOC of a power battery using multiple algorithms based on Bayesian probability according to claim 1, characterized in that: In step S1, the equivalent circuit model includes but is not limited to a Rint model, a Thevenin model and a DP model.

3. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: In step S1, the equivalent circuit model is the Thevenin model, and its circuit equation is as follows: Where: U t Indicates the terminal voltage of the power battery; U D Indicates the voltage drop of the RC parallel link of the power battery; U oc Represents the ideal voltage source of the power battery; R i Indicates the ohmic internal resistance of the power battery; R D Indicates the polarization resistance of the power battery; C D Represents the polarization capacitance of the power battery; i L Indicates the current flowing through the power battery.

4. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: In step S1, the associated parameters used to characterize the terminal voltage and SOC of the power battery include but are not limited to the terminal voltage, charge and discharge current, and surface temperature data of the power battery.

5. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: In step S1, the terminal voltage, charge and discharge current and surface temperature data of the power battery are collected at intervals of sampling time Δt, and then the model parameters corresponding to the equivalent circuit model are identified by the terminal voltage, charge and discharge current and surface temperature data of the power battery.

6. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: In step S1, the parameter identification method includes but is not limited to Kalman filtering, H∞ Kalman filtering, genetic algorithm, particle swarm algorithm, recursive least squares algorithm and least mean square algorithm.

7. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: In step S2, firstly, the corresponding state equation and measurement equation are constructed based on the equivalent circuit model, and then an observer for predicting the power battery terminal voltage and SOC is constructed based on the state equation and measurement equation in combination with the corresponding estimation algorithm.

8. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: In step S2, the multiple estimation algorithms include but are not limited to the EKF algorithm, the AEKF algorithm and the H∞ algorithm.

9. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 1, characterized in that: Step S4 specifically includes the following steps: S401: Calculate the deviation between the predicted value and the measured value of the terminal voltage output by the i-th observer at time k; in, Where: r k,i Indicates the deviation between the predicted value and the measured value of the terminal voltage; U k,i represents the measured value of the terminal voltage at time k; is the terminal voltage prediction value at time k output by the i-th observer; S402: Calculate the terminal voltage residual mean and the terminal voltage residual mean square error of each observer at time k based on the deviation between the terminal voltage prediction value and the measured value; in, Where: Represents the mean value of the terminal voltage residual; s k,i represents the residual mean square error of the terminal voltage; M represents the set moving window length; S403: Calculate the conditional probability density function of each observer at time k based on the terminal voltage residual mean and the terminal voltage residual mean square error; in, Where: represents the conditional probability density function; α i represents the i-th observer; S404: Calculating the weighted value of each observer at time k based on the conditional probability density function of each observer at time k; in, Where: k,i represents the weighted value of the i-th observer at time k; n represents the number of observers.

10. The power battery multi-algorithm fusion SOC estimation method based on Bayesian probability as claimed in claim 9, characterized in that: In step S5, the fused SOC estimation value is calculated by the following formula: Where: Z F,k Indicates the fusion SOC value; Z k,i represents the SOC estimation value output by the i-th observer at time k.

Images