A SOC estimation method based on low-voltage side sampling and high-voltage side cooperative correction
By setting up low-voltage sampling nodes and high-voltage side controllers in the battery pack, fault characteristic data is collected, individual cell confidence weights and correction coefficients are evaluated, virtual nodes are constructed to perform local SOC estimation, and global SOC reference values are optimized. This solves the problem of low SOC estimation accuracy of battery management systems under conditions of deterioration in individual cell consistency and water contamination, and achieves high-precision and robust SOC estimation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 智泰新能源(东台)有限公司
- Filing Date
- 2026-06-05
- Publication Date
- 2026-07-03
AI Technical Summary
Existing battery management systems have low SOC estimation accuracy under conditions of deteriorated cell consistency and water contamination, and lack a collaborative correction mechanism between the low-pressure and high-pressure sides, resulting in poor robustness of SOC estimation.
By setting up low-voltage sampling nodes in the battery pack, collecting independent data from individual cells and extracting fault characteristics, and combining the high-voltage side controller to evaluate the confidence weight and correction coefficient of individual cells, virtual nodes are constructed to perform local SOC estimation, optimize the global SOC reference value, and achieve coordinated correction between the low-voltage side and the high-voltage side.
It significantly improves the SOC estimation accuracy and robustness of the battery management system under complex operating conditions, effectively suppresses the interference of cell consistency deterioration and water ingress pollution, and realizes closed-loop collaborative correction of low-pressure side and high-pressure side information.
Smart Images
Figure CN122330740A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery management technology, and in particular to a SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction. Background Technology
[0002] Existing battery management systems typically acquire the voltage, temperature, and total current of each individual cell through low-voltage sampling circuits and calculate the state of charge (SOC) based on the ampere-hour integration method or open-circuit voltage method. However, when there is a significant deterioration in the consistency of individual cells within the battery pack or when external water intrusion occurs, traditional BMSs face the following problems: the differences between individual cells within the pack severely interfere with SOC estimation; the abnormal voltage and sharp increase in internal resistance of failed cells can cause the overall SOC to fail to accurately reflect the usable capacity. Existing technologies often rely on equalization circuits or filtering algorithms to suppress these issues, but these methods are prone to divergence under severe failure conditions; the multiple hidden effects of leakage current, corrosion gas generation, and local thermal field anomalies caused by water intrusion are only used as alarm outputs in traditional BMSs, failing to form a closed-loop correction with SOC estimation; in addition, existing systems typically process low-voltage side sampling and high-voltage side sampling independently, lacking a collaborative mechanism to correct low-voltage side SOC estimation using high-voltage side information.
[0003] Therefore, there is an urgent need to develop a SOC estimation method based on low-pressure side sampling and high-pressure side collaborative correction. This method can utilize the individual cell differences and water pollution characteristics within the battery pack, and coordinate the measurement data from the low-pressure side and the high-pressure side to perform specific correction of SOC. This would overcome the problems of low SOC estimation accuracy and poor robustness of existing technologies under complex operating conditions such as deterioration of individual cell consistency and intrusion of environmental water bodies. Summary of the Invention
[0004] To overcome the shortcomings of existing battery management systems, such as low SOC estimation accuracy and lack of a collaborative correction mechanism between the low-pressure and high-pressure sides under conditions of deteriorated cell consistency and water contamination, this invention provides an SOC estimation method based on low-pressure side sampling and high-pressure side collaborative correction.
[0005] The technical solution is as follows: A SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction, comprising the following steps:
[0006] S1: Set up a low-voltage sampling node for each individual cell in the battery pack, collect independent data of each individual cell at each low-voltage sampling node, and extract fault characteristic data and differential feature sets for each low-voltage sampling node.
[0007] S2: Classify nodes based on the fault feature data to obtain fault level indicators;
[0008] S3: Install a high-voltage side controller in the battery pack, evaluate the individual confidence weight of each low-voltage sampling node based on the differentiated feature set, and dynamically allocate correction coefficients according to the individual confidence weight and fault feature data;
[0009] S4: Based on the correction coefficient, obtain the local SOC estimate of the low-voltage sampling node; construct a virtual node according to the fault level flag to obtain the local SOC estimate of the virtual node; and obtain the global SOC reference value through the local SOC estimate.
[0010] S5: Obtain the local correction residuals of each low-voltage sampling node based on the global SOC reference value, and optimize and update the global parameters based on the local correction residuals.
[0011] Preferably, the step of setting a low-voltage sampling node for each individual cell in the battery pack, collecting independent data for each individual cell at each low-voltage sampling node, and extracting fault characteristic data and differential feature sets for each low-voltage sampling node includes: the independent data for each individual cell includes the terminal voltage, current, surface temperature, and ohmic internal resistance of the individual cell; the fault characteristic data includes voltage deviation characteristics, impedance surge detection characteristics, and individual cell surface temperature; the voltage deviation characteristic is the relative deviation between the mean of the terminal voltage of the corresponding node within the coordination period and the global average voltage issued by the high-voltage side. If the high-voltage side does not issue a global average voltage sequence, the historical voltage sequence of the corresponding node is used instead; the impedance surge detection characteristic is the rate of change of ohmic internal resistance between adjacent coordination periods; the differential feature set includes local outlier factors, dynamic time warping distance, voltage-capacity differential curve characteristic parameters, and moving mean square error.
[0012] Preferably, the differentiated feature set includes local outlier factors, dynamic time warping distance, voltage-capacity differential curve feature parameters, and moving mean square error, including: obtaining the voltage sequence of the corresponding node in the most recent coordinated cycle based on the individual independent data, and obtaining the local outlier factor based on the voltage sequence; obtaining the dynamic time warping distance based on the voltage sequence and the global average voltage sequence of the high-voltage side; obtaining the voltage-capacity differential curve based on the voltage and capacity during the charging process, and extracting the peak position offset and peak height change rate of the voltage-capacity differential curve as the voltage-capacity differential curve feature parameters; obtaining the moving mean square error based on the residual sequence historically received from the high-voltage side of the corresponding node, and using a preset residual default value if the residual sequence length is insufficient.
[0013] Preferably, the step of classifying nodes based on the fault feature data to obtain fault level indicators includes: extracting fault feature values from the fault feature data, wherein the fault feature values include voltage deviation. impedance surge rate and temperature The fault level indicators include normal, warning, and isolation. If any of the fault characteristic values is less than the first fault threshold, the fault level indicator of the node is normal. If any of the fault characteristic values is less than the second fault threshold and greater than the first fault threshold, the fault level indicator of the node is warning. If any of the fault characteristic values is greater than the second fault threshold, the fault level indicator of the node is isolation.
[0014] Preferably, the step of installing a high-voltage side controller in the battery pack and evaluating the individual cell confidence weights of each low-voltage sampling node based on the differentiated feature set includes: defining an individual cell differentiated feature vector set after normalizing the differentiated feature set. ,in for Local outlier factor after node normalization; for The normalized dynamic time-warped distance of the nodes; for The offset of the peak position after node normalization; for Rate of change of peak height after node normalization; for The moving average error after node normalization; the formula is obtained through individual weights. Get Initial individual confidence weights of nodes ,in Sensitivity coefficient; The scaling parameter is used; then, based on the average initial individual cell confidence weights of all low-voltage sampling nodes within the battery pack... Obtain the final individual confidence weights .
[0015] Preferably, the step of dynamically allocating correction coefficients based on the individual confidence weights and fault characteristic data includes: adjusting the individual confidence weights according to the fault level indicator; if the fault level indicator is normal, it remains unchanged; if the fault level indicator is a warning, it adjusts the coefficients according to the anomaly. By adjusting the formula Adjust the confidence weights of individual entities, where To calculate the final individual node confidence weight; if the fault level indicator is isolation, the individual node confidence weight is directly reset to zero and the corresponding node is marked as fault isolated; then, the correction coefficient is obtained based on the adjusted individual node confidence weight and fault feature data. , ;in The maximum permissible deviation is preset; This is the preset maximum allowable surge rate.
[0016] Preferably, obtaining the local SOC estimate of the low-voltage sampling node based on the correction coefficient includes: according to the low-voltage sampling node Construct a dynamic voltage SOC mapping table ,in This is the open-circuit voltage; Surface temperature; The average current is obtained by looking up a table. exist Preliminary SOC estimate at time 1 ; Introducing nodes Correction coefficient residuals of ampere time Corrections are made to obtain local SOC estimates. , ,in For nodes exist The time-dependent residual correction term.
[0017] Preferably, the step of constructing virtual nodes based on the fault level flags to obtain local SOC estimates for the virtual nodes includes: defining nodes with fault level flags of normal and warning as normal nodes, and defining nodes with fault level flags of isolation as isolated nodes; when low-voltage sampling nodes To isolate nodes, create virtual nodes. Replace the isolated node; define a set of normal nodes based on the adjacent normal nodes of the isolated node. The voltage of the virtual node is redefined based on the normal node set. Current ,temperature With correction factor The correction coefficient , For virtual nodes in Correction factor at time; A preset virtual conservative value is used; then, the local SOC estimate of the virtual node is obtained through the dynamic voltage SOC mapping table and the correction coefficient of the virtual node.
[0018] Preferably, obtaining the global SOC reference value through the local SOC estimate includes: defining low-voltage sampling nodes. Local SOC estimation uncertainty of local SOC estimates ,pass ,in For nodes The sliding window standard deviation of the terminal voltage; The preset voltage change threshold; For nodes Surface temperature; Preset reference temperature; The preset temperature change threshold; the local SOC estimation uncertainty of the virtual node. The arithmetic mean of the uncertainties of adjacent nodes is taken; finally, the global SOC reference value is obtained based on the local SOC estimation uncertainty, correction coefficient, and local SOC estimate. .
[0019] Preferably, the step of obtaining the local correction residuals of each low-voltage sampling node based on the global SOC reference value and optimizing and updating the global parameters based on the local correction residuals further includes: obtaining the local correction residuals based on the difference between the global SOC reference value and the local SOC estimates of each normal node, and then obtaining the mean of the local correction residuals; optimizing and updating the global parameters based on the mean of the local correction residuals, wherein the global parameters include a first fault threshold, a second fault threshold, a preset maximum allowable deviation, a preset maximum allowable surge rate, a preset virtual conservative value, a preset voltage change threshold, and a preset temperature change threshold.
[0020] The beneficial effects of this invention are as follows: By combining low-voltage side sampling with high-voltage side collaborative correction, this invention fully utilizes the fault characteristic data and differentiated feature sets of individual cells, dynamically allocates confidence weights and correction coefficients, and can effectively suppress the interference of factors such as the deterioration of intra-cell consistency and water contamination on SOC estimation. At the same time, virtual nodes are constructed for fault isolation nodes to perform local SOC estimation, and residual correction and parameter optimization are performed on each node based on the global SOC reference value, which significantly improves the accuracy, robustness and adaptability of SOC estimation of the battery management system under complex operating conditions, and realizes closed-loop collaborative correction of low-voltage side and high-voltage side information. Attached Figure Description
[0021] Figure 1 This is a flowchart of a SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction.
[0022] Figure 2 This is a schematic diagram of the signal flow and coordination relationship in the SOC estimation method based on low-voltage side sampling and high-voltage side coordinated correction. Detailed Implementation
[0023] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the present invention.
[0024] Example: A SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction, such as... Figure 1 and Figure 2 As shown, it includes the following steps:
[0025] S1: Set up a low-voltage sampling node for each individual cell in the battery pack, collect independent data of each individual cell at each low-voltage sampling node, and extract fault characteristic data and differential feature sets for each low-voltage sampling node.
[0026] The individual cell data includes the terminal voltage, current, surface temperature, and ohmic internal resistance of the individual cell; the fault characteristic data includes voltage deviation characteristics, impedance surge detection characteristics, and cell surface temperature; the voltage deviation characteristic is the relative deviation between the mean of the terminal voltage of the corresponding node within the coordinated cycle and the global average voltage issued by the high-voltage side. If the high-voltage side does not issue a global average voltage sequence, the historical voltage sequence of the corresponding node is used instead; the impedance surge detection characteristic is the rate of change of ohmic internal resistance between adjacent coordinated cycles; the differential feature set includes local outlier factors, dynamic time warping distance, voltage-capacity differential curve characteristic parameters, and moving mean square error.
[0027] Based on the individual independent data, obtain the voltage sequence of the corresponding node in the most recent coordinated cycle, and obtain the local outlier factor based on the voltage sequence; obtain the dynamic time warping distance based on the voltage sequence and the global average voltage sequence of the high-voltage side; obtain the voltage-capacity differential curve based on the voltage and capacity during the charging process, and extract the peak position offset and peak height change rate of the voltage-capacity differential curve as the characteristic parameters of the voltage-capacity differential curve; obtain the moving mean square error based on the residual sequence received historically from the high-voltage side of the corresponding node, and use the preset residual default value if the residual sequence length is insufficient.
[0028] It should also be noted that the voltage deviation feature is obtained by comparing the relative deviation between the average terminal voltage of the low-voltage sampling node within the coordination cycle and the global average voltage sent by the high-voltage side, and then comparing it with the reference voltage average. The impedance surge detection feature is obtained by collecting the ohmic internal resistance value of the low-voltage sampling node in the next coordination cycle, subtracting it from the ohmic internal resistance value of the previous cycle, and then comparing it with the ohmic internal resistance value of the previous cycle. The coordination cycle is dynamically adjusted according to the sliding window standard deviation of the battery pack's total voltage. When the standard deviation of the total voltage exceeds 10mV, the coordination cycle is shortened to 1 / 2, and the initial value is set to 60 seconds. The reference voltage average is calculated by the high-voltage side controller at the end of each coordination cycle, collecting the terminal voltages reported by all low-voltage sampling nodes, removing the maximum and minimum values, and then calculating the arithmetic mean, which is used as the global reference voltage average for that cycle. If the high-voltage side fails to send the voltage, each node temporarily uses the reference voltage average of the previous cycle or its own historical voltage sequence average as a substitute.
[0029] The local outlier factor is obtained by taking the voltage sequence of the low-voltage sampling node in the most recent coordinated cycle, calculating the ratio of the local reachability density of each sampling point to the average local reachability density of other points in the neighborhood, and then taking the average LOF of all points in the cycle as the local outlier factor of that node; the dynamic time warping distance is obtained by dynamically warping and aligning the voltage sequence of the node in the coordinated cycle with the global average voltage sequence sent from the high-voltage side, and calculating the minimum cumulative distance between the two sequences as the dynamic time warping distance; the voltage-capacity differential curve characteristic parameters are obtained by recording voltage and capacity during charging, calculating the curve of the differential of capacity / differential of voltage, extracting the voltage value and height of the main peak position of the curve, and comparing it with the curve in the initial healthy state to obtain the peak position offset and peak height change rate; the curve in the initial healthy state is obtained by calibrating the battery pack during its first use or full charge and discharge. Under a confirmed healthy state without faults, perform a standard constant current charge, record the voltage and capacity, and calculate the curve of the differential of capacity / differential of voltage. Extract the main peak position and peak height as the initial reference curve. If not in a charging state, take the effective value calculated during the most recent charging process. If no charging process occurs after more than 24 hours, reset both parameters to 0. The sliding mean square error is used to maintain the residual sequence received from the high-voltage side in the history of this node. Take the most recent L residual values and calculate their mean square error as the sliding mean square error. If the sequence length is less than L, use the preset residual default value for calculation. L is taken as an integer multiple of the number of residual data points in the collaborative cycle or directly uses the initial value, which is 10. The preset residual default value is obtained by taking the statistical average of the residuals of normal nodes under stable operating conditions during the first operation or calibration phase of the battery pack as the default value. If there is no reference data, the default value is 0.01.
[0030] S2: Classify nodes based on the fault feature data to obtain fault level indicators;
[0031] The fault feature data is extracted to obtain fault feature values, which include voltage deviation. impedance surge rate and temperature The fault level indicators include normal, warning, and isolation. If any of the fault characteristic values is less than the first fault threshold, the fault level indicator of the node is normal. If any of the fault characteristic values is less than the second fault threshold and greater than the first fault threshold, the fault level indicator of the node is warning. If any of the fault characteristic values is greater than the second fault threshold, the fault level indicator of the node is isolation.
[0032] It should also be noted that the first fault threshold is determined by collecting voltage deviation, impedance surge rate, and temperature data of a batch of healthy cells during the battery pack's factory calibration or initial power-on phase. The statistical mean of each data point is then added to twice the standard deviation as the initial value of the first fault threshold. If no calibration data is available, the default values are used: voltage deviation 1%, impedance surge rate 5%, and temperature 40 degrees Celsius. The first fault threshold is determined based on the battery pack's safe operation specifications or accelerated aging test data, using 4-6 times the standard deviation of the statistical mean of each characteristic of healthy cells as the initial value of the second fault threshold. If no calibration data is available, the default values are used: voltage deviation 5%, impedance surge rate 15%, and temperature 55 degrees Celsius. Only when the fault level indicator is normal must all items be less than the corresponding threshold. If only one item exceeds the first fault threshold, its final corresponding fault level indicator is used as the fault level indicator of the node. If multiple items exceed the threshold, the fault level indicator with the highest severity is used, with the severity ranking as: Isolation > Warning > Normal.
[0033] S3: Install a high-voltage side controller in the battery pack, evaluate the individual confidence weight of each low-voltage sampling node based on the differentiated feature set, and dynamically allocate correction coefficients according to the individual confidence weight and fault feature data;
[0034] After normalizing the differential feature set, a single-unit differential feature vector set is defined. ,in for Local outlier factor after node normalization; for The normalized dynamic time-warped distance of the nodes; for The offset of the peak position after node normalization; for Rate of change of peak height after node normalization; for The moving average error after node normalization; the formula is obtained through individual weights. Get Initial individual confidence weights of nodes ,in Sensitivity coefficient; The scaling parameter is used; then, based on the average initial individual cell confidence weights of all low-voltage sampling nodes within the battery pack... Obtain the final individual confidence weights .
[0035] The individual confidence weights are adjusted based on the fault level flag. If the fault level flag is normal, it remains unchanged; if the fault level flag is a warning, it is adjusted according to the abnormality adjustment coefficient. By adjusting the formula Adjust the confidence weights of individual entities, where To calculate the final individual node confidence weight; if the fault level indicator is isolation, the individual node confidence weight is directly reset to zero and the corresponding node is marked as fault isolated; then, the correction coefficient is obtained based on the adjusted individual node confidence weight and fault feature data. , ;in The maximum permissible deviation is preset; This is the preset maximum allowable surge rate.
[0036] It should also be noted that the normalization process for the differentiated feature set uses Min-Max normalization to linearly map each feature to the interval [0,1]; the sensitivity coefficient is based on the dynamic time-normalized distance distribution range of each node under normal battery pack operating conditions, and a typical value of this distance is taken so that the sensitivity coefficient multiplied by the typical value has a weight decay of about 20%. If there is no prior data, the default value of 2.5 is taken; the scale parameter takes the mean of the moving average error of the normal nodes of the battery pack as the reference value, and half of the reference value is taken as the initial value. If there is no data, the default value of 0.02 is taken; the individual confidence weight... ,in To obtain the expected mean, the statistical mean of the initial confidence weights of all nodes under normal operating conditions of the battery pack is taken as the initial value. If historical data is lacking, the default value of 0.8 is used. It should be a very small positive number to prevent the denominator from being zero.
[0037] The confidence weight for a single unit with a fault level indicator of "normal" remains unchanged; the confidence weight for a single unit with a fault level indicator of "warning" remains unchanged. ,in To calculate the final individual confidence weights; The abnormal adjustment coefficient is dynamically calculated based on the degree of voltage deviation, impedance surge rate, and temperature relative to the first fault threshold of the warning node. The weighted average of the excess ratios of the three factors is taken, with the weight set to the default value of 1 / 3. If there is no prior data, the default value of 0.5 is used. The confidence weight of the individual cells with the fault level indicator of isolation is set to zero. The initial value of the preset maximum allowable deviation is the mean of the voltage deviation of all nodes under normal operating conditions of the battery pack plus 4-6 times the standard deviation. If there is no prior data, the default value of 5% is used. The initial value of the preset maximum allowable surge rate is based on the distribution of internal resistance change rate measured by battery aging or water ingress tests, and is taken as 5-10 times the mean of the normal change rate. If there is no data, the default value of 15% is used.
[0038] S4: Based on the correction coefficient, obtain the local SOC estimate of the low-voltage sampling node; construct a virtual node according to the fault level flag to obtain the local SOC estimate of the virtual node; and obtain the global SOC reference value through the local SOC estimate.
[0039] According to the low-pressure sampling node Construct a dynamic voltage SOC mapping table ,in This is the open-circuit voltage; Surface temperature; The average current is obtained by looking up a table. exist Preliminary SOC estimate at time 1 ; Introducing nodes Correction coefficient residuals of ampere time Corrections are made to obtain local SOC estimates. , ,in For nodes exist The time-dependent residual correction term.
[0040] Nodes with fault level flags set to "normal" and "warning" are defined as normal nodes; nodes with fault level flags set to "isolated" are defined as isolated nodes; when a low-voltage sampling node... To isolate nodes, create virtual nodes. Replace the isolated node; define a set of normal nodes based on the adjacent normal nodes of the isolated node. The voltage of the virtual node is redefined based on the normal node set. Current ,temperature With correction factor The correction coefficient , For virtual nodes in Correction factor at time; A preset virtual conservative value is used; then, the local SOC estimate of the virtual node is obtained through the dynamic voltage SOC mapping table and the correction coefficient of the virtual node.
[0041] Define low-voltage sampling nodes Local SOC estimation uncertainty of local SOC estimates ,pass ,in For nodes The sliding window standard deviation of the terminal voltage; The preset voltage change threshold; For nodes Surface temperature; Preset reference temperature; The preset temperature change threshold; the local SOC estimation uncertainty of the virtual node. The arithmetic mean of the uncertainties of adjacent nodes is taken; finally, the global SOC reference value is obtained based on the local SOC estimation uncertainty, correction coefficient, and local SOC estimate. .
[0042] It should also be noted that the dynamic voltage SOC mapping table is constructed by performing low-rate charge-discharge tests on the same type of single battery at a reference temperature of 25°C to calibrate the reference curve of open-circuit voltage versus SOC. Temperature correction coefficients are obtained through experiments at different temperatures. Each low-voltage sampling node directly reuses this reference curve to generate a node-specific mapping table; exist Open circuit voltage at time ,in For nodes exist The terminal voltage at that moment; For nodes exist Current at any given moment; For nodes exist Ohmic internal resistance at time; node The average current is the average current during the cooperative cycle; node exist Time-of-flight residual correction term ,in For nodes exist Current at any moment ; For nodes The rated capacity; the ampere-hour residual correction term at the current moment. The initial value is set to the correction value from the previous time step. The first period is set to 0, and the local SOC estimation formula is substituted first for calculation. Recalculate once more And used for the next cycle.
[0043] When low-pressure sampling node To isolate nodes, construct a normal node set. Prioritize selecting nodes For two directly adjacent nodes, if an isolated node exists among the adjacent nodes, skip the isolated node and continue expanding the search outwards until at most two non-isolated nodes are obtained or the battery pack boundary is reached. If the final set of normal nodes is empty, the construction of virtual nodes is prohibited, and an alarm is recorded; Virtual nodes exist Voltage at time ,in Normal node exist Correction factor at time; Normal node exist Terminal voltage at any given time; virtual node exist Current at any moment ,in Normal node exist Current at any given moment; virtual node exist Temperature of a moment ,in Normal node exist The surface temperature at any given time; in the formula for calculating the correction coefficient of the virtual node, For normal node set The total number of virtual nodes; the preset virtual conservative value is taken as half of the mean of the statistical distribution of the normal node correction coefficient of the battery pack as the initial value, and if there is no prior data, the default value of 0.3 is taken; the ohmic internal resistance of the virtual node is taken as the average of the ohmic internal resistances of the adjacent nodes; the open circuit voltage is calculated based on the obtained voltage, current and ohmic internal resistance of the virtual node, and then the dynamic voltage SOC mapping table is used. The initial SOC estimate is obtained, and finally, the local SOC estimate of the virtual node is obtained by correcting its ampere-hour residual by calculating the correction coefficient of the obtained virtual node.
[0044] The node The sliding window standard deviation of the terminal voltage is taken at the node. The standard deviation of the terminal voltage sequence within the most recent coordinated cycle is calculated as the sliding window standard deviation. The preset voltage change threshold is set based on the typical fluctuation range of the terminal voltage under normal battery charging and discharging conditions, and is taken as 3-5 times the statistical mean of the standard deviation of the sliding window of normal nodes as the initial value. The preset reference temperature is taken as the median value of the normal operating temperature range of the battery pack. The preset temperature change threshold is taken as an empirical value of the allowable temperature difference range or temperature fluctuation amplitude of the battery pack, so that the local SOC estimation uncertainty does not exceed 1 / 2 when deviating from the normal temperature. If there is no data, the default value of 10 degrees Celsius is taken as the initial value. The local SOC estimation uncertainty of the virtual node is calculated as the arithmetic mean after removing the maximum and minimum values of the uncertainties of adjacent normal nodes. The global SOC reference value is used. ,in It is the set of all normal nodes and virtual nodes; For nodes exist The uncertainty of the local SOC estimation at time t.
[0045] S5: Obtain the local correction residuals of each low-voltage sampling node based on the global SOC reference value, and optimize and update the global parameters based on the local correction residuals.
[0046] The local correction residual is obtained based on the difference between the global SOC reference value and the local SOC estimate of each normal node, and then the mean of the local correction residual is obtained; the global parameters are optimized and updated based on the mean of the local correction residual. The global parameters include a first fault threshold, a second fault threshold, a preset maximum allowable deviation, a preset maximum allowable surge rate, a preset virtual conservative value, a preset voltage change threshold, and a preset temperature change threshold.
[0047] It should also be noted that the local corrected residual is obtained by subtracting the estimated local SOC of normal nodes from the global SOC reference value, and then the average of the local corrected residuals of all normal nodes is taken as the mean of the local corrected residuals; for each parameter, a signed incremental adjustment is used for updating, through... Optimization and updates were performed, including The updated parameters; The parameter for the current period; This is the locally corrected mean residual; To adjust the sensitivity coefficient, different fixed empirical values are taken according to the parameter type: 0.05 for the threshold class and 0.1 for the coefficient class. As a global scaling factor, the reciprocal of the standard deviation of the historical sliding window of the residual mean is taken, so that the tanh input corresponding to the residual within a certain standard deviation is approximately equal to 1; and the updated parameters are limited. ,in The initial values of the parameters are set; at the end of each collaborative cycle, the mean of the local correction residual is calculated, and the global parameters are updated; if the number of normal nodes is less than 3, the parameter update is stopped to avoid statistical bias; if the updated parameters exceed the limit or the updated parameters have a first fault threshold greater than the second fault threshold, the parameter values of the previous cycle are forcibly rolled back.
[0048] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction, characterized in that, Includes the following steps: S1: Set up a low-voltage sampling node for each individual cell in the battery pack, collect independent data of each individual cell at each low-voltage sampling node, and extract fault characteristic data and differential feature sets for each low-voltage sampling node. S2: Classify nodes based on the fault feature data to obtain fault level indicators; S3: Install a high-voltage side controller in the battery pack, evaluate the individual confidence weight of each low-voltage sampling node based on the differentiated feature set, and dynamically allocate correction coefficients according to the individual confidence weight and fault feature data; S4: Based on the correction coefficient, obtain the local SOC estimate of the low-voltage sampling node; construct a virtual node according to the fault level flag to obtain the local SOC estimate of the virtual node; and obtain the global SOC reference value through the local SOC estimate. S5: Obtain the local correction residuals of each low-voltage sampling node based on the global SOC reference value, and optimize and update the global parameters based on the local correction residuals.
2. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The method involves setting up low-voltage sampling nodes for each individual cell in the battery pack, collecting independent data for each cell at each low-voltage sampling node, and extracting fault characteristic data and differential feature sets for each low-voltage sampling node. These include: the independent data for each cell containing terminal voltage, current, surface temperature, and ohmic internal resistance; the fault characteristic data containing voltage deviation characteristics, impedance surge detection characteristics, and cell surface temperature; the voltage deviation characteristic being the relative deviation between the mean terminal voltage of the corresponding node within the coordinating cycle and the global average voltage issued by the high-voltage side (if the high-voltage side does not issue a global average voltage sequence, the historical voltage sequence of the corresponding node is used instead); the impedance surge detection characteristic being the rate of change of ohmic internal resistance between adjacent coordinating cycles; and the differential feature set containing local outlier factors, dynamic time warping distance, voltage-capacity differential curve characteristic parameters, and moving mean square error.
3. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 2, characterized in that, The differential feature set includes local outlier factors, dynamic time warping distance, voltage-capacity differential curve feature parameters, and moving mean square error. It includes: obtaining the voltage sequence of the corresponding node within the most recent coordinated cycle based on the individual node's independent data; obtaining the local outlier factor based on the voltage sequence; obtaining the dynamic time warping distance based on the voltage sequence and the global average voltage sequence of the high-voltage side; obtaining the voltage-capacity differential curve based on the voltage and capacity during the charging process; extracting the peak position offset and peak height change rate of the voltage-capacity differential curve as feature parameters; and obtaining the moving mean square error based on the residual sequence historically received from the high-voltage side by the corresponding node. If the residual sequence length is insufficient, a preset residual default value is used.
4. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of classifying nodes and obtaining fault level indicators based on the fault feature data includes: extracting fault feature values from the fault feature data, wherein the fault feature values include voltage deviation. impedance surge rate and temperature The fault level indicators include normal, warning, and isolation. If any of the fault characteristic values is less than the first fault threshold, the fault level indicator of the node is normal. If any of the fault characteristic values is less than the second fault threshold and greater than the first fault threshold, the fault level indicator of the node is warning. If any of the fault characteristic values is greater than the second fault threshold, the fault level indicator of the node is isolation.
5. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of installing a high-voltage side controller in the battery pack and evaluating the individual cell confidence weights of each low-voltage sampling node based on the differentiated feature set includes: defining an individual cell differential feature vector set after normalizing the differentiated feature set. ,in for Local outlier factor after node normalization; for The normalized dynamic time-warped distance of the nodes; for The offset of the peak position after node normalization; for Rate of change of peak height after node normalization; for The moving average error after node normalization; the formula is obtained through individual weights. Get Initial individual confidence weights of nodes ,in Sensitivity coefficient; The scaling parameter is used; then, based on the average initial individual cell confidence weights of all low-voltage sampling nodes within the battery pack... Obtain the final individual confidence weights .
6. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of dynamically allocating correction coefficients based on the individual confidence weights and fault characteristic data includes: adjusting the individual confidence weights according to the fault level indicator; if the fault level indicator is normal, it remains unchanged; if the fault level indicator is a warning, it adjusts the coefficients according to the anomaly. By adjusting the formula Adjust the confidence weights of individual entities, where To calculate the final individual node confidence weight; if the fault level indicator is isolation, the individual node confidence weight is directly reset to zero and the corresponding node is marked as fault isolated; then, the correction coefficient is obtained based on the adjusted individual node confidence weight and fault feature data. , ;in The maximum permissible deviation is preset; This is the preset maximum allowable surge rate.
7. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of obtaining the local SOC estimate of the low-voltage sampling node based on the correction coefficient includes: based on the low-voltage sampling node Construct a dynamic voltage SOC mapping table ,in This is the open-circuit voltage; Surface temperature; The average current is obtained by looking up a table. exist Preliminary SOC estimate at time 1 ; Introducing nodes Correction coefficient residuals of ampere time Corrections are made to obtain local SOC estimates. , ,in For nodes exist The time-dependent residual correction term.
8. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of constructing virtual nodes based on the fault level flags to obtain local SOC estimates for the virtual nodes includes: defining nodes with fault level flags of "normal" and "warning" as normal nodes, and defining nodes with fault level flags of "isolated" as isolated nodes; when low-voltage sampling nodes... To isolate nodes, create virtual nodes. Replace the isolated node; define a set of normal nodes based on the adjacent normal nodes of the isolated node. The voltage of the virtual node is redefined based on the normal node set. Current ,temperature With correction factor The correction coefficient , For virtual nodes in Correction factor at time; A virtual conservative value is preset; then, the local SOC estimate of the virtual node is obtained by using the dynamic voltage SOC mapping table and the correction coefficient of the virtual node.
9. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of obtaining the global SOC reference value through the local SOC estimate includes: defining low-voltage sampling nodes. Local SOC estimation uncertainty of local SOC estimates ,pass ,in For nodes The sliding window standard deviation of the terminal voltage; The preset voltage change threshold; For nodes Surface temperature; Preset reference temperature; The preset temperature change threshold; the local SOC estimation uncertainty of the virtual node. The arithmetic mean of the uncertainties of adjacent nodes is taken; finally, the global SOC reference value is obtained based on the local SOC estimation uncertainty, correction coefficient, and local SOC estimate. .
10. The SOC estimation method based on low-voltage side sampling and high-voltage side collaborative correction according to claim 1, characterized in that, The step of obtaining the local correction residuals of each low-voltage sampling node based on the global SOC reference value and optimizing and updating the global parameters based on the local correction residuals further includes: obtaining the local correction residuals based on the difference between the global SOC reference value and the local SOC estimate of each normal node, and then obtaining the mean of the local correction residuals; optimizing and updating the global parameters based on the mean of the local correction residuals, wherein the global parameters include a first fault threshold, a second fault threshold, a preset maximum allowable deviation, a preset maximum allowable surge rate, a preset virtual conservative value, a preset voltage change threshold, and a preset temperature change threshold.