A method for estimating the health status of lithium-ion batteries based on data model fusion

Abstract

The present invention discloses a method for estimating the health status of lithium-ion batteries based on data model fusion. In order to accurately estimate the health status of the battery, the present invention extracts all data of the entire life cycle of batteries of the same type as the battery to be estimated and that have been fully aged, and uses the Gaussian process regression method to fit the equal-time terminal voltage rebound and capacity mapping relationship after charging is completed, and constructs the observation equation of battery aging. The historical capacity data and cycle number of the battery to be estimated are extracted, and a multi-factor gray prediction model is used to construct the state transfer equation of battery aging. The state transfer equation and observation equation of battery aging are integrated to construct a data-driven state space model that characterizes battery aging. The equal-time terminal voltage rebound after charging of the battery to be estimated is extracted in real time, and the particle filtering method is used to track and update the changes in the battery health status online.

Description

Technical Field

[0001] The present invention belongs to the technical field of battery management and relates to a method for estimating the health status of a lithium-ion battery based on data model fusion. Background Art

[0002] Lithium-ion batteries are widely used in various fields such as consumer electronics and electric vehicles due to their advantages such as high operating voltage, high energy power density, long cycle life, and low environmental pollution. However, as the number of battery cycles increases, safety accidents caused by the inability of battery management systems to accurately determine the health status of lithium ions are common. Therefore, accurately estimating the health status of lithium-ion batteries is of great significance to ensuring the safety of lithium-ion batteries and equipment equipped with lithium-ion batteries.

[0003] The health status of lithium-ion batteries is a quantitative description of the ability of lithium-ion batteries to store electrical energy or energy relative to new batteries in percentage form. The current mainstream battery health status estimation methods mainly include model methods and data-driven methods. The model method establishes a battery equivalent circuit model or electrochemical model, estimates the battery capacity or impedance through model parameters, and then estimates the battery health status. However, the estimation accuracy of the model method is strongly dependent on the accuracy of the battery model. The data-driven method regards lithium-ion batteries as "black boxes" and directly mines the potential relationship between battery aging data and battery health status through intelligent algorithms to achieve battery health status estimation. Compared with the model method, the data-driven method does not need to understand the internal mechanism of the battery, and the prediction accuracy is not limited by the battery model, which has received more and more attention. The data-driven method needs to process a large amount of battery aging data, which consumes a lot of computational consumption, and it is difficult to guarantee generalization performance when the data volume is scarce or the data quality is poor. Summary of the invention

[0004] In view of the shortcomings of the above-mentioned background technology, this application proposes a lithium-ion battery health status estimation method based on data model fusion, establishes a state space model to characterize the aging of lithium-ion batteries based on a data-driven method, and combines a particle filtering algorithm to realize online estimation of the health status of lithium-ion batteries.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme:

[0006] A method for estimating the health state of a lithium-ion battery based on data model fusion, comprising: constructing an observation equation A and a state transfer equation B for battery aging; constructing a particle filter framework C for estimating battery capacity; and merging equations A and B to construct a data-driven state space model for characterizing battery aging.

[0007] The construction of the observation equation A for battery aging comprises the following steps:

[0008] Step 1: Extract all the data of the entire life cycle of batteries of the same type as the battery to be estimated and that have been fully aged, and perform preprocessing to remove outlier data;

[0009] Step 2: Extract characteristic data and battery capacity test data of the battery that has completed the aging process in step 1 after each charge;

[0010] Step 3: Based on the characteristic data in step 2 and the battery capacity test data, offline training is performed using a data-driven method to obtain the correlation characteristic data of multi-dimensional parameters of the same type of batteries and construct the observation equation y for battery aging. k =h(x k )+ω k ;

[0011] The state transition equation B for constructing the battery aging comprises the following steps:

[0012] Step 4: Extract the historical capacity data and battery cycle data of the battery to be estimated, perform preprocessing and remove outlier data as prediction sample data;

[0013] Step 5: Use the predicted sample data obtained in step 4 as the input data sequence of the data-driven method to construct the state transition equation x for battery aging k =f(x k-1 ,cycle)+v k , calculate the battery capacity of the current cycle;

[0014] The construction of the particle filter battery capacity estimation framework C includes the following steps:

[0015] Step 6: Initialize the particle filter algorithm, particle number, probability density distribution, and resampling strategy;

[0016] Step 7: Predict the prior particle distribution of the battery capacity data under the current cycle using the state transfer equation in step 5;

[0017] Step 8: extract characteristic data of the battery to be estimated in the current cycle after charging is completed, and use the obtained battery characteristic data to update the prior particle distribution of the battery capacity estimation data obtained in step 7 with the observation equation in step 3 to obtain a corrected battery capacity posterior particle distribution;

[0018] Step 9: Determine whether to resample the capacity posterior particle distribution obtained in step 8;

[0019] Step 10: Perform particle distribution on the resampled data in step 9 and calculate the mean value of the battery capacity particle distribution;

[0020] Step 11: Based on the definition of the battery health state, the battery health state and probability density distribution at the current moment are calculated using the final capacity data obtained and the battery capacity particle distribution in step 8 to obtain the final battery capacity estimation data;

[0021] Step 12: Output the estimated battery health status;

[0022] Step 13: Determine whether the estimation is completed; if the determination result is yes, execute step 10; if the determination result is no, skip step 10 and execute step 11;

[0023] Repeat step 7 to step 12 to iteratively calculate and output the battery health status and probability density distribution to be estimated.

[0024] Furthermore, the characteristic data in step 2 is the equal-time terminal voltage rebound after charging is completed, and the specific time is selected by the user according to the sampling frequency of the charging device.

[0025] Furthermore, the data driven method in step 3 is Gaussian process regression.

[0026] Furthermore, the data driven method in step 5 is a multi-factor GM (1, N) grey prediction model.

[0027] Compared with the existing lithium-ion battery health status estimation method, the present invention has the following advantages:

[0028] 1. The data-driven method proposed in the present invention is used to construct the battery aging state space equation, which is generalizable to different batteries and solves the problem of no effective battery aging model.

[0029] 2. Particle filtering is combined with the state-space equation of battery aging to build a closed-loop estimation framework to achieve battery capacity degradation tracking and health status estimation, effectively avoid the interference of outliers, and improve the accuracy and robustness of health status estimation.

[0030] Figure 1 Schematic diagram of a lithium-ion battery health status estimation method based on data model fusion of the present invention DETAILED DESCRIPTION

[0031] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0032] As attached Figure 1 As shown, the present invention proposes a lithium-ion battery health state estimation method based on data model fusion to accurately estimate the health state of lithium-ion batteries. The state space model of battery aging is established by using a data-driven method, and the battery capacity degradation tracking and health state estimation are realized by combining adaptive filtering. The method specifically includes the following steps:

[0033] Step 1: Construct the observation equation A of battery aging;

[0034] First, extract all the data of the entire life cycle of batteries of the same type as the battery to be estimated and that have been fully aged, and perform preprocessing to remove outlier data;

[0035] Then, based on the battery data set of the completed aging process obtained after preprocessing, the characteristic data and battery capacity test data after each charging is completed are extracted;

[0036] Finally, based on the obtained characteristic data and battery capacity test data, offline training is performed using the Gaussian process regression method to obtain the correlation characteristic data of multidimensional parameters of the same type of batteries and construct the observation equation y for battery aging. k =h(x k )+ω k ;

[0037] Step 2: Construct the state equation B of battery aging;

[0038] First, the historical capacity data and battery cycle data of the battery to be estimated are extracted, and then preprocessed and outlier data are removed as prediction sample data;

[0039] Then, the obtained prediction sample data is used as the input data sequence of the grey prediction model to construct the state transition equation x of battery aging. k =f(x k-1 ,cycle)+v k , calculate the battery capacity of the current cycle;

[0040] Finally, the observation equation A and the state equation B are combined to construct the state space equation of battery aging.

[0041] Step 3: Construct a particle filter framework C for estimating battery capacity:

[0042] First, initialize the particle filter algorithm, particle number, probability density distribution, and resampling strategy;

[0043] Second, the prior particle distribution of the battery capacity data under the current cycle is predicted using the state equation B;

[0044] Third, extract the characteristic data of the battery to be estimated in the current cycle after charging is completed, and use the obtained battery characteristic data observation equation A to update the prior particle distribution of the battery capacity estimation data obtained to obtain the corrected battery capacity posterior particle distribution;

[0045] Fourth, determine whether to resample the obtained capacity posterior particle distribution;

[0046] Fifth, perform particle distribution on the resampled data and calculate the mean value of the battery capacity particle distribution;

[0047] Sixth, based on the definition of battery health status, the final capacity data is obtained; at the same time, the battery health status and probability density distribution at the current moment are calculated according to the battery capacity particle distribution to obtain the final battery capacity estimation data;

[0048] Seventh: Determine whether the battery has reached the end of its life and decide to enter the next battery charging cycle; iteratively calculate the battery health status and probability density distribution to be estimated.

Claims

1. A method for estimating the health status of a lithium-ion battery based on data model fusion, characterized in that: include: Constructing an observation equation A for battery aging, constructing a state transfer equation B for battery aging, and constructing a particle filter framework C for estimating battery capacity; the equation A and equation B are integrated to construct a data-driven state space model that characterizes battery aging; wherein constructing the observation equation A for battery aging includes the following steps: Step 1: Extract all the data of the entire life cycle of batteries of the same type as the battery to be estimated and that have been fully aged, and perform preprocessing to remove outlier data; Step 2: Extract characteristic data and battery capacity test data of the battery that has completed the aging process in step 1 after each charge; Step 3: Based on the characteristic data in step 2 and the battery capacity test data, offline training is performed using a data-driven method to obtain the correlation characteristic data of multi-dimensional parameters of the same type of batteries and construct the observation equation y for battery aging. k =h(x k )+ω k ; The state transition equation B for constructing the battery aging comprises the following steps: Step 4: Extract the historical capacity data and battery cycle data of the battery to be estimated, perform preprocessing and remove outlier data as prediction sample data; Step 5: Use the predicted sample data obtained in step 4 as the input data sequence of the data-driven method to construct the state transition equation x for battery aging k =f(x k-1 ,cycle)+v k , calculate the battery capacity of the current cycle; The construction of the particle filter battery capacity estimation framework C includes the following steps: Step 6: Initialize the particle filter algorithm, particle number, probability density distribution, and resampling strategy; Step 7: Predict the prior particle distribution of the battery capacity data under the current cycle using the state transfer equation in step 5; Step 8: extract characteristic data of the battery to be estimated in the current cycle after charging is completed, and use the obtained battery characteristic data to update the prior particle distribution of the battery capacity estimation data obtained in step 7 with the observation equation in step 3 to obtain a corrected battery capacity posterior particle distribution; Step 9: Determine whether to resample the capacity posterior particle distribution obtained in step 8; Step 10: Perform particle distribution on the resampled data in step 9 and calculate the mean value of the battery capacity particle distribution; Step 11: Based on the definition of the battery health state, the battery health state and probability density distribution at the current moment are calculated using the final capacity data obtained and the battery capacity particle distribution in step 8 to obtain the final battery capacity estimation data; Step 12: Output the estimated battery health status; Step 13: Determine whether the estimation is completed; If the judgment result in step 9 is yes, execute step 10; if the judgment result in step 9 is no, skip step 10 and execute step 11; If the judgment result in step 13 is yes, the calculation is terminated; if the judgment result in step 13 is no, steps 7 to 12 are repeated to iteratively calculate and output the battery health status and probability density distribution to be estimated.

2. A lithium-ion battery health status estimation method based on data model fusion according to claim 1, characterized in that: The characteristic data in step 2 is the equal time terminal voltage rebound after charging is completed, and the specific time is selected by the user according to the sampling frequency of the charging device.

3. The method for estimating the health status of a lithium-ion battery based on data model fusion according to claim 1, characterized in that: The data driven method in step 3 is Gaussian process regression.

4. The method for estimating the health status of a lithium-ion battery based on data model fusion according to claim 1, characterized in that: The data driven method in step 5 is a multi-factor GM (1, N) grey prediction model.

Images