Lithium battery SOC estimation method driven by data behavior space and EKF update

Abstract

The present invention discloses a lithium battery SOC estimation method driven by data behavior space and EKF update. This method utilizes multi-source input and output data during battery operation, compresses the data through a polygonal event-triggered resampling method, and then uses the Willems theorem method to construct a data behavior space of the battery's dynamic behavior to achieve online estimation of SOC, avoiding the existing method's reliance on explicit state space models and parameter identification, and overcoming the parameter mismatch problem. At the same time, based on the estimation results of the data behavior space state, an extended Kalman filter algorithm based on the data behavior space and the Koopman operator is dynamically constructed, and the prior estimate of the extended Kalman filter algorithm of the Koopman operator is used as a reference trajectory for the rolling time domain optimization of the behavior space prediction. This method is suitable for dynamic high-noise scenarios during data-driven behavior trajectory prediction, and can improve the accuracy and real-time performance of battery SOC estimation.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical data processing and relates to a lithium battery SOC state estimation method, and specifically to a lithium battery SOC estimation method driven by data behavior space and EKF update. Background Art

[0002] Lithium batteries, with their high energy density and wide operating temperature range, have been widely used in systems such as electric vehicles and energy storage. Accurately estimating the state of charge (SOC) of lithium batteries can be used to monitor their long-term performance and optimize the management of available energy, thereby avoiding overcharging and overdischarging and extending the battery's cycle life.

[0003] Current SOC estimation methods for lithium battery systems primarily include the ampere-hour method, the open-circuit voltage function method, state observer-based estimation methods, and intelligent estimation methods. The ampere-hour method requires integrating the battery current and relies on the initial battery capacity. This can lead to high error in scenarios with high noise levels and inaccurate initial parameter settings. The open-circuit voltage function method estimates SOC by mapping the battery's open-circuit voltage to SOC. However, this method requires open-circuit voltage data after the battery has been stationary for a sufficient period of time to reach steady state, making it inapplicable under dynamic online conditions. State observer-based estimation methods, such as the Kalman filter, rely on modeling the battery's external voltage and current characteristics. This can lead to model parameter drift when addressing battery aging and capacity degradation, potentially resulting in significant estimation errors. Intelligent estimation methods typically employ deep neural networks. While these methods avoid the modeling challenges of state observers, they require extensive historical data to learn the dynamic characteristics of the battery's SOC. Furthermore, neural network methods lack physical interpretability, and parameter tuning requires significant manual intervention to improve estimation accuracy.

[0004] In recent years, in order to improve the estimation accuracy of battery SOC, some hybrid methods have been proposed, such as the joint estimation method based on multi-dimensional physical field coupling model and state observer, and the hybrid estimation method based on data-driven and state observer. However, they all rely on accurate lithium battery models. When adapting to complex working conditions, they are prone to problems such as model calculation complexity explosion and slow response time, making it difficult to complete the task of online real-time SOC estimation.

[0005] Therefore, compared with the current hybrid model based on explicit state-space modeling, there is an urgent need for a new lithium battery SOC online estimation method that does not rely on explicit state-space modeling. Under the premise of ensuring physical interpretability, it can solve the shortcomings of the explicit state-space model in the multi-dimensional physical state quantity coupling scenario, such as dependence on model parameter accuracy, real-time estimation response delay and weak adaptability to noise environment, and provide a more reliable and efficient online real-time estimation solution for the battery management and control system. Summary of the Invention

[0006] In response to the defects of the existing technology such as strong model and parameter dependence, poor real-time performance of data-driven methods and insufficient physical interpretability, the present invention proposes a lithium battery SOC estimation method driven by data behavior space and EKF update. It triggers the resampling of multi-source input and output data in battery operation through polygonal events, and then uses the Willems theorem method to construct an implicit representation space of battery dynamic behavior to realize online SOC estimation, avoiding the existing method's dependence on explicit state space models and parameter identification, and overcoming the parameter mismatch problem; at the same time, it integrates the estimation method based on state observer, uses the prior estimation value as the reference trajectory for rolling time domain optimization of behavior space prediction, adapts to the dynamic high-noise scenario during data-driven behavior trajectory prediction, and improves the accuracy of battery SOC estimation and the real-time performance of the algorithm.

[0007] The lithium battery SOC estimation method driven by data behavior space and EKF update includes the following steps:

[0008] Step 1: Collect raw data of lithium batteries

[0009] Collect multi-dimensional raw feature data of lithium batteries, including sampling timestamp, charging state, battery voltage, battery current, SOC, temperature, maximum voltage, minimum voltage, maximum current, maximum voltage, maximum temperature and minimum temperature, and set it as the raw data set .

[0010] Step 2: Resampling triggered by non-full data polygon events

[0011] Based on the original dataset in step 1 , set the valid value range of each feature data, preprocess the data, and perform event-triggered resampling for subsequent generation of input and output sequences for trajectory prediction:

[0012] S2.1. Outlier Removal and Missing Value Processing

[0013] For the original dataset Clean the data in the dataset, remove outliers that are beyond the valid value range, and use the mean or interpolation method to fill in missing values to ensure data integrity. Finally, delete the duplicate sample rows to obtain the preprocessed dataset. .

[0014] S2.2. Resampling triggered by non-full data polygon events

[0015] Most lithium-ion battery voltages change rapidly when the SOC values are between 0% and 20% and between 80% and 100%, but change more slowly when the SOC is between 20% and 80%. Temperature also exhibits similar characteristics under different operating conditions and durations. For example, for every 10°C increase in temperature, the SOC range in the fast-changing region expands by approximately 5%. Therefore, improperly selecting the sampling interval can result in the loss of a significant amount of critical data in the fast-changing region or the redundant collection of a significant amount of invalid data in the slow-changing region.

[0016] In view of the multi-time scale variation characteristics of lithium battery voltage in different SOC states, which have fast-changing and slow-changing regions, a non-full data polygon event trigger mechanism is designed to selectively sample the data and obtain a resampled data set. :

[0017]

[0018]

[0019] In the formula, h=1,2,3,4, represents the dimension identifier of the feature data, Represents the feature data of the hth dimension, , which represents the dataset Middle The data value at a time point, , 、 Respectively represent the i-th and i+1-th event triggering resampling moments, 、 They represent the values of the h-th dimension resampled at the i-th and i+1-th event triggers, respectively. Characteristic data The maximum value in Characteristic data The minimum value of It is an event-triggered scale variable that is dynamically adjusted according to actual needs. Representing feature data Event trigger thresholds, including battery current event trigger resampling thresholds , battery SOC event triggers resampling threshold , battery voltage event triggers resampling threshold and battery temperature event trigger resampling threshold .

[0020] S2.3. Resampling correction based on interpolation data error

[0021] Although the non-full data polygon event trigger mechanism improves the sampling efficiency, data may still be lost at the inflection point between the fast-changing area and the slow-changing area. Therefore, it is necessary to perform data resampling correction based on step S2.2, and interpolate and reconstruct the resampled data so that its time axis is aligned with the original data to form the reconstructed feature data. , and then find the difference between the reconstructed data and the original data. If the difference at time j is Greater than the proposed threshold , then the original data at time j Add resampled dataset To ensure the accuracy of subsequent estimates:

[0022] | - |

[0023]

[0024] in, represents the reconstructed feature data of the h-th dimension at the j-th time point, and || represents the absolute value calculation. It is the resampling precision variable, which is adjusted according to the actual estimation accuracy requirement.

[0025] Step 3: Constructing the data behavior space based on Willems theorem

[0026] Based on Willems' theorem, using resampled data sets Construct a Data Behavior Space (DBS) to implicitly describe the dynamic behavior of lithium batteries:

[0027] S3.1. Constructing Hankel data behavior space matrix

[0028] Using resampled data The process of constructing the Hankel matrix is as follows:

[0029]

[0030]

[0031] in 、 Indicates input current The resampled data; 、 Indicates output voltage , state of charge and temperature The resampled data, and are the input and output dimensions respectively, , ; , the superscript T represents the matrix transpose, represents the vector consisting of the input current In the The value at a time point, , Indicates the length of the resampled data; = , Indicated by the output voltage , state of charge and temperature Composed of The output matrix In the A column vector of time points, Indicates the length of the past moment. Indicates the length of a future moment.

[0032] S3.2, According to Willems's basic lemma, under the condition of sufficient excitation, any length is Input and output traces Can be expressed as a linear combination of the column vectors of the corresponding Hankel matrix in the historical data, that is, there is a coefficient vector So that:

[0033]

[0034] in is the known initial input and output trajectory, is the predicted future trajectory. The coefficient vector describing the future dynamic behavior of the system is obtained through the above linear equations , and then implicitly characterizes the Data-Behavior Space (DBS). This method can directly utilize historical input and output data without explicitly identifying physical model parameters, thus avoiding the parameter mismatch and nonlinear modeling difficulties in traditional physical modeling.

[0035] Step 4: Reference trajectory rolling optimization

[0036] This paper proposes an extended Kalman filter (EKF) algorithm that fuses the data behavior space (DBS) and the Koopman operator. By utilizing the nonlinear filtering characteristics of the EKF, multi-source measurement data such as battery current, voltage, and temperature are integrated in the lithium battery management and control system. The Koopman operator is then used to construct an implicit state space. The EKF reference trajectory rolling optimization is then achieved by fusing the data behavior space. The specific process is as follows:

[0037] S4.1. Initialization and key parameter construction

[0038] The coefficient vector describing the future dynamic behavior of the system at the kth time point Define the state vector at the kth time point , the EKF algorithm is improved based on the data behavior space, and the discrete dynamic equation of the lithium battery system is expressed as:

[0039]

[0040]

[0041]

[0042] Where, and represent the process noise and observation noise at time k respectively. are the nonlinear state function and measurement function of the system respectively. represents the observation value at time k.

[0043] Based on Koopman operator Constructing the data behavior space of output data with the data behavior space method , redefine the state transfer equation:

[0044]

[0045]

[0046] Where, , .

[0047] Define the initial covariance matrix using the data behavior space 、 、 :

[0048]

[0049]

[0050]

[0051] By solving the following minimization problem, we can obtain the finite-dimensional approximation operator :

[0052]

[0053] Where, Represents the square of the Euclidean norm.

[0054] S4.2. Operator-based Predictions and updates

[0055] The filtering process is divided into two stages: prediction and update. In the prediction stage, according to the operator M obtained by solution and the previous posterior observation value and the posterior covariance matrix , predict the current prior observation and the prior covariance matrix :

[0056]

[0057]

[0058] when hour, , , represents the covariance.

[0059] In the update phase, according to the prior observations , prior covariance matrix and the current observation , estimate the posterior observation and the posterior covariance matrix :

[0060]

[0061]

[0062]

[0063] in, By adjusting the estimated value online, the error caused by noise or parameter uncertainty is reduced and the accuracy of SOC estimation is further improved.

[0064] Step 5: Parameter update and online estimation

[0065] Combined with Observe the updated data at all times, use the incremental update learning method to learn the information of the updated data, and combine the data behavior space to give the state parameters of the lithium battery and further estimated values online, as follows:

[0066] S5.1. Data Behavior Space Prediction Update Method

[0067] Based on the data behavior space construction method proposed in step 3, according to the input of the lithium battery at time k With output , solve the following minimization problem and obtain the coefficient vector at time k :

[0068]

[0069] in is the prescribed reference trajectory for future output. ,matrix is a square matrix composed of penalty coefficients, is a cost function for controlling an action:

[0070]

[0071] in , is the regularization parameter, represents the output prediction estimate at time k obtained at time k-1, ,in is a constant.

[0072] S5.2. Update the estimated value of the extended Kalman filter algorithm that integrates the data behavior space and the Koopman operator:

[0073] The k-time parameter vector obtained by solving S5.1 Substitute into the discrete dynamic equation of the lithium battery system established in step 4 and solve Operators and prior observations , finally, through The posterior observation obtained by the operator filtering Mapping is performed to obtain the estimated output value of the lithium battery at time k to time k+1 :

[0074]

[0075] The predicted estimated value The SOC information component in the final lithium battery SOC online estimation result is used as the Add candidates to the initial input and output trajectories and use them for future prediction updates of battery trajectories.

[0076] The present invention has the following beneficial effects:

[0077] 1. The proposed non-full data polygonal data resampling technology compresses and resamples the data through a dynamic event trigger mechanism, and incorporates the error between the resampled data and the original data after re-expansion into the event trigger mechanism. This reduces the complexity of solving subsequent optimization problems and the error caused by event-triggered sampling, achieving more efficient data compression, and enabling lithium battery SOC estimation to still have high computational efficiency under complex working conditions.

[0078] 2. To address performance changes in lithium batteries caused by factors such as overcharging or over-discharging, we propose an online SOC estimation method based on the implicit state space and Koopman operator constructed using Willems' theorem, independent of the accuracy of explicit state space estimation model parameters. Compared to existing data-model hybrid SOC estimation methods, this overcomes the limitations of explicit state space modeling methods, such as strong model dependence, difficulty in parameter identification, and poor real-time performance of data-driven methods. This improves the accuracy and adaptability of the estimation, enabling more precise estimation of the real-time SOC of lithium batteries and enhancing robustness. This provides technical support for the efficient, safe, and optimized operation of lithium battery management and control systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 Flowchart of the lithium battery SOC estimation method driven by data behavior space and EKF update.

[0080] Figure 2 For the resampled dataset Schematic diagram of input and output data.

[0081] Figure 3 Schematic diagram of the resampling results triggered by polygon events.

[0082] Figure 4 Schematic diagram of the event-triggered resampling mechanism for voltage data;

[0083] Figure 5 Schematic diagram of spatial prediction of data behavior based on polygon event-triggered resampling.

[0084] Figure 6 Schematic diagram of parameter update. DETAILED DESCRIPTION

[0085] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific implementation methods. It should be noted that the embodiments and technical details described in this specification are intended to exemplarily illustrate the core concept of the present invention and do not constitute a limitation on the scope of protection of the claims. It should be pointed out in particular that all implementation methods that can be derived by those skilled in the art through conventional experimental means or customary technical replacements based on the technical concept of the present invention without departing from the principles of the invention, as well as all other embodiments obtained through equivalent transformations without creative work, should be deemed to fall within the scope of protection of the claims of the present invention.

[0086] like Figure 1 As shown, the lithium battery SOC estimation method driven by data behavior space and EKF update includes the following steps:

[0087] Step 1: Use the experimental data from Stanford University's "Experimental data of lithium-ion batteries under galvanostatic discharge tests at different rates ad temperatures of operation" as the original dataset , including sampling timestamp, charging status, battery voltage, battery current, SOC, temperature, maximum voltage, minimum voltage, maximum current, maximum voltage, maximum temperature and minimum temperature data.

[0088] Step 2: Resampling triggered by non-full data polygon events

[0089] Step 2.1: Original dataset Clean the data in the dataset, remove outliers that exceed the set valid value range, and use the mean or interpolation method to fill missing values to ensure data integrity. Finally, delete duplicate sample rows to obtain the preprocessed dataset. .

[0090] Step 2.2: If Figure 2 As shown, for the preprocessed dataset Timing characteristic data of voltage, current, SOC, and temperature in , respectively set the event trigger threshold , based on Figure 3 The timing data of the non-full polygon event triggering mechanism is shown Resample to obtain a resampled data set :

[0091]

[0092]

[0093] In the formula, h=1,2,3,4, represents the dimension identifier of the feature data, Represents the feature data of the hth dimension, , which represents the dataset Middle The data value at a time point, , 、 Respectively represent the i-th and i+1-th event triggering resampling moments, 、 They represent the values of the h-th dimension resampled at the i-th and i+1-th event triggers, respectively. Characteristic data The maximum value in Characteristic data The minimum value of is the event trigger scale variable, which is set to 100 in this embodiment. Representing feature data Event trigger thresholds, including battery current event trigger resampling thresholds , battery SOC event triggers resampling threshold , battery voltage event triggers resampling threshold and battery temperature event trigger resampling threshold . Figure 4 Schematic diagram of event-triggered sampling of voltage data.

[0094] Step 2.3: Resampling correction based on interpolation data error

[0095] By linear interpolation, the data resampled by the event trigger mechanism in step 2.2 is expanded so that its time length is consistent with the preprocessed data set. Match, find the difference between the interpolated reconstructed data and the original data, if the difference at time j Greater than the proposed threshold , then the original data at time j Add resampled dataset middle:

[0096] | - |

[0097]

[0098] in, represents the reconstructed feature data of the h-th dimension at the j-th time point, and || represents the absolute value calculation. is the resampling accuracy variable, which is set to 2% in this embodiment.

[0099] Step three, such as Figure 5 As shown in Figure 1, the data set constructed by resampling is used to construct the data behavior space of the lithium battery state based on Willems theorem, and then the battery state is estimated online.

[0100] Step 3.1: Dataset constructed using resampling , record the battery current data as input , SOC, temperature and voltage are recorded as output , construct the Hankel matrix for subsequent construction of the Data Behavior Space (DBS) for trajectory prediction:

[0101]

[0102]

[0103] in 、 Indicates input current The resampled data; 、 Indicates output voltage , state of charge and temperature The resampled data, and are the input and output dimensions respectively, , ; , the superscript T represents the matrix transpose, represents the vector consisting of the input current In the The value at a time point, , Indicates the length of the resampled data; = , Indicated by the output voltage , state of charge and temperature Composed of The output matrix In the A column vector of time points, Indicates the length of the past moment. Indicates the length of a future moment.

[0104] Step 3.2: Use the constructed Hankel matrix to construct the data behavior space , according to Willems's basic lemma, under the condition of sufficient excitation, there exists a vector So that:

[0105]

[0106] in The known initial input and output trajectory is used to predict the future trajectory Set the initial conditions.

[0107] By solving the above linear equations, we can obtain the coefficient vector describing the future dynamic behavior of the system: , and then implicitly characterizes the data behavior space.

[0108] Step 4: Reference trajectory rolling optimization

[0109] In order to further optimize the efficiency and accuracy of data behavior space prediction, an extended Kalman filter method based on the Koopman operator and data behavior space is introduced. The EKF is reconstructed under a data-driven framework. The nonlinear filtering characteristics of the EKF are utilized to fuse multi-source measurement data, and then the implicit state space is constructed through the Koopman operator. The rolling optimization of the EKF reference trajectory is achieved by fusing the data behavior space.

[0110] Step 4.1: Express the discrete dynamic equations of the lithium battery system as:

[0111]

[0112]

[0113] The state vector This is the coefficient vector describing the future dynamic behavior of the system in step 3 , are the nonlinear state function and measurement function of the system respectively. and are the process noise and observation noise at time k respectively. represents the observation value at time k.

[0114] Define the Koopman operator , for the above lithium battery system, use The operator acts on the function of the state space:

[0115]

[0116] Thus based on The operator reconstructs the dynamic equation of the lithium battery system to make it more suitable for the Kalman filter method:

[0117]

[0118]

[0119]

[0120] in , is the process noise, obey The normal distribution of is the observation noise, obey The normal distribution of , , 、 is the covariance matrix at time k.

[0121] By solving the following minimization problem, we can find the finite-dimensional approximation operator :

[0122]

[0123] Where, Represents the square of the Euclidean norm.

[0124] Step 4.2: Divide the filtering process into two stages: prediction and update.

[0125] In the prediction phase, based on the previous posterior observations and the posterior covariance matrix Predict the current prior observation and the prior covariance matrix :

[0126]

[0127]

[0128] when hour , . Wherein, the superscript T represents the matrix transpose, represents the covariance.

[0129] In the update phase, according to the prior observations , prior covariance matrix and the current observation Estimating the posterior observation and the posterior covariance matrix :

[0130]

[0131]

[0132]

[0133] in, By adjusting the estimated value online, the error caused by noise or parameter uncertainty is reduced and the accuracy of SOC estimation is further improved.

[0134] Step 5: Parameter update and online estimation

[0135] like Figure 4 As shown, combined with Observe the updated data at all times, use the incremental update learning method, only learn the information of the updated new data, and combine the data behavior space to give the state parameters of the lithium battery and further estimated values online.

[0136] Step 5.1: Based on the data behavior space method proposed in step 3, at time k, according to the input of the lithium battery With output Solve the following minimization problem to obtain the coefficient vector at time k :

[0137]

[0138] in It is the prescribed reference trajectory of future output, consisting of the terminal voltage calculated by the OCV-SOC function given by the battery manufacturer and the value observed by the sensor at the current moment. , where the matrix It is a square matrix composed of penalty coefficients, which is set as the unit matrix in this embodiment. is a cost function for controlling an action:

[0139]

[0140] in , is the regularization parameter, is the output prediction estimate at time k obtained at time k-1, In this example, Set to 0.1.

[0141] Step 5.2: Solve the k-time parameter vector obtained in S5.1 Substitute into the discrete dynamic equation of the lithium battery system established in step 4 and solve Operators and prior observations , finally, through The posterior observation obtained by the operator filtering Mapping is performed to obtain the output prediction estimate at time k for time k+1 , the predicted estimated value The SOC information component in is used as the final lithium battery SOC online estimation result:

[0142]

[0143] The denoised estimate The SOC information component in the final lithium battery SOC online estimation result is used as the Add candidates to the initial input and output trajectories and use them for future prediction updates of battery trajectories.

[0144] Among the output data of lithium batteries, in addition to SOC, voltage and temperature Both with input current Existence relationship:

[0145]

[0146]

[0147] in is the battery open circuit voltage, is the equivalent impedance of the battery, represents the gradient calculation, Representing irreversible resistive heating and reversible entropy heating, respectively. This method uses both voltage and temperature as output trajectories driven by the behavior space when predicting lithium battery SOC. This allows SOC estimation to integrate voltage and temperature models, rather than relying solely on single-dimensional observations. This mitigates error accumulation, improves error correction capabilities, and ultimately enhances prediction accuracy.

[0148] This embodiment addresses the problem of accurate state estimation of lithium battery systems under noisy conditions, describes in detail the specific implementation details of the lithium battery SOC estimation method driven by data behavior space and EKF update, and proposes a complete solution implementation plan including data behavior space construction, noise filtering and joint estimation. By introducing non-full event triggered resampling, the data is compressed to improve the computational efficiency of the algorithm; by constructing the data behavior space and designing the filtering algorithm, the anti-interference performance in uncertain environments is significantly improved. The SOC of lithium batteries is estimated by this method with an error of less than 2%. At the same time, the traditional extended Kalman filter method requires model parameter identification, and the parameters identified are prone to parameter mismatch, and the estimated SOC has a certain error. This method adopts a data-driven idea and establishes an implicit data behavior space based on the actual input and output data of the lithium battery. It not only effectively avoids the identification error caused by parameter mismatch, but also solves the problem of the lack of physical interpretability of the traditional data-driven method. Compared with the mainstream unscented Kalman filter, the maximum SOC prediction error of this method can be reduced by 59%. It should be noted that the above content merely illustrates the technical idea of the present invention and cannot be used to limit the scope of protection of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications all fall within the scope of protection of the claims of the present invention.

Claims

1. The lithium battery SOC estimation method driven by data behavior space and EKF update collects the input and output multi-dimensional feature data of the lithium battery to form the input trajectory u d and output trajectory y d Based on the data-driven method, the Hankel data behavior space matrix is constructed, and the input u of the lithium battery at time k is calculated. k With the output y k Calculate the coefficient vector g that describes the dynamic behavior of the system at time k k , predicting the lithium battery output data at time k+1, characterized by: The input trajectory u of the lithium battery d is the current data, output trajectory y d Including voltage, SOC and temperature data; For a length of T l The input trajectory u d and output trajectory y d , select the front T ini The data of the first moment is taken as the past moment, and the data of the next N moments is taken as the future moment. The constructed Hankel data behavior space matrix is as follows: in, m and n are the dimensions of the input trajectory and output trajectory respectively, H c =T l -T ini -N+1, L=T ini +N+1; The superscript T indicates the matrix transpose, u k Represents the input trajectory u d The value at the kth time point, k = 0, 1, ... T l -1, y k Represents the output trajectory y d Column vector at the kth time point; According to Willems's fundamental lemma, the coefficient vector g k Expressed as: where u ini 、y ini is the known initial input and output trajectory, u and y are the predicted future trajectories; The coefficient vector g at time k k As the system state x k , use the extended Kalman filter algorithm to predict and update, and estimate the posterior observation value x k|k That is, the coefficient vector g at time k+1 k+1 , and then use the Hankel data behavior space matrix, according to the coefficient vector g at time k+1 k+1 Get the estimated value of lithium battery output at time k for time k+1 Forecast estimate The SOC information component in is the online estimation result of the lithium battery SOC.

2. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 1, characterized in that: The original collected lithium battery input and output data are preprocessed, and the event trigger thresholds of the feature data of each dimension are designed, and the preprocessed feature data are resampled by event triggering.

3. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 2, characterized in that: Set the valid value range of each feature data, remove abnormal values that exceed the valid value range in the original collected data, and use the mean or interpolation method to fill in the missing values. Finally, delete the duplicate sample rows to obtain the preprocessed data.

4. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 2, characterized in that: The following event trigger mechanism is designed to perform event trigger resampling on each feature data after preprocessing to obtain the resampled data set O pd : In the formula, h = 1, 2, 3, 4, represents the dimension identifier of the feature data, x h Represents the feature data of the hth dimension; j∈[j i ,j i+1 ), j i 、j i+1 Respectively represent the i-th and i+1-th event triggering resampling moments, Respectively represent the values of the h-th dimension when the i-th and i+1-th events trigger resampling, max(x h )、min(x h ) are the feature data x h The maximum and minimum values in the scale are the event triggering scale variables; Δx h Represents feature data x h The event trigger threshold.

5. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 4, characterized in that: Interpolate and reconstruct the resampled data so that its time axis is aligned with the original data to form the reconstructed feature data Then find the difference between the reconstructed data and the original data. If the difference e at time j is j Greater than the proposed threshold Δe th , then the original data X at time j j Add the resampled dataset O pd middle.

6. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 1, characterized in that: The coefficient vector g at time k k As the system state x k , in the extended Kalman filter algorithm, the discrete dynamic equation of the lithium battery system is expressed as: x k+1 =f(x k )+w k z k =h(x k )+v k x k =g k Where w k With v k represents the process noise and observation noise at time k respectively; f(·) and h(·) are the nonlinear state function and measurement function of the system respectively; z k represents the observation value at time k.

7. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 6, characterized in that: Define the Koopman operator M and use the operator M to act on the nonlinear state function f(·): f(x) k+1 )=Mf(x k ) Data behavior space of output data constructed based on Koopman operator M and data behavior space method Redefine the state transition equation of the lithium battery system: x k+1 =Mx k +w k from k =Cx k +v k In the formula, z k =[y k-N+1 ,…,y k ], At time k, by solving the following minimization problem, we can obtain the finite-dimensional approximate operator M and implement the prediction and update process of the extended Kalman filter algorithm: Where, Represents the square of the Euclidean norm.

8. The lithium battery SOC estimation method driven by data behavior space and EKF update as claimed in claim 7, characterized in that: The posterior observation value x obtained by filtering with the M operator k|k Mapping is performed to calculate the output prediction estimate at time k for time k+1 9. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 8.

Images