Lithium battery charge state estimation model training method and device and lithium battery charge state estimation method and device

Abstract

The invention relates to a training method and device of a lithium battery state-of-charge estimation model and a lithium battery state-of-charge estimation method and device. The training method comprises the steps of obtaining a voltage time sequence, a current time sequence and an actual state-of-charge time sequence of a lithium battery under different working conditions; estimating a first estimated state-of-charge time sequence of the lithium battery by adopting an extended Kalman filtering algorithm; splicing the voltage time sequence, the current time sequence and the first estimated state-of-charge time sequence under each working condition into a fusion time sequence; intercepting a plurality of sample time sequences from the fusion time sequence by using a sliding window with a preset length; extracting actual charge state data corresponding to the last time step of the sample time sequence from the actual charge state time sequence under the working condition to which each sample time sequence belongs, and taking the actual charge state data as a tag corresponding to the actual charge state data; and training a preset lithium battery charge state estimation model by taking each sample time sequence and the corresponding label as input and output.

Description

Technical Field

[0001] This invention relates to the field of lithium battery technology, and more specifically to a training method and apparatus for a lithium battery state of charge estimation model, and a lithium battery state of charge estimation method and apparatus. Background Technology

[0002] The state of charge (SOC) of a lithium-ion battery is a crucial parameter for measuring its remaining usable capacity and is essential for battery energy management, lifespan prediction, and safety control. Accurately estimating SOC is vital for improving energy efficiency and extending battery life.

[0003] Methods for estimating the State of Charge (SOC) of lithium-ion batteries mainly include model-driven and data-driven approaches. Model-driven methods include mechanistic models, electrochemical models, and equivalent circuit models. However, due to the complexity of the internal workings of lithium-ion batteries, these models struggle to accurately reflect the actual battery state, resulting in low accuracy. Data-driven methods primarily employ machine learning or deep learning techniques (such as BP neural networks, LSTM, CNN, and GRU) to construct nonlinear mapping relationships and learn the estimation patterns of SOC from measurement data. However, these methods do not incorporate prior physical information, thus their estimation accuracy is also not high. Summary of the Invention

[0004] The purpose of this invention is to provide a training method and apparatus for a lithium battery state of charge estimation model, and a lithium battery state of charge estimation method and apparatus, which combine model-driven methods with data-driven methods to improve the estimation accuracy of SOC.

[0005] To achieve the above objectives, the present invention provides a training method for a lithium battery state of charge estimation model, comprising:

[0006] Acquire discharge data of lithium batteries, including voltage time series, current time series and actual state of charge time series of lithium batteries under different operating conditions;

[0007] Based on the discharge data of lithium batteries, the extended Kalman filter algorithm is used to estimate the state of charge of lithium batteries, and the first estimated state of charge time series of lithium batteries under different operating conditions is obtained.

[0008] The voltage time series, current time series and first estimated state of charge time series of the lithium battery under each operating condition are aligned by time step and spliced ​​into a fused time series. Each data point of the fused time series includes the voltage data, current data and first estimated state of charge data of the corresponding time step.

[0009] Set a sliding window of a preset length, which is less than the length of the fused time series under each operating condition. For the fused time series under each operating condition, the sliding window takes the starting data point of the fused time series under that operating condition as the starting point and uses the sliding window to extract a sample time series from the fused time series under that operating condition. The sample time series includes multiple consecutive data points with the same number as the preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a sample time series. When the sample time series extracted by the sliding window includes the last data point of the fused time series under that operating condition, the sliding window stops moving to obtain multiple sample time series under that operating condition.

[0010] For each sample time series, extract the actual state of charge data corresponding to the last time step of the sample time series from the actual state of charge time series under the operating condition to which the sample time series belongs, and use it as the label corresponding to the sample time series;

[0011] Using the time series of each sample as input and the label corresponding to each time series as output, the preset lithium battery state of charge estimation model is trained to obtain the trained lithium battery state of charge estimation model.

[0012] Optionally, based on the discharge data of the lithium battery, the extended Kalman filter algorithm is used to estimate the state of charge of the lithium battery, obtaining the first estimated state of charge time series of the lithium battery under different operating conditions, specifically including:

[0013] Establish an equivalent circuit model for a lithium battery;

[0014] Based on the discharge data of lithium batteries, the parameters of the equivalent circuit model are identified to obtain the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition.

[0015] Based on the discharge data of the lithium battery and the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition, the first estimated state of charge time series of the lithium battery under different operating conditions is obtained by iteratively using the extended Kalman filter algorithm.

[0016] Another aspect of the present invention provides a method for estimating the state of charge of a lithium battery, comprising:

[0017] Obtain the voltage and current time series of the lithium battery under test during the discharge process;

[0018] Based on the voltage and current time series of the lithium battery under test during the discharge process, the extended Kalman filter algorithm is used to estimate the state of charge of the lithium battery under test, and the first estimated state of charge time series of the lithium battery under test is obtained.

[0019] The voltage time series, current time series and first estimated state of charge time series of the lithium battery under test during the discharge process are aligned by time steps and spliced ​​into a fused time series.

[0020] A sliding window of a preset length is started from the beginning data point of the fusion time series of the lithium battery under test. A predicted sample time series is extracted from the fusion time series of the lithium battery under test using the sliding window. Then, the sliding window is moved to the right multiple times with a step size of 1. Each time the sliding window moves, a predicted sample time series is extracted. When the sample time series extracted by the sliding window includes the last data point of the fusion time series of the lithium battery under test, the sliding window is stopped to obtain the predicted sample time series of the lithium battery under test. Each predicted sample time series includes multiple consecutive data points with the same number as the preset length.

[0021] Each predicted sample time series of the lithium battery under test is input into a pre-trained lithium battery state of charge estimation model so that the lithium battery state of charge estimation model outputs the predicted state of charge value corresponding to the predicted sample time series; wherein the pre-trained lithium battery state of charge estimation model is obtained by the training method of the lithium battery state of charge estimation model described above.

[0022] The predicted state of charge values ​​corresponding to each predicted sample time series are arranged sequentially according to time steps to form a second estimated state of charge sequence.

[0023] The first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain the weighted fused state of charge time series.

[0024] Optionally, the first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain a weighted fused state of charge time series, specifically including:

[0025] The second estimated state of charge time series is padded with zeros at the beginning so that the length of the padded second estimated state of charge time series is consistent with that of the first estimated state of charge time series.

[0026] The first estimated state of charge time series and the padded second estimated state of charge time series are weighted and fused to obtain the weighted fused state of charge time series.

[0027] Optionally, the weighted and fused state-of-charge time series satisfies the following relationship:

[0028] ,

[0029] ,

[0030] ,

[0031] ,

[0032] Where t is the time step, For the weighted fusion, This is the first estimated state-of-charge time series. This is the second estimated state-of-charge time series. This is the second estimated state-of-charge time series after filling. The corrected weights, The uncertainty of the first estimated state-of-charge time series, The uncertainty of the second estimated state-of-charge time series is given by L, where L is the length of the sliding window. The weights before correction. is the midpoint of the smoothing, and p is the slope control coefficient.

[0033] Another aspect of the present invention provides a training device for a lithium battery state of charge estimation model, comprising:

[0034] The first acquisition module is used to acquire the discharge data of the lithium battery, including the voltage time series, current time series and actual state of charge time series of the lithium battery under different operating conditions.

[0035] The first Kalman filter module is used to estimate the state of charge of the lithium battery based on the discharge data of the lithium battery using the extended Kalman filter algorithm, so as to obtain the first estimated state of charge time series of the lithium battery under different operating conditions.

[0036] The first splicing module is used to align the voltage time series, current time series and first estimated state of charge time series of each operating condition of the lithium battery by time step, and splice them into a fused time series. Each data point of the fused time series includes the voltage data, current data and first estimated state of charge data of the corresponding time step.

[0037] The first extraction module is used to set a sliding window of a preset length, which is less than the length of the fused time series under each working condition. For the fused time series under each working condition, the sliding window takes the starting data point of the fused time series under that working condition as the starting point and uses the sliding window to extract a sample time series from the fused time series under that working condition. The sample time series includes multiple consecutive data points with the same number as the preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a sample time series. When the sample time series extracted by the sliding window includes the last data point of the fused time series under that working condition, the sliding window stops moving to obtain multiple sample time series under that working condition.

[0038] The extraction module is used to extract the actual state of charge data corresponding to the last time step of the sample time series from the actual state of charge time series under the working condition to which the sample time series belongs, and use it as the label corresponding to the sample time series.

[0039] The training module is used to train a preset lithium battery state of charge estimation model by taking the time series of each sample as input and the label corresponding to each sample time series as output, so as to obtain a trained lithium battery state of charge estimation model.

[0040] Optionally, based on the discharge data of the lithium battery, the extended Kalman filter algorithm is used to estimate the state of charge of the lithium battery, obtaining the first estimated state of charge time series of the lithium battery under different operating conditions, specifically including:

[0041] Establish an equivalent circuit model for a lithium battery;

[0042] Based on the discharge data of lithium batteries, the parameters of the equivalent circuit model are identified to obtain the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition.

[0043] Based on the discharge data of the lithium battery and the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition, the first estimated state of charge time series of the lithium battery under different operating conditions is obtained by iteratively using the extended Kalman filter algorithm.

[0044] Another aspect of the present invention provides a lithium battery state of charge estimation device, comprising:

[0045] The second acquisition module is used to acquire the voltage time series and current time series of the lithium battery under test during the discharge process;

[0046] The second extended Kalman filter module is used to estimate the state of charge of the lithium battery under test based on the voltage time series and current time series of the lithium battery under test during the discharge process, and obtain the first estimated state of charge time series of the lithium battery under test.

[0047] The second splicing module is used to align the voltage time series, current time series and first estimated state of charge time series of the lithium battery under test during the discharge process according to time steps, and splice them into a fused time series.

[0048] The second extraction module is used to extract a predicted sample time series from the fusion time series of the lithium battery under test by starting from the starting data point of the fusion time series of the lithium battery under test using a sliding window of a preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a predicted sample time series. When the sample time series extracted by the sliding window includes the last data point of the fusion time series of the lithium battery under test, the sliding window stops moving to obtain the predicted sample time series of the lithium battery under test. Each predicted sample time series includes multiple consecutive data points of the same number as the preset length.

[0049] The prediction module is used to input each predicted sample time series of the lithium battery under test into a pre-trained lithium battery state of charge estimation model, so that the lithium battery state of charge estimation model outputs the predicted state of charge value corresponding to the predicted sample time series; wherein the pre-trained lithium battery state of charge estimation model is obtained by the training method of the lithium battery state of charge estimation model as described above.

[0050] The arrangement module is used to arrange the predicted state of charge values ​​corresponding to each predicted sample time series in sequence according to time steps to form a second estimated state of charge sequence.

[0051] The weighted fusion module is used to perform weighted fusion of the first estimated state of charge time series and the second estimated state of charge time series to obtain the weighted fused state of charge time series.

[0052] Optionally, the first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain a weighted fused state of charge time series, specifically including:

[0053] The second estimated state of charge time series is padded with zeros at the beginning so that the length of the padded second estimated state of charge time series is consistent with that of the first estimated state of charge time series.

[0054] The first estimated state of charge time series and the padded second estimated state of charge time series are weighted and fused to obtain the weighted fused state of charge time series.

[0055] Optionally, the weighted and fused state-of-charge time series satisfies the following relationship:

[0056] ,

[0057] ,

[0058] ,

[0059] ,

[0060] Where t is the time step, For the weighted fusion, This is the first estimated state-of-charge time series. This is the second estimated state-of-charge time series. This is the second estimated state-of-charge time series after filling. The corrected weights, The uncertainty of the first estimated state-of-charge time series, The uncertainty of the second estimated state-of-charge time series is given by L, where L is the length of the sliding window. The weights before correction. is the midpoint of the smoothing, and p is the slope control coefficient. Attached Figure Description

[0061] Figure 1 A flowchart illustrating the training method for a lithium battery state-of-charge estimation model according to an embodiment of the present invention;

[0062] Figure 2 A circuit diagram of an equivalent circuit model of a lithium battery according to an embodiment of the present invention;

[0063] Figure 3 This is a flowchart of a lithium battery state-of-charge estimation method according to an embodiment of the present invention;

[0064] Figure 4 This is a structural block diagram of a training device for a lithium battery state of charge estimation model according to an embodiment of the present invention.

[0065] Figure 5 This is a structural block diagram of a lithium battery state of charge estimation device according to an embodiment of the present invention. Detailed Implementation

[0066] The preferred embodiments of the present invention are given below with reference to the accompanying drawings and described in detail.

[0067] like Figure 1 As shown, this embodiment of the invention provides a training method for a lithium battery state of charge estimation model, which includes the following steps S110-S160:

[0068] S110: Acquire discharge data of lithium battery, including voltage time series, current time series and actual state of charge time series of lithium battery under different operating conditions.

[0069] Discharge data for lithium-ion batteries can be obtained from publicly available experimental datasets (such as the LG18650HG2 battery dataset from McMaster University). The voltage time series, current time series, and actual state of charge (SPC) time series for each operating condition of a lithium-ion battery include multiple consecutive data points arranged chronologically, with a one-to-one correspondence between the voltage, current, and SPC data at the same time step (i.e., time index). The voltage, current, and SPC data for each time step are obtained by measurement or calculation using existing methods.

[0070] S120: Based on the discharge data of the lithium battery, the extended Kalman filter algorithm is used to estimate the state of charge of the lithium battery, and the first estimated state of charge time series of the lithium battery under different operating conditions is obtained.

[0071] In some embodiments, step S120 specifically includes the following steps S121-S123:

[0072] S121: Establish the equivalent circuit model of the lithium battery.

[0073] A battery model is a way to mathematically and circuitically describe the electrochemical behavior of a battery, used to characterize the dynamic relationships between state variables such as voltage, current, and state of charge (SOC) during charging and discharging. By establishing a reasonable battery model, we can gain a deeper understanding of the battery's electrochemical mechanism, predict its performance change trends, and provide a physical basis for subsequent state estimation and control algorithms. Figure 2 As shown, the Equivalent Circuit Model (ECM) of this invention represents the complex electrochemical reaction process as a network of circuit elements consisting of an ideal voltage source, an ohmic internal resistance, and an RC network. This network can simulate the dynamic response behavior of a battery under an external load in an electrical sense. Here, the open-circuit voltage source (OCV) represents the thermodynamic equilibrium potential of the battery, which has a nonlinear relationship with the state of charge (SOC), and Uoc is the open-circuit voltage; the ohmic internal resistance... The DC impedance characterizing the internal conductive path of the battery; the resistance of the polarization branch in series. and capacitor This model is used to describe the hysteresis and dynamic response characteristics caused by electrochemical reactions and ion diffusion. The equivalent model is connected to the real battery through the relationship between port voltage and current; that is, the model's input is the external current I, and its output is the terminal voltage U, thus reflecting the dynamic characteristics of the real battery.

[0074] S122: Based on the discharge data of the lithium battery, the parameters of the equivalent circuit model are identified to obtain the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition.

[0075] The parameters of the equivalent circuit model include the ohmic internal resistance. ,resistance and capacitor Parameter identification is a crucial step in transforming a model from a theoretical structure into a practically computable one, and its accuracy directly affects the accuracy and stability of subsequent SOC estimation. Inaccurate parameter identification will cause the model output to deviate from the actual measurement, thereby reducing the effectiveness of subsequent filtering and prediction processes.

[0076] In some embodiments, the variable forgetting factor recursive least squares (VFF-RLS) method can be used to identify parameters of the equivalent battery model. Specifically, the voltage and current data at each time step of the voltage and current time series under each operating condition are used to iteratively update the parameters of that operating condition to minimize the error. The parameter update formula is as follows:

[0077]

[0078]

[0079]

[0080] in, The observation vector consists of voltage time series, current time series, and their integral terms, where k represents the k-th time step. This represents the voltage data at the k-th time step of the voltage time series. , where K represents the values ​​of the parameters at the k-th time step; Here is the gain matrix. Let covariance be the parameter. It is a forgetting factor. , The baseline forgetting factor, typically ranging from 0.95 to 0.99 (here we take an initial fixed value of 0.95), is used to control the degree to which the algorithm retains historical data; This is an adjustment coefficient used to balance the effect of error on the magnitude of change in the forgetting factor; The residual is given in the formula below; hyperbolic tangent function. Limit output to The interval ensures that the forgetting factor is dynamically adjusted within a reasonable range: when the error is large, the parameter response is accelerated, and when the error is small, the estimation remains stable.

[0081] It is important to note that the voltage data in the voltage time series is used as input in the observation vector, while... As outputs, the two play different roles in the formula: the terminal voltage in the observation vector is used to describe the current state and historical information of the system, providing a basis for model prediction; the output terminal voltage is the actual measured value, which is the target of parameter update.

[0082] Understandable is the forgetting factor. Alternatively, it can be directly used as the baseline forgetting factor, without needing to calculate it based on the residuals. .

[0083] S123: Based on the discharge data of the lithium battery and the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition, the first estimated state of charge time series of the lithium battery under different operating conditions is obtained by iteratively using the extended Kalman filter algorithm (EKF).

[0084] The iterative formula for the extended Kalman filter algorithm is as follows:

[0085]

[0086]

[0087] in, This is the estimated state of charge at time step k+1, i.e., the value at time step k+1 of the first estimated state of charge time series. This is the estimated state of charge at time step k, i.e., the value of time step k in the first estimated state of charge time series; This represents the polarization voltage of the RC network at time step k, used to describe the transient response of the battery under dynamic load; , , These are the values ​​of each parameter of the equivalent circuit model obtained in step S122 at time step k; This represents the coulomb efficiency, used for correction in SOC integral calculations. It is the time step used for SOC integral calculation; Indicates the battery's rated capacity; The battery charging and discharging current represents time step k, i.e., the value of the current time series at time step k; It is process noise, reflecting dynamic changes that are not modeled in the model; It is the derivative of the open-circuit voltage with respect to the state of charge (SOC), used to map changes in SOC to the terminal voltage; Indicates the load current; It refers to measurement noise, describing the error in terminal voltage measurement.

[0088] In the above formula, This corresponds to the time step in step S122. The discrete parameter values, i.e. Therefore, the state prediction and observation update at each time step use the latest identified parameters from that time step, ensuring that the model's dynamic characteristics are consistent with the actual battery behavior. Observation equations state variables and polarization voltage Mapping to observables and taking into account ohmic voltage drop. With measurement noise The extended Kalman filter (EPF) model, used as the input to the observation equations, corrects the state predictions against actual measurements. Its primary function is to represent the battery's dynamic behavior (SOC and polarization voltage variations over time) in matrix form, facilitating recursive calculations using the Kalman filter. The EPF model linearizes the nonlinear system, enabling real-time SOC estimation. The initial SOC can be an approximate value provided by experimental data or obtained from the initial battery voltage via the OCV-SOC curve; this initial value serves as the starting point for state recursion. The model's dynamic characteristics are ensured to match actual battery behavior through online-identified parameters (R0, R1, C1).

[0089] At each time step k, the prediction error (i.e., residual) of SOC. The calculation formula is:

[0090]

[0091] in It is the voltage value (i.e., the actual voltage) at time step k of the voltage time series. This is the terminal voltage (i.e., the predicted voltage) predicted by EKF based on the state equation and the observation equation. Residual The forgetting factor used in step S122 is dynamically adjusted to achieve closed-loop optimization of parameter identification.

[0092] The formula for calculating the uncertainty of the first estimated state-of-charge time series is as follows:

[0093]

[0094] in, The uncertainty of the time step k for the first estimated state-of-charge time series. Let be the element in the first row and first column of the covariance matrix at time step k.

[0095] The first estimated state of charge time series of a lithium battery under each operating condition has the same length as the time series of its voltage and current under that operating condition, and each time step corresponds one-to-one.

[0096] S130: Align the voltage time series, current time series and first estimated state of charge time series of the lithium battery under each operating condition by time step, and splice them into a fused time series. Each data point of the fused time series includes the voltage data, current data and first estimated state of charge data of the corresponding time step.

[0097] Since the voltage time series, current time series, and first estimated state of charge time series have the same length and each time step corresponds to another, they can be aligned in time and spliced ​​together into a fused time series. The length of the fused time series is the same as that of the original time series and each time step corresponds to another. Each data point of the fused time series includes three data points corresponding to the time step, namely voltage data, current data, and first estimated state of charge data.

[0098] One fused time series can be obtained for each operating condition, and multiple fused time series can be obtained for multiple operating conditions.

[0099] S140: Set a sliding window of a preset length, which is less than the length of the fused time series under each working condition; for the fused time series under each working condition, make the sliding window take the starting data point of the fused time series under that working condition as the starting point, and use the sliding window to extract a sample time series from the fused time series under that working condition. The sample time series includes multiple consecutive data points with the same number as the preset length; then make the sliding window move to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a sample time series. When the sample time series extracted by the sliding window includes the last data point of the fused time series under that working condition (i.e., the data point corresponding to the last time step), stop moving the sliding window to obtain multiple sample time series under that working condition.

[0100] Since the lengths of the fused time series under different operating conditions are not necessarily the same, preprocessing is required to generate samples for training the model. Specifically, a sliding window of the same length can be set for the fused time series under each operating condition, and the length of the sliding window should be less than the length of the fused time series under each operating condition. Then, using the set sliding window, multiple sample time series are extracted from the fused time series under each operating condition, and each sample time series includes multiple continuous data points with the same number as the length of the sliding window. Assume the length of the fused time series under a certain operating condition is M, and the length of the sliding window is L, where M and L are both positive integers, and L < M. The time steps of the fused time series are 0 to M-1. The truncation method is as follows: First, the sliding window truncates L consecutive data points starting from the starting data point of the fused time series (i.e., the data point corresponding to time step 0), obtaining the first sample time series, which includes L data points from time step 0 to L-1. Then, the sliding window moves one step to the right, truncates L consecutive data points starting from the second data point of the fused time series (i.e., the data point corresponding to time step 1), obtaining the second sample time series, which includes L data points from time step 1 to L. This process continues, with the sliding window moving multiple times in steps of 1, until it moves to the starting point corresponding to time step ML. The last sample time series obtained at this point includes L data points from time step ML to M-1. Through the above truncation method, N = M - L + 1 sample time series can be obtained.

[0101] Since the sliding window has the same length, the time series of each sample under different working conditions has the same length, which can be used to train the model.

[0102] S150: For each sample time series, extract the actual state of charge data corresponding to the last time step of the sample time series from the actual state of charge time series under the operating condition to which the sample time series belongs, and use it as the label corresponding to the sample time series.

[0103] As described in step S140, each sample time series is obtained by extracting a portion of the fused time series under a certain operating condition. The operating condition to which the extracted fused time series belongs is the operating condition to which the sample time series belongs. Each sample time series includes multiple continuous data points, and each data point corresponds to a time step of the extracted fused time series. Since the time steps of the actual state of charge time series and the fused time series under the same operating condition are in one-to-one correspondence, the actual state of charge data that is the same as the last time step of the sample time series can be extracted from the actual state of charge time series under the operating condition to which the sample time series belongs, and used as the label of the sample time series for subsequent training of the model.

[0104] S160: Using the time series of each sample as input and the label corresponding to each time series of the sample as output, train the preset lithium battery state of charge estimation model to obtain the trained lithium battery state of charge estimation model.

[0105] The preset lithium battery state of charge estimation model can be any suitable machine learning model, such as RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory Network), Bi-LSTM (Bidirectional Long Short-Term Memory Network), etc. For example, the lithium battery state of charge estimation model can be a Bi-LSTM network model, the specific principle of which is well known in the art and will not be elaborated here.

[0106] The training method for the lithium battery state of charge (SOC) estimation model in this embodiment of the invention uses training samples derived from a first estimated SOC time series based on the extended Kalman filter algorithm. This time series implicitly contains information about the nonlinear mapping of battery dynamics and terminal voltage to SOC, thus providing physical constraints for the model and guiding it to learn an SOC evolution pattern that better reflects actual battery behavior, achieving an organic integration of physical model and data-driven approaches. By introducing the first estimated SOC time series, the lithium battery SOC estimation model can perform error correction based on existing estimates without having to learn complex nonlinear mapping relationships from scratch. This not only reduces training difficulty... The speed of convergence is significantly accelerated, and the accuracy of SOC prediction is improved. In the case of rapid changes in battery conditions or the presence of measurement noise, the data of the first estimated state of charge time series can help the state of charge estimation model capture key dynamic trends, mitigate the risk of overfitting, and thus improve the model's adaptability and robustness to unknown conditions. Multiple sample time series can be extracted from the fused time series of each condition by using a sliding window, thereby increasing the number of samples. The label corresponding to each sample time series is the actual state of charge value corresponding to the last time step of the sample time series. This allows us to focus on the prediction of the state at the end of the sequence, reduce output complexity, and improve training stability.

[0107] like Figure 3 As shown, this embodiment of the invention provides a method for estimating the state of charge of a lithium battery, which includes the following steps S210-S270:

[0108] S210: Obtain the voltage and current time series of the lithium battery under test during the discharge process (at this time, it is only necessary to obtain the voltage and current time series under a certain operating condition).

[0109] S220: Based on the voltage and current time series of the lithium battery under test during the discharge process, the extended Kalman filter algorithm is used to estimate the state of charge of the lithium battery under test, and the first estimated state of charge time series of the lithium battery under test is obtained.

[0110] S230: Align the voltage time series, current time series and first estimated state of charge time series of the lithium battery under test during the discharge process by time steps and splice them into a fused time series;

[0111] S240: Starting from the initial data point of the fusion time series of the lithium battery under test, a sliding window of a preset length is used to extract a predicted sample time series from the fusion time series of the lithium battery under test. Then, the sliding window is moved to the right multiple times with a step size of 1. Each time the sliding window moves, a predicted sample time series is extracted. When the sample time series extracted by the sliding window includes the last data point of the fusion time series of the lithium battery under test, the sliding window is stopped to obtain the predicted sample time series of the lithium battery under test. Each predicted sample time series includes multiple consecutive data points of the same number as the preset length.

[0112] S250: Input each predicted sample time series of the lithium battery to be tested into the pre-trained lithium battery state of charge estimation model, so that the lithium battery state of charge estimation model outputs the predicted state of charge value corresponding to the predicted sample time series; wherein the pre-trained lithium battery state of charge estimation model is trained using the method in steps S110-S160.

[0113] S260: Arrange the predicted state of charge values ​​corresponding to each predicted sample time series in sequence according to time steps to form a second estimated state of charge sequence;

[0114] S270: The first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain a weighted fused state of charge time series, which is used as the final state of charge estimate.

[0115] The weighted fusion formula is as follows:

[0116]

[0117] Where t is the time step, This is the weighted and fused state-of-charge time series. This is the first estimated state-of-charge time series. This is the second estimated state-of-charge time series; The corrected weights satisfy the following relationship:

[0118]

[0119] in, The uncertainty of the first estimated state-of-charge time series, The uncertainty of the second estimated state-of-charge time series, is the weight before correction.

[0120] satisfies the following relational expression:

[0121]

[0122] where L is the length of the sliding window.

[0123] In the SOC fusion strategy, the initial, middle, and end stages are divided according to the battery state of charge percentage interval, representing the state of the battery at different discharge stages. Different stages correspond to different reliability preferences for methods. In the initial stage, SOC > 80%. The battery has just started discharging, the available capacity rate is high, the SOC changes slowly, and the battery voltage curve is relatively stable. The first estimated state of charge time series is robust and has small errors. The second estimated state of charge time series is slightly inaccurate due to the influence of data sparsity; in the middle stage, SOC is in the middle interval, 20% < SOC < 80%. The battery discharges relatively frequently, and the nonlinear characteristics are obvious. The machine learning model can learn the complex nonlinear SOC evolution trend, with obvious advantages; at the end stage, SOC is close to low power, less than 20%. The battery is close to the end of discharge, and the polarization effect and voltage drop are obvious. The extended Kalman filter algorithm is more robust in modeling under extreme voltages. Based on the above theory, it is possible to design satisfies the following relational expression:

[0124]

[0125] where is the midpoint of the smooth change of the Sigmoid curve, which can be selected according to experience. For example, it can be taken as 50%; p is the slope control coefficient of the Sigmoid function, used to adjust the steepness of the weight change, and the value range is [5, 15].

[0126] Since the length of the second estimated state of charge time series obtained in step S260 is less than the length of the first estimated state of charge time series, the second estimated state of charge time series is only aligned with the latter part of the first estimated state of charge time series in terms of time steps. Therefore, before weighted fusion, it is necessary to fill the second estimated state of charge time series with zeros, that is, fill 0 at the front end of the second estimated state of charge time series (the position of the missing time steps not aligned with the first estimated state of charge time series) to make the length of the filled second estimated state of charge time series consistent with that of the first estimated state of charge time series, and all time steps of the two are completely aligned; then the two are weighted and fused according to the above weighted fusion formula.

[0127] The lithium battery state of charge estimation method of this invention integrates the first estimated state of charge time series and the second estimated state of charge time series based on time-dependent modified weights to obtain the final estimated value of the state of charge time series. This achieves the complementary advantages of machine learning models and extended Kalman filters under different SOC stages and different uncertainty conditions, significantly improving the overall prediction accuracy and robustness, especially in the boundary segments of the early and late SOC stages.

[0128] like Figure 4 As shown, this embodiment of the invention provides a training device for a lithium battery state of charge estimation model, which includes a first acquisition model 11, a first extended Kalman filter module 12, a first splicing module 13, a first truncation module 14, an extraction module 15, and a training module 16.

[0129] The first acquisition module 11 is used to acquire the discharge data of the lithium battery, including the voltage time series, current time series and actual state of charge time series of the lithium battery under different operating conditions.

[0130] The first Kalman filter module 12 is used to estimate the state of charge of the lithium battery based on the discharge data of the lithium battery and to obtain the first estimated state of charge time series of the lithium battery under different operating conditions.

[0131] The first splicing module 13 is used to align the voltage time series, current time series and first estimated state of charge time series of each operating condition of the lithium battery by time step, and splice them into a fused time series. Each data point of the fused time series includes the voltage data, current data and first estimated state of charge data of the corresponding time step.

[0132] The first extraction module 14 is used to set a sliding window of a preset length, which is less than the length of the fused time series under each working condition. For the fused time series under each working condition, the sliding window takes the starting data point of the fused time series under that working condition as the starting point and uses the sliding window to extract a sample time series from the fused time series under that working condition. The sample time series includes multiple consecutive data points with the same number as the preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a sample time series. When the sample time series extracted by the sliding window includes the last data point of the fused time series under that working condition (i.e., the data point corresponding to the last time step), the sliding window stops moving to obtain multiple sample time series under that working condition.

[0133] The extraction module 15 is used to extract the actual state of charge data corresponding to the last time step of the sample time series from the actual state of charge time series under the working condition to which the sample time series belongs, and use it as the label corresponding to the sample time series for each sample time series.

[0134] The training module 16 is used to train the preset lithium battery state of charge estimation model with each sample time series as input and the corresponding label of each sample time series as output, so as to obtain the trained lithium battery state of charge estimation model.

[0135] The first acquisition model 11, the first extended Kalman filter module 12, the first splicing module 13, the first truncation module 14, the extraction module 15, and the training module 16 are functional modules that implement steps S110-S160 of the training method. Their specific implementation methods can be found in the training method embodiment, and will not be repeated here.

[0136] like Figure 5 As shown, this embodiment of the invention provides a lithium battery state of charge estimation device, which includes a second acquisition module 21, a second extended Kalman filter module 22, a second splicing module 23, a second truncation module 24, a prediction module 25, an arrangement module 26, and a weighted fusion module 27.

[0137] The second acquisition module 21 is used to acquire the voltage time series and current time series of the lithium battery under test during the discharge process;

[0138] The second extended Kalman filter module 22 is used to estimate the state of charge of the lithium battery under test based on the voltage time series and current time series of the lithium battery under test during the discharge process, and obtain the first estimated state of charge time series of the lithium battery under test.

[0139] The second splicing module 23 is used to align the voltage time series, current time series and first estimated state of charge time series of the lithium battery under test during the discharge process according to time steps, and splice them into a fused time series.

[0140] The second extraction module 24 is used to extract a predicted sample time series from the fusion time series of the lithium battery under test by starting a sliding window of a preset length from the starting data point of the fusion time series of the lithium battery under test. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a predicted sample time series. When the sample time series extracted by the sliding window includes the last data point of the fusion time series of the lithium battery under test, the sliding window stops moving to obtain the predicted sample time series of the lithium battery under test. Each predicted sample time series includes multiple consecutive data points of the same number as the preset length.

[0141] The prediction module 25 is used to input each predicted sample time series of the lithium battery under test into the pre-trained lithium battery state of charge estimation model, so that the lithium battery state of charge estimation model outputs the predicted state of charge value corresponding to the predicted sample time series; wherein the pre-trained lithium battery state of charge estimation model is obtained by using the training method of the lithium battery state of charge estimation model described above.

[0142] The arrangement module 26 is used to arrange the predicted state of charge values ​​corresponding to each predicted sample time series in sequence according to time steps to form a second estimated state of charge sequence.

[0143] The weighted fusion module 27 is used to perform weighted fusion of the first estimated state of charge time series and the second estimated state of charge time series to obtain the weighted fused state of charge time series.

[0144] The second acquisition module 21, the second extended Kalman filter module 22, the second splicing module 23, the second truncation module 24, the prediction module 25, the arrangement module 26, and the weighted fusion module 27 are functional modules that implement steps S210-S270 of the estimation method. Their specific implementation methods can be found in the estimation method, and will not be repeated here.

[0145] Another embodiment of the present invention provides a readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to implement the training method or estimation method in the above embodiments of the present invention.

[0146] Another embodiment of the present invention provides an electronic device, which includes a memory and a processor. The memory stores executable code, and when the processor executes the executable code, it implements the training method or estimation method in the above embodiments of the present invention.

[0147] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0148] For ease of description, the above apparatus is described in terms of its functions, divided into various units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0149] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0150] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0151] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0152] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0153] In a typical configuration, an electronic device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0154] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0155] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by electronic devices. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0156] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0157] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0158] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0159] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0160] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of the invention. Various variations can be made to the above embodiments of the present invention. That is, all simple and equivalent changes and modifications made based on the claims and description of this invention fall within the protection scope of the claims of this patent. All aspects not described in detail in this invention are conventional technical content.

Claims

1. A training method for a lithium battery state-of-charge estimation model, characterized in that, include: Acquire discharge data of lithium batteries, including voltage time series, current time series and actual state of charge time series of lithium batteries under different operating conditions; Based on the discharge data of lithium batteries, the extended Kalman filter algorithm is used to estimate the state of charge of lithium batteries, and the first estimated state of charge time series of lithium batteries under different operating conditions is obtained. The voltage time series, current time series and first estimated state of charge time series of the lithium battery under each operating condition are aligned by time step and spliced ​​into a fused time series. Each data point of the fused time series includes the voltage data, current data and first estimated state of charge data of the corresponding time step. Set a sliding window of a preset length, which is less than the length of the fused time series under each operating condition. For the fused time series under each operating condition, the sliding window takes the starting data point of the fused time series under that operating condition as the starting point and uses the sliding window to extract a sample time series from the fused time series under that operating condition. The sample time series includes multiple consecutive data points with the same number as the preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a sample time series. When the sample time series extracted by the sliding window includes the last data point of the fused time series under that operating condition, the sliding window stops moving to obtain multiple sample time series under that operating condition. For each sample time series, extract the actual state of charge data corresponding to the last time step of the sample time series from the actual state of charge time series under the operating condition to which the sample time series belongs, and use it as the label corresponding to the sample time series; Using the time series of each sample as input and the label corresponding to each time series as output, the preset lithium battery state of charge estimation model is trained to obtain the trained lithium battery state of charge estimation model.

2. The training method for the lithium battery state-of-charge estimation model according to claim 1, characterized in that, Based on the discharge data of lithium batteries, the extended Kalman filter algorithm is used to estimate the state of charge (SOC) of the lithium batteries, obtaining the first estimated SOC time series under different operating conditions, specifically including: Establish an equivalent circuit model for a lithium battery; Based on the discharge data of lithium batteries, the parameters of the equivalent circuit model are identified to obtain the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition. Based on the discharge data of the lithium battery and the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition, the first estimated state of charge time series of the lithium battery under different operating conditions is obtained by iteratively using the extended Kalman filter algorithm.

3. A method for estimating the state of charge of a lithium battery, characterized in that, include: Obtain the voltage and current time series of the lithium battery under test during the discharge process; Based on the voltage and current time series of the lithium battery under test during the discharge process, the extended Kalman filter algorithm is used to estimate the state of charge of the lithium battery under test, and the first estimated state of charge time series of the lithium battery under test is obtained. The voltage time series, current time series and first estimated state of charge time series of the lithium battery under test during the discharge process are aligned by time steps and spliced ​​into a fused time series. A sliding window of a preset length is started from the beginning data point of the fusion time series of the lithium battery under test. A predicted sample time series is extracted from the fusion time series of the lithium battery under test using the sliding window. Then, the sliding window is moved to the right multiple times with a step size of 1. Each time the sliding window moves, a predicted sample time series is extracted. When the sample time series extracted by the sliding window includes the last data point of the fusion time series of the lithium battery under test, the sliding window is stopped to obtain the predicted sample time series of the lithium battery under test. Each predicted sample time series includes multiple consecutive data points with the same number as the preset length. Each predicted sample time series of the lithium battery under test is input into a pre-trained lithium battery state of charge estimation model, so that the lithium battery state of charge estimation model outputs the predicted state of charge value corresponding to the predicted sample time series; wherein the pre-trained lithium battery state of charge estimation model is obtained by the training method of the lithium battery state of charge estimation model according to any one of claims 1-2; The predicted state of charge values ​​corresponding to each predicted sample time series are arranged sequentially according to time steps to form a second estimated state of charge sequence. The first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain the weighted fused state of charge time series.

4. The lithium battery state of charge estimation method according to claim 3, characterized in that, The first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain a weighted fused state of charge time series, specifically including: The second estimated state of charge time series is padded with zeros at the beginning so that the length of the padded second estimated state of charge time series is consistent with that of the first estimated state of charge time series. The first estimated state of charge time series and the padded second estimated state of charge time series are weighted and fused to obtain the weighted fused state of charge time series.

5. The lithium battery state of charge estimation method according to claim 4, characterized in that, The weighted and fused state-of-charge time series satisfies the following relationship: ， ， ， ， Where t is the time step, For the weighted fusion, This is the first estimated state-of-charge time series. This is the second estimated state-of-charge time series. This is the second estimated state-of-charge time series after filling. The corrected weights, The uncertainty of the first estimated state-of-charge time series, The uncertainty of the second estimated state-of-charge time series is given by L, where L is the length of the sliding window. The weights before correction. is the midpoint of the smoothing, and p is the slope control coefficient.

6. A training device for a lithium battery state-of-charge estimation model, characterized in that, include: The first acquisition module is used to acquire the discharge data of the lithium battery, including the voltage time series, current time series and actual state of charge time series of the lithium battery under different operating conditions. The first Kalman filter module is used to estimate the state of charge of the lithium battery based on the discharge data of the lithium battery using the extended Kalman filter algorithm, so as to obtain the first estimated state of charge time series of the lithium battery under different operating conditions. The first splicing module is used to align the voltage time series, current time series and first estimated state of charge time series of each operating condition of the lithium battery by time step, and splice them into a fused time series. Each data point of the fused time series includes the voltage data, current data and first estimated state of charge data of the corresponding time step. The first extraction module is used to set a sliding window of a preset length, which is less than the length of the fused time series under each working condition. For the fused time series under each working condition, the sliding window takes the starting data point of the fused time series under that working condition as the starting point and uses the sliding window to extract a sample time series from the fused time series under that working condition. The sample time series includes multiple consecutive data points with the same number as the preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a sample time series. When the sample time series extracted by the sliding window includes the last data point of the fused time series under that working condition, the sliding window stops moving to obtain multiple sample time series under that working condition. The extraction module is used to extract the actual state of charge data corresponding to the last time step of the sample time series from the actual state of charge time series under the working condition to which the sample time series belongs, and use it as the label corresponding to the sample time series. The training module is used to train a preset lithium battery state of charge estimation model by taking the time series of each sample as input and the label corresponding to each sample time series as output, so as to obtain a trained lithium battery state of charge estimation model.

7. The training device for the lithium battery state-of-charge estimation model according to claim 6, characterized in that, Based on the discharge data of lithium batteries, the extended Kalman filter algorithm is used to estimate the state of charge (SOC) of the lithium batteries, obtaining the first estimated SOC time series under different operating conditions, specifically including: Establish an equivalent circuit model for a lithium battery; Based on the discharge data of lithium batteries, the parameters of the equivalent circuit model are identified to obtain the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition. Based on the discharge data of the lithium battery and the values ​​of each parameter of the equivalent circuit model at different time steps under each operating condition, the first estimated state of charge time series of the lithium battery under different operating conditions is obtained by iteratively using the extended Kalman filter algorithm.

8. A lithium battery state of charge estimation device, characterized in that, include: The second acquisition module is used to acquire the voltage time series and current time series of the lithium battery under test during the discharge process; The second extended Kalman filter module is used to estimate the state of charge of the lithium battery under test based on the voltage time series and current time series of the lithium battery under test during the discharge process, and obtain the first estimated state of charge time series of the lithium battery under test. The second splicing module is used to align the voltage time series, current time series and first estimated state of charge time series of the lithium battery under test during the discharge process according to time steps, and splice them into a fused time series. The second extraction module is used to extract a predicted sample time series from the fusion time series of the lithium battery under test by starting from the starting data point of the fusion time series of the lithium battery under test using a sliding window of a preset length. Then, the sliding window moves to the right multiple times with a step size of 1. Each time the sliding window moves, it extracts a predicted sample time series. When the sample time series extracted by the sliding window includes the last data point of the fusion time series of the lithium battery under test, the sliding window stops moving to obtain the predicted sample time series of the lithium battery under test. Each predicted sample time series includes multiple consecutive data points of the same number as the preset length. The prediction module is used to input each predicted sample time series of the lithium battery under test into the lithium battery state of charge estimation model that is pre-trained, so that the lithium battery state of charge estimation model outputs the predicted state of charge value corresponding to the predicted sample time series. The pre-trained lithium battery state of charge estimation model is obtained by using the training method of the lithium battery state of charge estimation model according to any one of claims 1-2; The arrangement module is used to arrange the predicted state of charge values ​​corresponding to each predicted sample time series in sequence according to time steps to form a second estimated state of charge sequence. The weighted fusion module is used to perform weighted fusion of the first estimated state of charge time series and the second estimated state of charge time series to obtain the weighted fused state of charge time series.

9. The lithium battery state of charge estimation device according to claim 8, characterized in that, The first estimated state of charge time series and the second estimated state of charge time series are weighted and fused to obtain a weighted fused state of charge time series, specifically including: The second estimated state of charge time series is padded with zeros at the beginning so that the length of the padded second estimated state of charge time series is consistent with that of the first estimated state of charge time series. The first estimated state of charge time series and the padded second estimated state of charge time series are weighted and fused to obtain the weighted fused state of charge time series.

10. The lithium battery state of charge estimation device according to claim 9, characterized in that, The weighted and fused state-of-charge time series satisfies the following relationship: ， ， ， ， Where t is the time step, For the weighted fusion, This is the first estimated state-of-charge time series. This is the second estimated state-of-charge time series. This is the second estimated state-of-charge time series after filling. The corrected weights, The uncertainty of the first estimated state-of-charge time series, The uncertainty of the second estimated state-of-charge time series is given by L, where L is the length of the sliding window. The weights before correction. is the midpoint of the smoothing, and p is the slope control coefficient.

Images