A lithium battery full life cycle soc self-correction method and system

Abstract

This invention relates to the field of lithium battery lifecycle management technology, and discloses a lithium battery lifecycle SOC self-correction method and system. The method includes acquiring the terminal voltage change sequence, current change sequence, surface temperature change sequence, and cycle characteristic parameters of the lithium battery; performing time-series mapping on the terminal voltage and surface temperature to obtain a dynamic change trajectory, matching a reference trajectory to calculate the trajectory offset, and combining the cycle characteristic parameters to deduce the aging state; updating the battery state model based on the aging state to obtain a calculation benchmark, and mapping to calculate the initial state of charge; analyzing the dynamic internal resistance and predicting the voltage residual, and combining temperature and residual compensation to obtain the target state of charge; if the target state of charge is lower than the derating threshold, the maximum discharge current is deduced to generate an over-discharge protection command. This method can achieve self-correction and safe management of the lithium battery's state of charge throughout its lifecycle.

Description

Technical Field

[0001] This invention relates to the field of lithium battery lifecycle management technology, and in particular to a lithium battery lifecycle SOC self-correction method and system. Background Technology

[0002] Currently, in the field of lithium battery life cycle management technology, health prediction and fault management, as key links in preventing over-discharge risks and ensuring operational safety, have become the focus of research in the field.

[0003] In one existing technology, a SOC estimation method based on fixed parameter calibration or lookup table mapping is mainly used to ensure the judgment of state of charge. This method mainly relies on factory calibration rules or voltage and temperature indicators for evaluation. For example, it uses a combination of ampere-hour integration and open-circuit voltage lookup table to perform full-process estimation, and judges the remaining capacity by preset capacity and internal resistance parameters. However, this method is significantly insufficient in long-term service and multi-stage usage scenarios of batteries. Fixed calibration parameters cannot adapt to the dynamic migration of internal resistance, capacity shrinkage and voltage and temperature response trajectories during battery aging. The evaluation of a single indicator is easily affected by environmental temperature fluctuations and cyclic aging. It is also affected by changes in battery life cycle stages and complex operating conditions, which can easily lead to continuous drift of the state of charge estimation benchmark. Moreover, it lacks the ability to identify and compensate for the deviation of the dynamic change trajectory of voltage and temperature. Especially in the high-frequency use and variable operating conditions of the battery throughout its life cycle, it is difficult to avoid the cumulative error and over-discharge risk caused by benchmark drift, resulting in insufficient reliability of battery state judgment.

[0004] In summary, existing technologies are insufficient to achieve self-correction and safe management of the state of charge throughout the entire life cycle of lithium batteries, and cannot meet the core requirements for the reliability of lithium batteries throughout their entire life cycle. Summary of the Invention

[0005] This invention provides a method and system for self-correction of the state of charge (SOC) throughout the entire life cycle of a lithium battery, so as to realize the self-correction and safe management of the state of charge throughout the entire life cycle of the lithium battery, and meet the core requirements of the reliability of lithium battery operation throughout the entire life cycle.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for self-correcting the state of charge (SOC) of a lithium battery throughout its entire life cycle, comprising: Obtain the terminal voltage change sequence, current change sequence, surface temperature change sequence, and cycle characteristic parameters of the lithium battery; A dynamic trajectory is obtained by performing time-series mapping calculation on the terminal voltage change sequence and the surface temperature change sequence. The dynamic trajectory is calculated by matching the current change sequence and the surface temperature change sequence with a pre-obtained reference trajectory to obtain the trajectory offset. The aging state is obtained by performing deduction calculation in combination with the cyclic feature parameters. The preset battery state model is updated based on the aging state to obtain the calculation benchmark. The initial state of charge is obtained by performing state mapping calculation based on the calculation benchmark, the terminal voltage change sequence and the current change sequence. The dynamic internal resistance is obtained by performing analytical calculation based on the terminal voltage change sequence and the current change sequence. The voltage residual sequence is obtained by performing predictive calculation based on the calculation benchmark. The initial state of charge is compensated by combining the surface temperature change sequence and the voltage residual sequence to obtain the target state of charge. If the target state of charge is lower than the preset derating threshold, the maximum discharge current is calculated based on the target state of charge and the dynamic internal resistance, and an over-discharge protection command is generated; if it is not lower than the derating threshold, a normal operation command is generated.

[0007] Secondly, the present invention provides a lithium battery full life cycle SOC self-correction system, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the lithium battery full life cycle SOC self-correction method as described above.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains the terminal voltage change sequence, current change sequence, surface temperature change sequence and cycle characteristic parameters of lithium battery, performs time-series mapping calculation on terminal voltage and surface temperature to obtain dynamic change trajectory, and matches the reference trajectory from the reference trajectory set according to current and temperature, calculates the difference between dynamic trajectory and reference trajectory to obtain trajectory offset, and deduces aging state by combining cycle characteristic parameters. It breaks through the limitation of traditional fixed parameter calibration that cannot adapt to the continuous migration of voltage and temperature response trajectory caused by internal resistance increase and capacity shrinkage in the entire life cycle of lithium battery. It integrates multi-source operating signals to dynamically identify the current aging stage of battery, accurately depicts the evolution characteristics of battery electrothermal characteristics under different life cycle stages, solves the problem of reference drift and continuous error accumulation after aging of single ampere-hour integral and open circuit voltage lookup table, and provides accurate aging state basis for subsequent dynamic updating of SOC estimation reference.

[0009] (2) This invention extracts capacity correction coefficients and internal resistance correction coefficients based on aging state, performs update calculations of capacity parameters and internal resistance parameters on the battery state model, and constructs a calculation benchmark by combining state covariance, observation matrix and observation noise parameters. Based on this calculation benchmark, the current sequence is integrated to obtain the integrated state of charge, the terminal voltage is compensated and the voltage state of charge is obtained by looking up the table, and the initial state of charge is obtained by state fusion gain. This breaks through the limitation of traditional methods that use the factory-calibrated capacity and internal resistance parameters throughout the entire life cycle, so that the core model parameters on which SOC estimation depends can be synchronously calibrated with the degree of battery aging, solves the systematic estimation deviation caused by the failure to include capacity decay and internal resistance increase in the estimation benchmark, and provides a dynamic calculation benchmark that matches the current real state of the battery for accurate judgment of remaining power.

[0010] (3) This invention obtains the dynamic internal resistance after temperature and aging correction by analytical calculation of the terminal voltage change sequence and the current change sequence. Combined with the calculation benchmark, it performs prediction calculation to generate the predicted terminal voltage sequence and obtains the voltage residual. After dynamically updating the observation noise by incorporating the surface temperature change, it calculates the correction gain. Using the voltage residual and the local slope at the current state of charge in the voltage state mapping table, it performs linear superposition compensation on the initial state of charge to obtain the target state of charge. When the target state of charge is lower than the derating judgment threshold, it deduces the maximum discharge current to generate an over-discharge protection command. This invention breaks through the limitations of traditional SOC estimation, which lacks real-time compensation for dynamic internal resistance and voltage residual and over-discharge early warning capabilities. It constructs a full-link self-correction system from aging perception, benchmark update, state estimation to dynamic compensation and over-discharge protection, which significantly reduces the accumulated error and improves the operating safety of the battery under complex conditions. This invention meets the core requirements of lithium battery life cycle for SOC estimation accuracy and safety protection. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a lithium battery full life cycle SOC self-correction method provided in the first embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a method for self-correction of SOC throughout the entire life cycle of a lithium battery, including the following steps: S11, obtain the terminal voltage change sequence, current change sequence, surface temperature change sequence and cycle characteristic parameters of the lithium battery; S12, perform time-series mapping calculation on the terminal voltage change sequence and the surface temperature change sequence to obtain a dynamic change trajectory. Based on the current change sequence and the surface temperature change sequence matching the pre-obtained reference trajectory, calculate the difference between the dynamic change trajectory and the reference trajectory to obtain the trajectory offset. Combine the cyclic feature parameters to perform deduction calculation to obtain the aging state. S13, update the preset battery state model according to the aging state to obtain the calculation benchmark, and perform state mapping calculation according to the calculation benchmark, the terminal voltage change sequence and the current change sequence to obtain the initial state of charge; S14, perform analytical calculations based on the terminal voltage change sequence and the current change sequence to obtain the dynamic internal resistance, perform predictive calculations based on the calculation benchmark to obtain the voltage residual sequence, and perform compensation calculations on the initial state of charge by combining the surface temperature change sequence and the voltage residual sequence to obtain the target state of charge; S15, if the target state of charge is lower than the preset derating threshold, the maximum discharge current is calculated based on the target state of charge and the dynamic internal resistance and an over-discharge protection command is generated; if it is not lower than the derating threshold, a normal operation command is generated.

[0014] In step S11, the terminal voltage change sequence, current change sequence, surface temperature change sequence, and cycle characteristic parameters of the lithium battery are obtained.

[0015] It should be noted that, firstly, during the charging and discharging process of the lithium battery, synchronous sampling is performed using voltage sensors, current sensors, and surface temperature sensors deployed on the battery module. The sampling frequency is set to 10Hz, with each sampling lasting 100 seconds, resulting in 1000 data points for each signal. The original voltage signal ranges from 3000mV to 4200mV, the original current signal ranges from -100A to 0A during discharge and from 0A to 100A during charging, and the original surface temperature signal ranges from 5℃ to 45℃. The initial signals collected by these sensors reflect the external manifestations of the electrochemical reactions inside the battery.

[0016] Next, due to the response delays of different sensors and differences in communication links, the three initial signals—voltage, current, and temperature—are misaligned on the time axis. Using the timestamp of the voltage signal as a reference, a dynamic time warping algorithm is employed to align the time axes of the current and temperature signals. Dynamic time warping stretches or compresses the time axis by finding the shortest path between two time series, outputting initial voltage, initial current, and initial surface temperature sequences with completely consistent time dimensions, thus avoiding feature mismatches during subsequent fusion analysis.

[0017] Secondly, the aligned initial sequences are often superimposed with high-frequency noise caused by electromagnetic interference. To suppress this noise, the initial voltage, initial current, and initial surface temperature sequences are each subjected to moving average filtering. The moving window size is set to 5 data points, and the arithmetic mean of the data within the window is calculated point by point to replace the center point value. After this processing, smoothed voltage, smoothed current, and smoothed surface temperature sequences are obtained, which significantly improve the signal-to-noise ratio while preserving the macroscopic trend of battery charging and discharging.

[0018] Subsequently, to obtain continuous and high-resolution signal curves, cubic spline interpolation was performed on the smoothed voltage, current, and surface temperature sequences, respectively. Cubic spline interpolation generates continuous curves while preserving discrete data points by constructing piecewise cubic polynomials, outputting voltage, current, and surface temperature change sequences. The lengths of the three sequences are all 1000, with units of mV, A, and ℃, respectively.

[0019] Then, the cycle characteristic parameters are read from the battery management device. These parameters include the cumulative number of charge-discharge cycles, the percentage of time spent in the constant current charging phase, and the cumulative throughput. The cumulative number of charge-discharge cycles is accumulated by the battery management device after each complete charge-discharge cycle; the percentage of time spent in the constant current charging phase is calculated as the ratio of the constant current charging time to the total charging time; and the cumulative throughput is obtained by accumulating the integral values ​​of the current throughout all historical charge-discharge processes. These three parameters together constitute the three-dimensional cycle characteristic parameters.

[0020] In step S12, the step of performing time-series mapping calculations on the terminal voltage change sequence and the surface temperature change sequence to obtain the dynamic change trajectory includes: Resampling alignment and sliding filtering are performed on the terminal voltage change sequence and the surface temperature change sequence to obtain an aligned time sequence; The aligned time series is subjected to window segmentation to obtain a multidimensional time series input matrix; The multidimensional temporal input matrix is ​​input into a preset temporal inference network to perform nonlinear feature filtering and cross-weighted calculation to obtain the coupled hidden state; Spatial coordinate mapping is performed on the coupled hidden state to obtain single-step trajectory nodes, and temporal splicing is performed on the single-step trajectory nodes to obtain dynamically changing trajectories.

[0021] It should be noted that, firstly, resampling alignment and sliding filtering are performed on the terminal voltage change sequence and the surface temperature change sequence to obtain an aligned time series sequence. Although the terminal voltage change sequence and the surface temperature change sequence have already undergone the aforementioned preprocessing, since the two sequences need to maintain a strictly unified time scale and consistent data point positions before entering the time series extrapolation network, the surface temperature change sequence is resampled again using linear interpolation based on the timestamp of the terminal voltage change sequence, so that the two sequences correspond one-to-one at every time point. Subsequently, a sliding average filter with a window size of 5 is performed on the two resampled sequences. The filter window slides point by point, and the arithmetic mean of the 5 data points within the window is taken as the filtered value at each point. After this processing, the terminal voltage change sequence and the surface temperature change sequence are integrated into an aligned time series sequence with a dimension of 1000×1.

[0022] Next, windowing is performed on the aligned time series to obtain a multi-dimensional time series input matrix. The window length is set to 50 time points, corresponding to a 5-second time span at a 10Hz sampling frequency. This time span covers the continuous changes in battery terminal voltage and temperature during short-term charge and discharge, while avoiding a significant increase in computational complexity due to excessively long windows. The window sliding step is set to 10 time points, moving window by window from the beginning of the aligned time series until less than 50 time points remain. The terminal voltage change sequence and surface temperature change sequence within each window are stacked into a 50×2 matrix, with one dimension corresponding to the 50 time points and the other dimension corresponding to the two feature channels of terminal voltage and surface temperature. After combining all windows, the overall dimension of the multi-dimensional time series input matrix is ​​W×50×2, where W is the number of windows. The multi-dimensional time series input matrix contains W windows, each a 50×2 matrix, which serve as input to the time series extrapolation network, which processes the data window by window.

[0023] Secondly, the multi-dimensional temporal input matrix is ​​input into a pre-defined temporal extrapolation network to perform nonlinear feature filtering and cross-weighting calculations to obtain the coupled hidden state. This temporal extrapolation network adopts a long short-term memory (LSTM) network structure to capture the temporal nonlinear coupling relationship between terminal voltage and surface temperature. The network input is a 50×2 matrix corresponding to a single window. The first LSM layer contains 64 hidden units, with an output dimension of 50×64. The second LSM layer also contains 64 hidden units, with an output dimension of 50×64. The hidden state at the last time step of the second layer is taken as the coupled hidden state, with a dimension of 64×1. Inside the LSM network, the input gate, forget gate, and output gate perform nonlinear feature filtering and cross-weighting on the input sequence through a gating mechanism. The weight values ​​are automatically updated through gradient backpropagation during network training, without using manually set fixed weights. The activation functions used within the network are sigmoid and tanh. The sigmoid function is used for three gated output signals, with values ​​ranging from 0 to 1. The tanh function is used for calculating candidate and hidden states, with values ​​ranging from -1 to 1. This network suppresses noise components in the sequence through cross-weighting while preserving temporal correlation features useful for subsequent trajectory construction.

[0024] The training process of the time-series extrapolation network is as follows. The training set consists of terminal voltage sequences and surface temperature sequences collected by a high-precision battery test bench at various temperature points of 5℃, 15℃, 25℃, 35℃, and 45℃, as well as at charge / discharge rates of 0.5C, 1C, and 2C. This covers the actual operating temperature range, ensuring that the network has training data to support it across all temperature segments from low to high temperatures, and avoiding out-of-range extrapolation at high temperatures such as 35℃ and 45℃. The training samples are divided into training samples according to a window length of 50 and a sliding step size of 10. Each training sample is a 50×2 matrix, and the single-step trajectory coordinates after the last time step of each window are retained as labels. The trajectory coordinates in the labels are directly obtained from the terminal voltage and surface temperature data measured by high-precision sensors under the same operating conditions and calibrated. The voltage normalization range is 3000mV to 4200mV, and the temperature normalization range is 5℃ to 45℃, consistent with the normalization range of the input sequence. The two values ​​are then combined into two-dimensional coordinates. The sensors used were calibrated with standard voltage and temperature sources before the experiment, with a voltage measurement error of ±0.1% and a temperature measurement error of ±0.2℃, ensuring that the labels accurately reflect the actual characteristics of the battery and avoiding system bias introduced by model extrapolation. The loss function is the mean squared error loss, the optimizer is Adam, the initial learning rate is set to 0.001, and a total of 200 training rounds are conducted. After each training round, the mean squared error on the validation set is calculated, and the network parameters corresponding to the round with the smallest validation error are selected as the final time series extrapolation network parameters. Both voltage and temperature sequences in the training set are processed using min-max normalization before input. Voltage values ​​are normalized to the 0-1 range from 3000mV to 4200mV, and temperature values ​​are normalized to the 0-1 range from 5℃ to 45℃.

[0025] Subsequently, spatial coordinate mapping is performed on the coupled hidden state to obtain single-step trajectory nodes. Temporal concatenation is then performed on these single-step trajectory nodes to obtain the dynamically changing trajectory. Spatial coordinate mapping is implemented through a fully connected layer with an input dimension of 64 and an output dimension of 2, using a linear activation function. The 64×1 coupled hidden state is input into this fully connected layer, which outputs a 2×1 coordinate vector as the single-step trajectory node. The two coordinate values ​​in the single-step trajectory node correspond to the positions of the voltage change trend and surface temperature change trend at that time point in the two-dimensional trajectory space, respectively. Following the temporal order of the windows in the original time sequence, the single-step trajectory nodes obtained from all windows are sequentially concatenated to form the complete dynamically changing trajectory. The dynamic trajectory has a dimension of W×2, where W is the number of windows.

[0026] In one implementation, the step of matching the current change sequence with the surface temperature change sequence to a pre-obtained reference trajectory, calculating the difference between the dynamic change trajectory and the reference trajectory to obtain a trajectory offset, and performing a deductive calculation based on the cyclic feature parameters to obtain the aging state includes: The charge / discharge rate features are obtained by performing extraction calculations based on the current change sequence, and the temperature range features are obtained by performing interval mapping calculations based on the surface temperature change sequence. Based on the terminal voltage change sequence and the current change sequence, the state of charge estimation calculation is performed to obtain the state of charge interval characteristics; Based on the charge / discharge rate characteristics, the state of charge range characteristics, and the temperature range characteristics, a baseline trajectory is obtained by traversing and matching from a pre-acquired baseline trajectory set. A spatial distance sequence is obtained by performing path matching calculation between the dynamically changing trajectory and the reference trajectory, and a discrete integral calculation is performed on the spatial distance sequence to obtain the trajectory offset; An aging feature vector is constructed based on the trajectory offset and the cyclic feature parameters. The aging feature vector is then input into a preset mapping classification network to perform state discrimination calculation, thereby obtaining the aging state.

[0027] It should be noted that, firstly, charge / discharge rate characteristics are obtained through extraction calculations based on the current change sequence, and temperature range characteristics are obtained through interval mapping calculations based on the surface temperature change sequence. The charge / discharge rate characteristics include the average charge / discharge rate and the rate fluctuation value. The average charge / discharge rate is calculated as follows: first, the average absolute value of the current in the current change sequence is calculated, then divided by the battery's rated capacity. The rated capacity is directly given in the battery specifications, for example, a rated capacity of 50Ah. The unit is C. The rate fluctuation value is calculated as follows: the standard deviation of the absolute value of the current in the current change sequence is calculated, then divided by the battery's rated capacity. The unit is C. By considering both the average rate and the rate fluctuation simultaneously, the change process of the charge / discharge rate within a segment can be described more completely, avoiding the limitation that the average value alone cannot reflect the fluctuations under actual operating conditions.

[0028] The temperature range characteristics are determined based on the actual operating temperature range of the battery. Four temperature ranges—5℃ to 15℃, 15℃ to 25℃, 25℃ to 35℃, and 35℃ to 45℃—are numbered 0, 1, 2, and 3, respectively. The percentage of the cumulative duration of surface temperature change sequences falling within each range is statistically analyzed, and the range number with the highest percentage is taken as the temperature range characteristic. The boundaries of the four temperature ranges are derived from the lower limit of the temperature sensor's range (5℃) and the upper limit (45℃) in the battery management device. The range of 5℃ to 45℃ is divided into equal widths, with each range having a width of 10℃. When the battery material system changes, the upper and lower limits can be adjusted within the range of 0℃ to 50℃ while maintaining the equal width division of the four ranges.

[0029] The determination method for the state of charge (SOC) interval characteristics is as follows: Based on the terminal voltage change sequence and the current change sequence, the current SOC is estimated using the ampere-hour integration method combined with open-circuit voltage correction. The SOC is then divided into five intervals at 20% intervals: 0%~20%, 20%~40%, 40%~60%, 60%~80%, and 80%~100%. The percentage of time within each SOC segment that falls into each interval is statistically analyzed, and the interval with the largest percentage is taken as the SOC interval characteristic. Introducing this characteristic ensures that the baseline trajectory matching is performed at the same SOC operating point, avoiding the overlap of trajectory shape differences and aging differences caused by different SOCs.

[0030] Next, based on the charge / discharge rate characteristics, state of charge (SCC) interval characteristics, and temperature interval characteristics, a baseline trajectory is obtained by traversing and matching from a pre-acquired baseline trajectory set. The baseline trajectory set is constructed offline when the battery is brand new and in a stable state. During construction, at three charge / discharge rates (0.5C, 1C, 2C), five temperature points (5℃, 15℃, 25℃, 35℃, 45℃), and the aforementioned five SCC intervals, the terminal voltage change sequence and surface temperature change sequence of the quasi-static charging segment are collected and combined to obtain the baseline trajectory set. Each baseline trajectory is stored with a charge / discharge rate field, a temperature interval number field, a SCC interval number field, and a trajectory data field, with each trajectory data having a dimension of Lb×2. Before online traversal matching, a segment applicability determination is performed on the current data segment: if the current values ​​in the current change sequence are all positive values ​​in the charging direction and the rate fluctuation value does not exceed 0.05C, the current segment is determined to be a quasi-static charging segment, and traversal matching is performed; otherwise, the current segment is not used for aging state updates, and the baseline trajectory matching for this segment is skipped. During the traversal matching process, firstly, baseline trajectories are selected where the temperature range number and the state of charge (SOC) range number are completely identical to the current temperature range. Then, the trajectory with the smallest absolute value of the deviation between the average charge / discharge rate and the current average charge / discharge rate (within 0.05C) and the smallest absolute value of the deviation between the rate fluctuation value and the corresponding rate fluctuation value when the baseline trajectory was constructed is selected as the baseline trajectory. The 0.05C setting is based on the current sensor's measurement error at a 10Hz sampling frequency not exceeding 0.1A, which is converted to 0.002C for a 50Ah rated capacity. 0.05C serves as a matching deviation margin to cover sensor drift and battery management device recording errors. If multiple trajectories still meet the criteria after filtering, the one with the smallest absolute value of the average charge / discharge rate deviation is selected. If the average charge / discharge rate deviations of all baseline trajectories with temperature range numbers and SOC range numbers exceed 0.05C, the one with the smallest absolute value of the average charge / discharge rate deviation is selected as the baseline trajectory. If there is no reference trajectory with the same temperature range number and state of charge range number, then select the reference trajectory with the smallest absolute value of average charge / discharge rate deviation among those with adjacent temperature range numbers and adjacent state of charge range numbers.

[0031] Secondly, path matching is performed between the dynamically changing trajectory and the reference trajectory to obtain a spatial distance sequence. Discrete integration is then performed on the spatial distance sequence to obtain the trajectory offset. Path matching employs a dynamic time warping algorithm, which allows nonlinear stretching or compression of the two trajectories on the time axis to handle situations where the length W of the dynamically changing trajectory may not be equal to the length Lb of the reference trajectory. Dynamic time warping calculates the Euclidean distance between the points of the dynamically changing trajectory and the reference trajectory at each alignment node. The Euclidean distances at all alignment nodes are arranged in chronological order to form a spatial distance sequence with a dimension of Ld×1, where Ld is the number of alignment nodes found by dynamic time warping. To ensure the comparability of trajectory offsets between segments of different lengths, the trajectory coordinates are uniformly normalized to min-max before entering dynamic time warping. Voltage coordinates are mapped from 3000mV to 4200mV to 0 to 1, and temperature coordinates are mapped from 5℃ to 45℃ to 0 to 1. Therefore, each distance value in the spatial distance sequence is a dimensionless number, ranging from 0 to √2. Discrete integrals employ a normalized average distance method, which calculates the arithmetic mean of all distance values ​​in the spatial distance sequence as the trajectory offset. Since both the dynamically changing trajectory and the baseline trajectory are taken from the same state of charge range and are both quasi-static charging segments, the difference in trajectory shape is mainly caused by aging, and the difference in operating conditions is effectively controlled, thus ensuring that the trajectory offset primarily reflects aging information. This trajectory offset reflects the average deviation of the current battery's dynamically changing trajectory from the brand-new healthy baseline trajectory; a larger offset indicates more significant aging. Because an average distance is used, the trajectory offset does not change with the number of alignment nodes or the segment length, and can be directly used for comparison of segments of different lengths.

[0032] Subsequently, an aging feature vector is constructed based on the trajectory offset and cycle feature parameters. This aging feature vector is then input into a pre-defined mapping classification network to perform state discrimination calculations, yielding the aging state. The aging feature vector consists of three features: trajectory offset, cumulative charge / discharge cycle count, and the proportion of time spent in the constant current charging phase, with a dimension of 3×1. Before inputting into the mapping classification network, min-max normalization is performed on each of the three features. The upper and lower bounds of the normalization are taken from the maximum and minimum values ​​of the three features in the training set. These upper and lower bounds are saved during training and used directly during online inference. After normalization, all three features are located in the range of 0 to 1. The mapping classification network specifically uses a support vector machine with a radial basis function kernel. The input dimension is 3, and the output is the aging state category: 0, 1, and 2. Aging state 0 represents mild aging, corresponding to a capacity retention rate between 90% and 100%; aging state 1 represents moderate aging, corresponding to a capacity retention rate between 80% and 90%; and aging state 2 represents severe aging, corresponding to a capacity retention rate between 70% and 80%. The 80% capacity retention boundary is set according to industry standards, while 90% and 70% serve as adjacent state boundaries around 80%, allowing for adjustments within the range of 70% to 95% depending on the battery material system.

[0033] The training process of the Support Vector Machine (SVM) is as follows: The training set is constructed from aging experimental data of lithium batteries from the same batch. Under laboratory conditions, after every 50 complete charge-discharge cycles, the trajectory offset, the cumulative number of charge-discharge cycles, and the percentage of time spent in the constant current charging phase are recorded, and a capacity calibration test is performed simultaneously. The actual capacity obtained from the capacity calibration test divided by the rated capacity is the capacity retention rate, which serves as the true value label for the aging state. Samples are labeled as 0, 1, or 2 according to the interval in which the capacity retention rate falls. The total number of samples in the training set is no less than 300, and the difference in the number of samples in the three states does not exceed 10% of the total number of samples. Before training, min-max normalization is performed on the three features. The radial basis function kernel parameters and penalty coefficients are determined using grid search. Candidate penalty coefficients are set to 0.1, 1, and 10, and candidate kernel parameters are set to 0.01, 0.1, and 1. Five-fold cross-validation is performed on each hyperparameter combination, and the combination with the highest average classification accuracy is selected as the final hyperparameters. The training solution uses a sequential minimum optimization algorithm, iterating until the stopping condition is met. The internal loss function of the model is the hinge loss. After training, the support vectors, weights, and biases are saved for online calculation of aging states.

[0034] In step S13, updating the preset battery state model based on the aging state to obtain the calculation benchmark includes: Extract capacity correction coefficients and internal resistance correction coefficients based on the aging state; The target capacity parameter is obtained by performing capacity parameter update calculation on the battery state model according to the capacity correction coefficient, and the target internal resistance parameter is obtained by performing internal resistance parameter update calculation according to the internal resistance correction coefficient. Based on the aging state, parameter analysis is performed to obtain the state covariance, observation matrix, and observation noise parameters; The target capacity parameter, the target internal resistance parameter, the state covariance, the observation matrix, and the observation noise parameter are combined to construct the calculation benchmark.

[0035] It should be noted that, firstly, capacity correction coefficients and internal resistance correction coefficients are extracted based on the aging state. The battery state model involves two basic parameters: rated capacity and initial internal resistance. The rated capacity is given by the battery datasheet, for example, 50 Ah, and the initial internal resistance is also given by the battery datasheet, for example, 2.0 mΩ. During the battery's entire lifespan, capacity decreases with increasing cycle count, and internal resistance increases with increasing cycle count; this change is directly related to the aging state.

[0036] To avoid abrupt changes in target capacity parameters when switching between three fixed correction coefficients, which would reduce the continuous correction accuracy of the ampere-hour integrated state of charge throughout the entire lifespan, this embodiment utilizes the cumulative throughput in the cycle characteristic parameters to continuously calculate the capacity correction coefficient and internal resistance correction coefficient. Continuous mapping curves of "cumulative throughput - capacity retention rate" and "cumulative throughput - internal resistance growth rate" are established in advance through cycle aging tests of the same batch of lithium batteries. The test sample size is no less than 5 batteries, and each battery completes 500 full charge-discharge cycles. The actual capacity, internal resistance, and cumulative throughput are recorded every 50 cycles. Using cumulative throughput as the independent variable and capacity retention rate and internal resistance growth rate as dependent variables, a cubic polynomial is used to fit the continuous mapping curves, with a goodness of fit R² of no less than 0.98. During online use, the cumulative throughput in the current cycle characteristic parameters is obtained, substituted into the fitted curve for interpolation to obtain the current capacity retention rate, and then the capacity correction coefficient is directly determined based on the capacity retention rate: Capacity correction coefficient = Current capacity retention rate; Internal resistance correction coefficient = 1 + Current internal resistance growth rate. The aging state is used to constrain the reasonable range of values ​​for the correction coefficient. If the interpolation result exceeds the capacity retention rate range or internal resistance growth rate range corresponding to that aging state, the correction coefficient is truncated to the boundary value of that range. For example, aging state 0 corresponds to a capacity retention rate of 0.90 to 1.00, so the capacity correction coefficient is truncated to this range; aging state 1 corresponds to 0.80 to 0.90; and aging state 2 corresponds to 0.70 to 0.80. The corresponding truncation range for the internal resistance growth rate is: aging state 0 corresponds to 0 to 0.10, aging state 1 corresponds to 0.10 to 0.30, and aging state 2 corresponds to 0.30 to 0.50. In this way, the capacity correction coefficient changes continuously with the cumulative throughput, eliminating inter-level jumps and achieving continuous self-correction throughout the entire life cycle.

[0037] Next, the target capacity parameter is obtained by updating the battery state model based on the capacity correction factor, and the target internal resistance parameter is obtained by updating the internal resistance parameter based on the internal resistance correction factor. The target capacity parameter equals the rated capacity multiplied by the capacity correction factor. For example, in aging state 1 with a capacity retention rate of 0.83 obtained by interpolation of cumulative throughput, the target capacity parameter is 50Ah multiplied by 0.83, equaling 41.5Ah. The target internal resistance parameter equals the initial internal resistance multiplied by the internal resistance correction factor. For example, in aging state 1 with an internal resistance growth rate of 0.20 obtained by interpolation of cumulative throughput, the target internal resistance parameter is 2.0mΩ multiplied by 1.20, equaling 2.4mΩ. Both the capacity correction factor and the internal resistance correction factor have a dimension of 1. The unit for the target capacity parameter is Ah, and the unit for the target internal resistance parameter is mΩ.

[0038] Secondly, parameter analysis is performed based on the aging state to obtain the state covariance, observation matrix, and observation noise parameters. The parameter analysis is implemented based on the battery state model, which uses a first-order equivalent circuit model. The state vector consists of two components: state of charge (SCC) and polarization voltage. Therefore, the state covariance is a 2×2 matrix, the observation matrix is ​​a 1×2 matrix, and the observation noise parameter is a scalar. The diagonal elements of the state covariance reflect the SCC estimation variance and the polarization voltage estimation variance, respectively. The initial SCC at power-on is obtained by reading the open-circuit voltage from the battery management device. The standard deviation of the SCC error corresponding to the open-circuit voltage measurement error is approximately 10%, so the SCC estimation variance is initialized to 0.01. The initial estimated value of the polarization voltage is 0mV, and its standard deviation during the charge-discharge dynamic process is approximately 200mV; therefore, the polarization voltage estimation variance is initialized to 0.04. The off-diagonal elements of the state covariance are set to 0, indicating that the SCC estimation error and the polarization voltage estimation error are uncorrelated in the initial state. The observation matrix is ​​a 1×2 matrix. The first element is the derivative of the open-circuit voltage with respect to the state of charge, determined by the first-order difference of the voltage state mapping table at the current state of charge. The second element is 1, corresponding to the observation coefficient of the polarization voltage term. The observation noise parameter originates from the voltage sensor measurement noise and the unmodeled error of the model. The measurement accuracy of the voltage sampling link in a mass-produced battery management system is typically ±5mV. Taking three times the standard deviation and squaring it, the baseline observation noise is approximately 2.8mV². Therefore, the observation noise parameter corresponding to aging state 0 is set to 2.8mV². As the aging state increases, the terminal voltage fluctuation intensifies, and the unmodeled error of the model increases. Therefore, the observation noise parameter corresponding to aging state 1 is set to 3.5mV², and the observation noise parameter corresponding to aging state 2 is set to 4.2mV². This parameter can be adjusted within the range of 1.0mV² to 10.0mV² according to the actual accuracy of the voltage sampling link and the degree of terminal voltage fluctuation, based on the noise variance index provided in the voltage sampling chip datasheet.

[0039] Subsequently, the target capacity parameter, target internal resistance parameter, state covariance, observation matrix, and observation noise parameter are combined and constructed to obtain the calculation baseline. The combined construction process writes the five data items into structured storage units in a fixed order, with each field name identical to the above, without any changes. The target capacity parameter is stored as a scalar, the target internal resistance parameter as a scalar, the state covariance as a 2×2 array, the observation matrix as a 1×2 array, and the observation noise parameter as a scalar. This allows for direct reading of the corresponding fields from the calculation baseline during subsequent initial state of charge estimation and dynamic internal resistance compensation.

[0040] In one implementation, the step of performing state mapping calculation to obtain the initial state of charge based on the calculation reference, the terminal voltage change sequence, and the current change sequence includes: The integral state of charge is obtained by performing an integral calculation on the current change sequence based on the capacity parameters in the calculation reference. The ohmic voltage drop is obtained by multiplying the current change sequence with the internal resistance parameter in the calculation reference, and the polarization voltage is obtained by performing polarization analytical calculation based on the pre-acquired battery state model. Based on the ohmic voltage drop and the polarization voltage, voltage compensation calculation is performed on the terminal voltage change sequence to obtain the open-circuit voltage sequence. Based on the open-circuit voltage sequence, interpolation calculation is performed in a pre-acquired voltage state mapping table to obtain the voltage state of charge. The state fusion gain is obtained by performing gain calculation based on the state covariance, observation matrix and observation noise parameters in the calculation benchmark. The integrated charged state and the voltage charged state are fused based on the state fusion gain to obtain the initial charged state.

[0041] It should be noted that, firstly, the current change sequence is integrally calculated based on the target capacity parameter in the calculation benchmark to obtain the integrated state of charge (SPC). The integral calculation uses the ampere-hour integration method. Each sampling point in the current change sequence is multiplied by the sampling period of 0.1s and then accumulated to obtain the cumulative charge in As. This cumulative charge is then divided by 3600 to convert to Ah, and then divided by the target capacity parameter and multiplied by 100% to obtain the integrated SPC in %. For example, if the target capacity parameter is 42.5 Ah, and the average current over 10 seconds is 5A, the cumulative charge is 50As, approximately 0.0139 Ah, corresponding to a SPC change of approximately 0.0327%. This calculation does not involve filtering or normalization; the original current sequence is used directly.

[0042] Next, the ohmic voltage drop is obtained by multiplying the current change sequence with the target internal resistance parameter in the calculation reference. Then, the polarization voltage is obtained by performing polarization analytical calculation based on the pre-acquired battery state model. The battery state model adopts a first-order equivalent circuit model, which includes an ohmic internal resistance branch and a polarization branch, where the ohmic internal resistance branch corresponds to the target internal resistance parameter. The ohmic voltage drop is equal to each current value in the current change sequence multiplied by the target internal resistance parameter, in mV. The polarization voltage is calculated through the discrete recursive equation of the polarization branch in the first-order equivalent circuit model. The polarization internal resistance and polarization capacitance parameters in the recursive equation are given by the battery datasheet during battery state model initialization. In this embodiment, the polarization internal resistance is 1.5mΩ and the polarization capacitance is 1000F. The recursive calculation takes the current change sequence as input and outputs a polarization voltage sequence, in mV.

[0043] Then, based on the ohmic voltage drop and polarization voltage change sequence, voltage compensation calculation is performed to obtain the open-circuit voltage sequence. The voltage compensation formula is:

[0044] in, It is an open-circuit voltage sequence. This is a sequence of terminal voltage changes. Let I(t) be the ohmic voltage drop sequence, I(t) be the current change sequence, and R be the target internal resistance parameter. It is a polarization voltage sequence.

[0045] If I(t) < 0 during discharge, then When I(t) < 0, subtracting a negative value from the terminal voltage is equivalent to adding a positive value; therefore, the open-circuit voltage is higher than the terminal voltage, consistent with the actual physical relationship during discharge. During charging, I(t) > 0, then... A value greater than 0 indicates that the open-circuit voltage is lower than the terminal voltage when the positive value is subtracted from the terminal voltage, consistent with the actual physical relationship during charging. The polarization voltage is negative during discharging (terminal voltage lower than open-circuit voltage) and positive during charging (terminal voltage higher than open-circuit voltage), and is uniformly processed by subtraction. All three sequences have the same length, 1000 × 1, and the unit is mV.

[0046] The voltage state of charge (VCC) is obtained by interpolation calculation based on the open-circuit voltage sequence in a pre-acquired voltage state mapping table. To eliminate dynamic hysteresis and temperature effects, the voltage state mapping table is obtained from a 0.05C low-current charge-discharge experiment before the battery leaves the factory, and open-circuit voltage-VCC mapping sub-tables are established for three temperature points: 5℃, 25℃, and 45℃. During online calculation, the closest temperature sub-table is selected from the surface temperature change sequence for linear interpolation. Before interpolation, the open-circuit voltage needs to be corrected for hysteresis. The hysteresis voltage is experimentally calibrated and is typically between 10mV and 20mV. When the current direction is discharge, the back-calculated open-circuit voltage is lower than the equilibrium open-circuit voltage due to the hysteresis effect, so the hysteresis voltage is added to the calculated open-circuit voltage. When the current direction is charging, the back-calculated open-circuit voltage is higher than the equilibrium open-circuit voltage due to the hysteresis effect, so the hysteresis voltage is subtracted from the calculated open-circuit voltage. The corrected open-circuit voltage is then interpolated again to obtain the voltage VCC, in percentages (%).

[0047] Subsequently, the state fusion gain is calculated based on the state covariance, observation matrix, and observation noise parameters in the calculation benchmark. To address the issue of inconsistent dimensions, volts (V) are uniformly adopted as the unit for voltage-related quantities, while the state of charge (SOC) still uses dimensionless decimals. The state covariance matrix P is 2×2, where the element in the first row and first column is the SOC estimation variance, in the square of a dimensionless decimal; the element in the second row and second column is the polarization voltage estimation variance, in V², with an initial value of 0.04V² based on the square of the 0.2V standard deviation; the initial value of the element in the first row and first column remains 0.01. The observation matrix H is... ,in The first column represents the derivative of the open-circuit voltage with respect to the state of charge, obtained from the first-order difference of the voltage state mapping table at the current SOC, in units of V / dimensionless decimal. The second column corresponds to the polarization voltage, with a coefficient of 1 before the polarization voltage in the observation equation. The observation noise parameter R is in units of V², determined by the voltage sampling link noise. During calculation, the aforementioned mV² values ​​are first converted to V², for example, 2.8 mV² is converted to 2.8 × 10⁻⁶. -6 V². The gain calculation formula is: Where K is a 2×1 vector, P is the state covariance matrix, and H is the observation matrix. Let H be the transpose of H, and R be the observation noise parameter. This calculation directly uses the values ​​from the calculation baseline. The resulting state fusion gain is a 2×1 vector, with the first element used to correct the state of charge estimate and the second element used to correct the polarization voltage estimate. During fusion, the difference between the voltage state of charge and the integrated state of charge is multiplied by the local slope of the open-circuit voltage with respect to the state of charge, and converted into an equivalent observation residual in volts. The initial state of charge is equal to the integrated state of charge plus the first element of the state fusion gain multiplied by this equivalent observation residual, and the polarization voltage estimate is simultaneously added to the second element of the state fusion gain multiplied by this equivalent observation residual.

[0048] In step S14, the step of performing a prediction calculation based on the calculation benchmark to obtain a voltage residual sequence, and combining the surface temperature change sequence with the voltage residual sequence to perform a compensation calculation on the initial state of charge to obtain a target state of charge, includes: Based on the initial state of charge, the current change sequence, and the calculation reference, a state prediction calculation is performed to obtain the predicted terminal voltage sequence. Calculate the algebraic difference between the predicted terminal voltage sequence and the terminal voltage change sequence to obtain the voltage residual sequence; A temperature gradient is obtained by performing a difference operation on the surface temperature change sequence. The pre-acquired observation noise parameters are then dynamically updated based on the temperature gradient and the voltage residual sequence to obtain the target observation noise parameters. The corrected gain is obtained by performing gain extrapolation calculation based on the target observation noise parameters and the state covariance in the calculation benchmark. The target state of charge is obtained by performing linear superposition calculation on the initial state of charge based on the corrected gain and the voltage residual sequence.

[0049] First, state prediction calculations are performed based on the initial state of charge, current change sequence, and calculation benchmark to obtain the predicted terminal voltage sequence. The state prediction calculation uses a first-order equivalent circuit model, which includes an open-circuit voltage source, an ohmic internal resistance branch, and a polarization branch. Starting from the initial state of charge, recursion is performed for each sampling point in the current change sequence. At each step, the open-circuit voltage is first obtained by interpolating the current state of charge in the voltage state mapping table. Then, the open-circuit voltage is added to the ohmic voltage drop obtained by multiplying the current value by the target internal resistance parameter, and finally, the polarization voltage is added to obtain the predicted terminal voltage. The polarization voltage recursion uses a first-order difference equation, with a polarization internal resistance of 1.5 mΩ, a polarization capacitance of 1000 F, and a recursion step size of 0.1 s. The predicted terminal voltage sequence has a dimension of 1000 × 1, and the unit is mV.

[0050] Next, the algebraic difference between the predicted terminal voltage sequence and the terminal voltage change sequence is calculated to obtain the voltage residual sequence. Subtracting the corresponding sampling points from the predicted terminal voltage sequence and the terminal voltage change sequence yields 1000 voltage residuals. The voltage residual sequence has a dimension of 1000×1 and the unit is mV. The voltage residuals reflect the deviation between the battery state model output and the actual measured output.

[0051] Secondly, a differential operation is performed on the surface temperature change sequence to obtain the temperature gradient. Based on the temperature gradient and the voltage residual sequence, the pre-acquired observation noise parameters are dynamically updated to obtain the target observation noise parameters. The differential operation uses backward difference, dividing the difference between two adjacent temperature sampling points in the surface temperature change sequence by the sampling period of 0.1s to obtain 999 temperature gradient values. The temperature gradient dimension is 999×1, and the unit is ℃ / s. The pre-acquired observation noise parameters are derived from the calculation baseline. The baseline observation noise parameters are taken from the observation noise parameters corresponding to the current aging state in the calculation baseline, for example, 2.8×10⁻⁶. -6 V². The extreme value of the temperature gradient is taken as the maximum value among 999 absolute values ​​of the temperature gradient. The standard deviation of the voltage residual is calculated from the voltage residual sequence. The target observation noise parameter is calculated by multiplying the baseline observation noise parameter by the temperature correction factor and then by the voltage residual correction factor. The temperature correction factor is 1 plus the extreme value of the temperature gradient multiplied by 0.1 s / ℃, and the voltage residual correction factor is 1 plus the standard deviation of the voltage residual multiplied by 0.21 / mV. 0.1 s / ℃ and 0.21 / mV are the conversion factors for the temperature gradient and the standard deviation of the voltage residual, respectively, determined based on the normal fluctuation range of the temperature gradient and the standard deviation of the voltage residual, to ensure that the target observation noise parameter is not lower than the baseline observation noise parameter. The target observation noise parameter is a scalar, with the unit being V².

[0052] Subsequently, a corrected gain is calculated by performing gain derivation based on the target observation noise parameters and the state covariance in the calculation benchmark. Then, a linear superposition calculation is performed on the initial state of charge using the corrected gain and the voltage residual sequence to obtain the target state of charge. The corrected gain is calculated using the Kalman filter gain formula. Where K represents the correction gain vector, P represents the state covariance matrix, and H represents the observation matrix. Represents the transpose of the observation matrix. This represents the target observation noise parameter. The correction gain dimension is 2×1, with the first element corresponding to the state of charge correction and the second element corresponding to the polarization voltage correction. During linear superposition calculation, the average value of the voltage residual sequence is first calculated. Then, based on the current initial state of charge, the local slope of the open-circuit voltage with respect to the state of charge is obtained from a pre-acquired voltage state mapping table. The average voltage residual is divided by this local slope to obtain the state of charge correction. The local slope is determined by the first-order difference between two adjacent nodes centered on the current initial state of charge in the voltage state mapping table, in mV / %. This correction method considers the differences in the rate of change of open-circuit voltage with the state of charge in different segments, avoiding inconsistencies in the direction and magnitude of the correction deviation caused by fixed-span conversion. Finally, the initial state of charge is added to the first element of the correction gain and multiplied by the state of charge correction to obtain the target state of charge, which is a scalar in percentage (%).

[0053] In step S14, the step of performing analytical calculation to obtain the dynamic internal resistance based on the terminal voltage change sequence and the current change sequence includes: Based on the current change sequence, a sudden change detection calculation is performed to obtain the current step interval, and based on the current step interval, a time window truncation process is performed on the terminal voltage change sequence to obtain the voltage response interval. Calculate the current change in the current step interval and the instantaneous voltage change in the voltage response interval, and perform a division operation based on the current change and the instantaneous voltage change to obtain the ohmic internal resistance; Calculate the delayed voltage change within the voltage response interval, and calculate the polarization resistance by using the ratio of the difference between the delayed voltage change and the instantaneous voltage change to the current change. Based on the surface temperature change sequence and the aging state, feature fusion compensation calculations are performed on the ohmic internal resistance and the polarization internal resistance to obtain the dynamic internal resistance.

[0054] It should be noted that, firstly, a sudden change detection calculation is performed on the current change sequence to obtain the current step interval. Then, a time window is applied to the terminal voltage change sequence based on the current step interval to obtain the voltage response interval. The sudden change detection calculation traverses all adjacent sampling points in the current change sequence, calculating the absolute value of the current difference between adjacent points. When this absolute value is greater than 1A, the position is recorded as the current step start point. The 1A threshold is set based on the measurement error of the current sensor at a sampling frequency of 10Hz. The sensor's measurement error does not exceed 0.1A, and 1A is 10 times the error, which can stably distinguish between real current steps and noise fluctuations. To avoid interval overlap caused by consecutive steps, after recording a step start point, the subsequent 20 sampling points are skipped before continuing detection. The current step interval is the current data segment of the 50 sampling points before and after the step start point. If the boundary has less than 50 sampling points, it is padded with the maximum available length. The voltage response interval is truncated from the terminal voltage change sequence according to the same index range as the current step interval, and the lengths of the two intervals are consistent.

[0055] Next, the current change during the current step interval and the instantaneous voltage change during the voltage response interval are calculated. The ohmic resistance is obtained by dividing the current change and the instantaneous voltage change. The current change equals the average current of the last 20 sampling points within the current step interval minus the average current of the first 20 sampling points, in amperes (A). The instantaneous voltage change is the voltage of the second sampling point after the step in the voltage response interval minus the voltage of the last sampling point before the step, in mV. The ohmic resistance equals the instantaneous voltage change divided by the current change, in mΩ, because mV divided by A equals mΩ. This calculation uses the raw values ​​directly without filtering or normalization.

[0056] Secondly, the delayed voltage change within the voltage response interval is calculated. The polarization resistance is then calculated by ratioing the difference between the delayed voltage change and the instantaneous voltage change to the current change. The delayed voltage change is taken as the voltage at the 50th sampling point after the step change in the voltage response interval minus the voltage at the last sampling point before the step change, in mV. The polarization resistance is equal to the delayed voltage change minus the instantaneous voltage change, divided by the current change, in mΩ. The difference between the delayed voltage change and the instantaneous voltage change reflects the establishment process of the polarization effect, which has a time scale of 5s, corresponding to the time difference between the 50th sampling point and the 2nd sampling point.

[0057] Subsequently, based on the surface temperature change sequence and aging state, characteristic fusion compensation calculations were performed on the ohmic resistance and polarization resistance to obtain the dynamic internal resistance. Characteristic fusion compensation calculations require first determining the temperature correction coefficient and the aging correction coefficient. Both the temperature correction coefficient and the aging correction coefficient were calibrated based on aging test data from the same batch of lithium batteries. During the calibration process, the ohmic resistance and polarization resistance were measured at four temperature points (25℃, 30℃, 35℃, and 40℃) and aging states 0, 1, and 2, and their ratios to the reference internal resistance at 25℃ and aging state 0 were recorded. The temperature correction factor is the ratio of the reference internal resistance at 25℃ to the average measured internal resistance at each temperature point, used to convert the measured internal resistance to the 25℃ reference operating condition. The values ​​are 1.00, 1.08, 1.17, and 1.26, corresponding to a cumulative relationship where the ionic conductivity increases by approximately 8% for every 5℃ increase in temperature, and the measured internal resistance decreases accordingly. The aging correction factor is the ratio of the average internal resistance of the two types under three aging states, at 1.00, 1.20, and 1.50. The temperature correction factor is 1.00 when the average value of the surface temperature change sequence is less than 25℃, 1.08 when it is between 25℃ and less than 30℃, 1.17 when it is between 30℃ and less than 35℃, 1.26 when it is between 35℃ and 40℃, and 1.26 when it is above 40℃. The aging correction factor is selected corresponding to the current aging state. The converted dynamic internal resistance is not lower than the measured internal resistance under the current operating condition, and is conservatively applied when extrapolating the over-discharge protection voltage drop. The feature fusion compensation calculation is as follows: dynamic internal resistance equals the sum of ohmic internal resistance and polarization internal resistance multiplied by a temperature correction factor, and then multiplied by an aging correction factor. Both ohmic and polarization internal resistances are in mΩ, and both the temperature and aging correction factors are dimensionless ratios. The dynamic internal resistance is also in mΩ. This calculation method simultaneously covers the combined effects of temperature and aging on both ohmic and polarization internal resistances, avoiding systematic biases caused by correcting only a single type of internal resistance. Both the temperature and aging correction factors have been calibrated through actual testing with the same batch of batteries and can be adjusted from the calibrated values ​​by 0.8 to 1.2 times, depending on the battery material system.

[0058] In step S15, if the target state of charge is lower than a preset derating threshold, then the maximum discharge current is calculated based on the target state of charge and the dynamic internal resistance, and an over-discharge protection command is generated, including: If the target state of charge is lower than the derating determination threshold, the remaining available capacity is obtained by multiplying the target state of charge with the capacity parameter in the calculation benchmark. The current open-circuit voltage is obtained by performing a lookup and address matching in a pre-acquired voltage state mapping table based on the target state of charge. Based on the remaining available capacity, a subset of candidate discharge currents is selected from the pre-acquired set of candidate discharge currents. A set of predicted terminal voltages is obtained by performing voltage drop extrapolation calculations based on the current open-circuit voltage, the dynamic internal resistance, and the subset of candidate currents. The predicted terminal voltages in the predicted terminal voltage set are compared with a preset cutoff threshold. Based on the comparison results, the maximum discharge current is obtained by filtering the candidate current subset. An over-discharge protection command is generated based on the maximum discharge current.

[0059] In one implementation, if the threshold value is not lower than the derating determination threshold, a normal operation instruction is generated.

[0060] It should be noted that, firstly, if the target state of charge (SOC) is lower than the derating threshold, the remaining usable capacity is obtained by multiplying the target SOC by the target capacity parameter in the calculation benchmark. The derating threshold is preset to 20%, which is determined based on the maximum depth of discharge and cycle life requirements specified in the battery datasheet and can be adjusted within the range of 15% to 25%. The target SOC is expressed as a percentage, first divided by 100 revolutions per minute to obtain a decimal, and then multiplied by the target capacity parameter to obtain the remaining usable capacity, in Ah.

[0061] Next, the current open-circuit voltage is obtained by performing a lookup and address matching in a pre-acquired voltage state mapping table based on the target state of charge. The voltage state mapping table is constructed before the battery leaves the factory through a 0.05C low-current charge-discharge experiment, recording the correspondence between the state of charge and the open-circuit voltage at 10mV intervals. During the lookup and addressing, the target state of charge is used as the lookup key value. The two closest state of charge nodes are found from the voltage state mapping table, and the current open-circuit voltage is calculated by linear interpolation, with the unit being mV.

[0062] Secondly, a subset of candidate discharge currents is obtained by filtering from the pre-acquired candidate discharge current set based on the remaining available capacity. The candidate discharge current set is determined by the rated capacity and the maximum continuous discharge rate specified in the datasheet. In step S11, the current acquisition range is -100A to 100A. The acquisition link has no effective range support for discharge currents above 100A. Therefore, the upper limit of the candidate discharge current set is constrained by the upper limit of the current acquisition of 100A. When the rated capacity is 50Ah, the corresponding rate is 2C. Twenty candidate current values ​​are generated from 0.1C to 2C with a step size of 0.1C, in units of A. During the screening, the ratio of the remaining available capacity to the rated capacity is calculated. This ratio is used as the upper limit of the allowable rate. Elements in the candidate discharge current set that have a current value greater than the upper limit of the allowable rate are removed to obtain the candidate current subset.

[0063] Subsequently, a set of predicted terminal voltages is calculated based on the current open-circuit voltage, dynamic internal resistance, and candidate current subsets. For each current value in the candidate current subset, the dynamic internal resistance is multiplied by the current value to obtain the internal resistance voltage drop (in mV). The internal resistance voltage drop is then subtracted from the current open-circuit voltage to obtain the predicted terminal voltage corresponding to that current. The predicted terminal voltages corresponding to all candidate currents constitute the set of predicted terminal voltages, with the number of elements in the set matching that of the candidate current subsets.

[0064] Then, the predicted terminal voltages in the predicted terminal voltage set are compared with a preset cutoff threshold. Based on the comparison result, the maximum discharge current is selected from the candidate current subset, and an over-discharge protection command is generated based on the maximum discharge current. The cutoff threshold is preset to 3000mV, which is determined according to the discharge cutoff voltage specified in the battery datasheet and can be adjusted within the range of 2800mV to 3200mV. The candidate current subset and the predicted terminal voltage set are traversed, and candidate currents with predicted terminal voltages greater than the cutoff threshold are selected. The candidate current with the largest current amplitude is taken as the maximum discharge current. If all predicted terminal voltages are not greater than the cutoff threshold, the maximum discharge current is taken as the minimum current value in the candidate current subset, and an over-discharge protection command is generated, limiting the allowable discharge current below the maximum discharge current. The generated over-discharge protection command includes the maximum discharge current limit value.

[0065] In summary, this invention discloses a self-correction method for the state of charge (SOC) of a lithium battery throughout its entire life cycle, which can realize self-correction and safe management of the state of charge of the lithium battery throughout its entire life cycle.

[0066] The second embodiment of the present invention provides a lithium battery full life cycle SOC self-correction system, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the lithium battery full life cycle SOC self-correction method as described above.

[0067] It should be noted that the lithium battery full life cycle SOC self-correction system provided in this embodiment of the invention is used to execute all the process steps of the lithium battery full life cycle SOC self-correction method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0068] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for self-correcting the state of charge (SOC) of a lithium battery throughout its entire life cycle, characterized in that, include: Obtain the terminal voltage change sequence, current change sequence, surface temperature change sequence, and cycle characteristic parameters of the lithium battery; A dynamic trajectory is obtained by performing time-series mapping calculation on the terminal voltage change sequence and the surface temperature change sequence. The dynamic trajectory is calculated by matching the current change sequence and the surface temperature change sequence with a pre-obtained reference trajectory to obtain the trajectory offset. The aging state is obtained by performing deduction calculation in combination with the cyclic feature parameters. The preset battery state model is updated based on the aging state to obtain the calculation benchmark. The initial state of charge is obtained by performing state mapping calculation based on the calculation benchmark, the terminal voltage change sequence and the current change sequence. The dynamic internal resistance is obtained by performing analytical calculation based on the terminal voltage change sequence and the current change sequence. The voltage residual sequence is obtained by performing predictive calculation based on the calculation benchmark. The initial state of charge is compensated by combining the surface temperature change sequence and the voltage residual sequence to obtain the target state of charge. If the target state of charge is lower than the preset derating threshold, the maximum discharge current is calculated based on the target state of charge and the dynamic internal resistance, and an over-discharge protection command is generated; if it is not lower than the derating threshold, a normal operation command is generated.

2. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, The step of performing time-series mapping calculations on the terminal voltage change sequence and the surface temperature change sequence to obtain the dynamic change trajectory includes: The terminal voltage change sequence and the surface temperature change sequence are subjected to resampling alignment and sliding filtering to obtain an aligned time sequence; the aligned time sequence is subjected to window segmentation to obtain a multi-dimensional time input matrix. The multidimensional temporal input matrix is ​​input into a preset temporal inference network to perform nonlinear feature filtering and cross-weighted calculation to obtain the coupled hidden state; Spatial coordinate mapping is performed on the coupled hidden state to obtain single-step trajectory nodes, and temporal splicing is performed on the single-step trajectory nodes to obtain dynamically changing trajectories.

3. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, The process involves matching the current change sequence with the surface temperature change sequence to a pre-obtained reference trajectory, calculating the difference between the dynamic change trajectory and the reference trajectory to obtain the trajectory offset, and then performing a deductive calculation based on the cyclic feature parameters to obtain the aging state, including: The charge / discharge rate features are obtained by performing extraction calculations based on the current change sequence, and the temperature range features are obtained by performing interval mapping calculations based on the surface temperature change sequence. Based on the terminal voltage change sequence and the current change sequence, the state of charge estimation calculation is performed to obtain the state of charge interval characteristics; Based on the charge / discharge rate characteristics, the state of charge range characteristics, and the temperature range characteristics, a baseline trajectory is obtained by traversing and matching from a pre-acquired baseline trajectory set. A spatial distance sequence is obtained by performing path matching calculation between the dynamically changing trajectory and the reference trajectory, and a discrete integral calculation is performed on the spatial distance sequence to obtain the trajectory offset; An aging feature vector is constructed based on the trajectory offset and the cyclic feature parameters. The aging feature vector is then input into a preset mapping classification network to perform state discrimination calculation, thereby obtaining the aging state.

4. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, The step of updating the preset battery state model based on the aging state to obtain the calculation benchmark includes: Capacity correction coefficient and internal resistance correction coefficient are extracted based on the aging state; the capacity parameter is updated and calculated based on the capacity correction coefficient to obtain the target capacity parameter; the internal resistance parameter is updated and calculated based on the internal resistance correction coefficient to obtain the target internal resistance parameter. Based on the aging state, parameter analysis is performed to obtain the state covariance, observation matrix, and observation noise parameters; The target capacity parameter, the target internal resistance parameter, the state covariance, the observation matrix, and the observation noise parameter are combined to construct the calculation benchmark.

5. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, The initial state of charge is obtained by performing state mapping calculation based on the calculation benchmark, the terminal voltage change sequence, and the current change sequence, including: The integral state of charge is obtained by performing an integral calculation on the current change sequence based on the capacity parameters in the calculation reference. The ohmic voltage drop is obtained by multiplying the current change sequence with the internal resistance parameter in the calculation reference, and the polarization voltage is obtained by performing polarization analytical calculation based on the pre-acquired battery state model. Based on the ohmic voltage drop and the polarization voltage, voltage compensation calculation is performed on the terminal voltage change sequence to obtain the open-circuit voltage sequence. Based on the open-circuit voltage sequence, interpolation calculation is performed in a pre-acquired voltage state mapping table to obtain the voltage state of charge. The state fusion gain is obtained by performing gain calculation based on the state covariance, observation matrix and observation noise parameters in the calculation benchmark. The integrated charged state and the voltage charged state are fused based on the state fusion gain to obtain the initial charged state.

6. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, The step of performing a prediction calculation based on the calculation benchmark to obtain a voltage residual sequence, and then combining the surface temperature change sequence with the voltage residual sequence to perform a compensation calculation on the initial state of charge to obtain the target state of charge includes: Based on the initial state of charge, the current change sequence, and the calculation reference, a state prediction calculation is performed to obtain the predicted terminal voltage sequence. Calculate the algebraic difference between the predicted terminal voltage sequence and the terminal voltage change sequence to obtain the voltage residual sequence; A temperature gradient is obtained by performing a difference operation on the surface temperature change sequence. The pre-acquired observation noise parameters are then dynamically updated based on the temperature gradient and the voltage residual sequence to obtain the target observation noise parameters. The corrected gain is obtained by performing gain extrapolation calculation based on the target observation noise parameters and the state covariance in the calculation benchmark. The target state of charge is obtained by performing linear superposition calculation on the initial state of charge based on the corrected gain and the voltage residual sequence.

7. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, The step of performing analytical calculations to obtain the dynamic internal resistance based on the terminal voltage change sequence and the current change sequence includes: Based on the current change sequence, a sudden change detection calculation is performed to obtain the current step interval, and based on the current step interval, a time window truncation process is performed on the terminal voltage change sequence to obtain the voltage response interval. Calculate the current change in the current step interval and the instantaneous voltage change in the voltage response interval, and perform a division operation based on the current change and the instantaneous voltage change to obtain the ohmic internal resistance; Calculate the delayed voltage change within the voltage response interval, and calculate the polarization resistance by using the ratio of the difference between the delayed voltage change and the instantaneous voltage change to the current change. Based on the surface temperature change sequence and the aging state, feature fusion compensation calculations are performed on the ohmic internal resistance and the polarization internal resistance to obtain the dynamic internal resistance.

8. The lithium battery full life cycle SOC self-correction method according to claim 1, characterized in that, If the target state of charge is lower than a preset derating threshold, the maximum discharge current is calculated based on the target state of charge and the dynamic internal resistance, and an over-discharge protection command is generated, including: If the target state of charge is lower than the derating determination threshold, the remaining available capacity is obtained by multiplying the target state of charge with the capacity parameter in the calculation benchmark. The current open-circuit voltage is obtained by performing a lookup and address matching in a pre-acquired voltage state mapping table based on the target state of charge. Based on the remaining available capacity, a subset of candidate discharge currents is selected from the pre-acquired set of candidate discharge currents. A set of predicted terminal voltages is obtained by performing voltage drop extrapolation calculations based on the current open-circuit voltage, the dynamic internal resistance, and the subset of candidate currents. The predicted terminal voltages in the predicted terminal voltage set are compared with a preset cutoff threshold. Based on the comparison results, the maximum discharge current is obtained by filtering the candidate current subset. An over-discharge protection command is generated based on the maximum discharge current.

9. A lithium battery full life cycle SOC self-correction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the lithium battery full life cycle SOC self-correction method according to any one of claims 1 to 8.

Images