Parameter self-learning method and system for battery cell equalization and storage medium

Abstract

The invention provides a parameter self-learning method and system for battery cell equalization and a storage medium, and relates to the technical field of battery management systems. The method is applied to a battery management system, and comprises the following steps: monitoring and acquiring operation data including voltage and charging and discharging current of each battery cell in a historical operation cycle of a battery pack; analyzing the operation data, and identifying a change characteristic between a first voltage difference at a starting moment of a typical working condition and a second voltage difference at an ending moment of the typical working condition of the target battery cell and the reference battery cell; according to the change characteristics, dynamically determining an optimized value of at least one equalization control parameter through a self-learning algorithm, the equalization control parameter being used for controlling start and stop of a passive equalization circuit; and performing cell equalization control on the battery pack by using the optimized equalization control parameters.

Description

Technical Field

[0001] This invention relates to the field of battery management system technology, and in particular to a parameter self-learning method, system, and storage medium for cell balancing. Background Technology

[0002] Cell balancing technology is a crucial component of the Battery Management System (BMS) to eliminate inconsistencies among cells in a series-connected battery pack caused by initial differences, uneven operating temperatures, and varying aging rates. This improves the pack's usable capacity, cycle life, and safety. Passive balancing is the most widely used method due to its simple structure and low cost. Traditional passive balancing control typically relies on a set of preset fixed parameter rules. For example, when the difference between a single cell's voltage and its lowest voltage exceeds a certain fixed threshold, balancing is initiated after the system has been allowed to rest and the temperature is within safe limits. Balancing stops when the voltage difference falls below another fixed threshold.

[0003] However, the state of a battery pack in actual operation is dynamic. Battery inconsistency not only intensifies with the increase in the number of cycles, but is also profoundly affected by specific operating conditions (such as charge / discharge rate, depth, and resting time). A fixed set of equalization trigger and exit thresholds is difficult to adapt to the state changes of the battery pack throughout its entire life cycle from initial stage to aging, and it cannot be optimized for different typical application scenarios (such as backup power supplies with frequent float charging or energy storage systems with deep cycles). This static strategy often leads to untimely or inaccurate equalization actions, or unnecessary energy dissipation, which restricts the optimization of equalization efficiency and overall battery pack performance.

[0004] Therefore, the industry needs a more intelligent equalization control method that can break through the limitations of fixed parameters and enable the equalization strategy to be adaptively adjusted according to the actual operating history and performance evolution of the battery pack itself. Summary of the Invention

[0005] The purpose of this invention is to provide a parameter self-learning method, system, and storage medium for battery cell balancing, in order to solve the problem mentioned in the background art: the fixed threshold strategy is difficult to adapt to the dynamic changes of battery packs and the management of the entire life cycle.

[0006] To achieve the above objectives, according to one aspect of the present invention, a parameter self-learning method for cell balancing is provided, the method being applied to a battery management system, the method comprising:

[0007] Throughout the battery pack's historical operating cycle, monitor and acquire operating data, including the voltage and charging / discharging current of each cell.

[0008] Analyze the operational data to identify the variation characteristics between the first voltage difference at the start time and the second voltage difference at the end time of the typical operating condition between the target cell and the reference cell;

[0009] Based on the aforementioned change characteristics, an optimized value for at least one equalization control parameter is dynamically determined using a self-learning algorithm, wherein the equalization control parameter is used to control the start and stop of the passive equalization circuit.

[0010] The optimized equalization control parameters are used to perform cell equalization control on the battery pack.

[0011] In one possible implementation, the equalization control parameters include at least one of an equalization on-state voltage difference threshold, an equalization off-state voltage difference threshold, a resting current threshold for determining the resting state of the battery pack, and a resting time threshold.

[0012] In one possible implementation, the step of dynamically determining the optimized value of at least one equilibrium control parameter based on the changing characteristics using a self-learning algorithm specifically includes:

[0013] Based on the aforementioned variation characteristics, calculate the compensation voltage difference used to compensate for the inconsistency between the target cell and the reference cell at the target state of charge point;

[0014] Based on the compensation voltage difference, the optimized values ​​of the equalization start voltage difference threshold and the equalization stop voltage difference threshold are determined.

[0015] In one possible implementation, the step of calculating the compensation voltage difference for compensating the inconsistency between the target cell and the reference cell at the target state of charge point further includes:

[0016] Based on the preset equalization target mode of the battery pack, select the corresponding calculation strategy;

[0017] The equilibrium target mode includes a top equilibrium mode, a bottom equilibrium mode, or a middle equilibrium mode.

[0018] In one possible implementation, if the target equalization mode is the top equalization mode, the compensation voltage difference is calculated based on the statistical value of the difference between the second voltage difference and the first voltage difference recorded multiple times under the charging condition.

[0019] If the target equalization mode is the bottom equalization mode, the compensation voltage difference is calculated based on the statistical value of the difference between the first voltage difference and the second voltage difference recorded multiple times under the discharge condition;

[0020] If the target balancing mode is the central balancing mode, the compensation voltage difference is calculated based on the statistical value of the absolute value of the voltage difference recorded multiple times under static or low-current steady-state conditions.

[0021] In one possible implementation, the step of dynamically determining the optimized value of at least one equilibrium control parameter through a self-learning algorithm specifically includes:

[0022] Statistical analysis was performed on the distribution of durations during which the system current was below multiple candidate thresholds within the historical operating cycle.

[0023] Based on the duration distribution, candidate thresholds that satisfy preset frequency conditions and / or duration conditions are selected as optimized resting current thresholds.

[0024] In one possible implementation, after determining the optimized value of the resting current threshold, the optimized value of the resting time threshold is dynamically determined, specifically including:

[0025] From a sample of multiple durations corresponding to the static current threshold, a typical value is selected as the optimized static time threshold according to a preset rule.

[0026] In one possible implementation, the method further includes:

[0027] After running an evaluation cycle with the optimized equalization control parameters, the equalization effect evaluation score is calculated based on at least two of the following: voltage consistency improvement, equalization energy consumption, and number of equalization actions.

[0028] Based on historical evaluation scores, multiple sets of equilibrium control parameters are screened, optimized iteratively, or locked.

[0029] According to another aspect of the embodiments of this disclosure, a parameter self-learning system for cell balancing is provided, applied to a battery management system, the parameter self-learning system for cell balancing comprising:

[0030] The data monitoring module is used to monitor and acquire operating data, including the voltage and charging / discharging current of each cell, during the historical operating cycle of the battery pack.

[0031] The feature analysis module is used to analyze the operating data and identify the variation characteristics between the first voltage difference at the start time and the second voltage difference at the end time of the target cell and the reference cell under typical operating conditions.

[0032] The parameter learning module is used to dynamically determine the optimized value of at least one equalization control parameter based on the change characteristics through a self-learning algorithm, wherein the equalization control parameter is used to control the start and stop of the passive equalization circuit.

[0033] The equalization execution module is used to perform cell equalization control on the battery pack using the optimized equalization control parameters.

[0034] According to another aspect of the present disclosure, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements a parameter self-learning method for cell balancing as described in any of the above possible implementations.

[0035] The above-described one or more technical solutions in the embodiments of this application have at least one or more of the following technical effects:

[0036] This invention provides a parameter self-learning method for cell balancing. By analyzing the voltage difference variation characteristics at the start and end times of typical operating conditions within a historical cycle, this method can automatically identify the actual evolution law of battery pack inconsistency, thus overcoming the inherent defect of fixed parameters being disconnected from the real state. Secondly, based on this characteristic, a self-learning algorithm dynamically determines the optimized value of the balancing control parameters, enabling the balancing start and stop thresholds to be adaptively adjusted according to the aging degree and usage mode of the battery pack, ensuring the accuracy of balancing judgment from the initial stage to the aging stage of the battery pack. Finally, using this optimized parameter to execute control, balancing can be initiated in a timely manner when inconsistency increases and exited at an appropriate time. This significantly improves the targeting and timeliness of balancing actions, effectively suppresses voltage divergence, and minimizes unnecessary energy dissipation. Overall, this method significantly improves the usable capacity, cycle life, and operational safety of the battery pack throughout its entire life cycle and in different application scenarios.

[0037] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of a parameter self-learning method for cell balancing provided according to an exemplary embodiment.

[0039] Figure 2 This is a schematic diagram of the composition structure of a parameter self-learning system module for cell equalization according to an exemplary embodiment.

[0040] Figure labeling: 100, Data monitoring module; 200, Feature analysis module; 300, Parameter learning module; 400, Balanced execution module. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this disclosure clearer, the embodiments of this disclosure will be described in further detail below with reference to the accompanying drawings.

[0042] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of systems and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0043] Figure 1 Here is a flowchart of a parameter self-learning method for cell balancing according to an exemplary embodiment, as shown below. Figure 1 As shown, the method is applied to a battery management system (BMS), and the method includes:

[0044] In step S100, during the historical operating cycle of the battery pack, operating data including the voltage of each cell and the charging and discharging current are monitored and acquired; the historical operating cycle can be defined based on duration, number of cycles, or cumulative throughput capacity.

[0045] Specifically, the data monitored and recorded by the BMS includes at least the terminal voltage of each individual cell in the series-connected battery pack, as well as the charging and discharging current flowing through the battery pack. Cell voltage is a key parameter directly characterizing its state of charge and inconsistencies; while the charging and discharging current is used to accurately determine the battery pack's operating state, such as charging, discharging, or resting, and provides a basis for subsequent identification of typical operating conditions for analysis. Preferably, the BMS also simultaneously records the temperature data of each cell or battery module and a precise timestamp; the temperature data ensures the safety of subsequent equalization operations and can serve as an auxiliary analytical factor; the timestamp ensures that all data has a temporal sequence, thereby enabling the reconstruction of the battery pack's operating history.

[0046] During data acquisition, the BMS synchronously samples the voltage of all cells at a certain sampling frequency through its analog front-end (AFE) circuit, and performs necessary filtering on the sampled values ​​to suppress noise; current data is also collected synchronously. This operational data is stored in real-time in the BMS's non-volatile memory, forming a time-stamped historical operational data sequence. This data sequence constitutes the original database for subsequent self-learning analysis, aiming to fully capture the dynamic process of the battery pack's inconsistencies evolving with operating conditions, time, and aging during actual use, thus laying the data foundation for overcoming the limitations of fixed-parameter strategies.

[0047] In step S200, the operating data is analyzed to identify the variation characteristics between the first voltage difference at the start time and the second voltage difference at the end time of a typical operating condition between the target cell and the reference cell. This analysis process first identifies typical operating condition segments from the historical operating data sequence that clearly reflect the performance differences between cells and occur frequently. The system analyzes the charging and discharging current ( The magnitude, direction, and frequency of change of the charge / discharge pattern are used to identify and extract the typical operating conditions, such as identifying the most frequent charging / discharging mode during daily operation and the time interval between charging / discharging events. ), and the system's usual static current level.

[0048] In one specific embodiment, the typical operating condition can be further refined to the stage where the battery pack is in constant current charging. By analyzing current data, the system identifies continuous time periods in which the current value exceeds a preset threshold and remains relatively stable, thereby defining the start and end times of this typical operating condition.

[0049] After determining the typical operating condition, the system performs feature analysis on this segment. First, it identifies the target cell and a reference cell as a comparison benchmark. For example, under constant current charging, the target cell could be the cell with the highest voltage at the end of charging, and the reference cell could be the cell with the lowest voltage under this condition. Subsequently, the system retrieves and reads the voltage values ​​of the target cell and the reference cell at the beginning of the typical operating condition from the stored data, and calculates their difference to obtain the first voltage difference (…). Similarly, retrieve and read the voltage values ​​at the end of a typical operating condition (e.g., the end of the constant current charging phase, the switch to constant voltage charging, or the moment the charging cutoff condition is reached), calculate the difference, and obtain the second voltage difference. ).

[0050] By calculating the difference between the second voltage difference and the first voltage difference, i.e. This allows us to obtain a core characteristic value of change. This characteristic value of change ( The value has a clear physical meaning: a positive value indicates that during this typical operating condition, the voltage difference between the target cell and the reference cell is further widened, and the battery pack inconsistency shows a divergent trend; a negative value indicates that the voltage difference between the two is reduced, and the inconsistency shows a convergent trend. By statistically analyzing the changing characteristic values ​​corresponding to multiple consecutive historical typical operating condition segments, the system can quantitatively evaluate the overall evolution law and rate of battery pack inconsistency at the current stage. At the same time, the extracted information about typical operating condition modes, such as frequency and interval, provides contextual basis directly related to the actual usage habits of the product for the subsequent self-learning optimization of equalization control parameters.

[0051] In step S300, based on the change characteristics, an optimized value of at least one equalization control parameter is dynamically determined through a self-learning algorithm. The equalization control parameter is used to control the start and stop of the passive equalization circuit. Specifically, the equalization control parameter includes, but is not limited to, the voltage difference threshold for triggering equalization action and the voltage difference threshold for stopping equalization action.

[0052] The self-learning algorithm is configured to analyze the changing characteristics to determine the macroscopic evolution trend of battery pack inconsistency under typical operating conditions. Specifically, the algorithm adaptively adjusts the equalization control parameters based on the trend direction and degree represented by the changing characteristics.

[0053] When the analysis indicates that, under continuous typical operating conditions, the changing characteristics consistently indicate a diverging trend in battery pack inconsistency, the self-learning algorithm generates optimized instructions to lower the equalization initiation threshold. This allows the equalization control system to intervene more sensitively in the early stages of widening voltage differences, thereby enhancing its ability to suppress inconsistency divergence.

[0054] Conversely, when the analysis indicates that the changing characteristics continuously suggest a convergence trend in battery pack inconsistency, the self-learning algorithm generates optimized instructions to increase the equalization initiation threshold. This allows the equalization control system to reduce unnecessary equalization operations when the cell conditions are relatively consistent, thereby reducing additional energy loss in the system.

[0055] Through the above method, the self-learning algorithm enables the balancing control parameters to move away from preset fixed values ​​and become dynamically adjusted variables that follow the actual operating state and aging process of the battery pack. The optimized parameters are applied to the real-time balancing control logic, achieving adaptive matching between the balancing strategy and the actual needs of the battery pack.

[0056] In step S400, the optimized equalization control parameters are used to perform cell equalization control on the battery pack. Specifically, the BMS continuously and synchronously monitors the real-time voltage of each cell. During the control process, the system compares the voltage difference between the target cell and the reference cell, calculated in real time, with the optimized equalization start threshold. When it is determined that the real-time voltage difference exceeds the optimized start threshold and simultaneously meets the system's preset allowable equalization conditions, the BMS generates a control command to trigger the passive equalization discharge circuit of the target cell, initiating the equalization process.

[0057] During the equalization process, the BMS continuously monitors changes in the real-time voltage difference. When the voltage difference decreases to be equal to or below the optimized equalization stop threshold, the BMS generates a corresponding control command to shut down the equalization discharge circuit for the target cell, ending the equalization operation. In this way, the optimized parameters directly determine the start and stop criteria for the equalization action, thereby achieving targeted management of battery pack inconsistencies.

[0058] By analyzing the voltage difference variation characteristics at the start and end points of typical operating conditions throughout historical cycles, this method can automatically identify the actual evolution patterns of battery pack inconsistencies, thus overcoming the inherent defect of fixed parameters being disconnected from real-world conditions. Secondly, based on this characteristic, a self-learning algorithm dynamically determines the optimized values ​​of the equalization control parameters, enabling the equalization start / stop thresholds to adaptively adjust with the battery pack's aging level and usage patterns, ensuring the accuracy of equalization judgments from the initial stage to the aging phase. Finally, using these optimized parameters for control execution allows for timely initiation of equalization when inconsistencies increase and timely exit at appropriate times. This significantly improves the targeting and timeliness of equalization actions, effectively suppresses voltage divergence, and minimizes unnecessary energy dissipation. Overall, this method significantly improves the battery pack's usable capacity, cycle life, and operational safety throughout its entire lifecycle and under different application scenarios.

[0059] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] In an exemplary embodiment, the equalization control parameters include at least one of an equalization on-state voltage difference threshold, an equalization off-state voltage difference threshold, a resting current threshold for determining the battery pack's resting state, and a resting time threshold. The battery management system performs self-learning optimization on the above parameters by analyzing the feature dataset extracted from the historical operating data sequence.

[0061] Specifically, for the static state determination parameters, the system optimizes the static current threshold and static time threshold by analyzing recorded quasi-static state samples, i.e., a series of current values ​​when the system current is maintained at a low level and their corresponding stable durations. The battery management system collects the system current through a current sensor, such as a shunt or a Hall sensor. In actual operation, due to factors such as zero drift and static load power consumption, the system current is not fixed when there is no active charging or discharging. Therefore, the system monitors current values ​​and records all stable periods with small absolute current values ​​and their durations, forming a raw sample dataset about the quasi-static state, including a static current sample set. and the corresponding set of static time samples .

[0062] The system performs statistical analysis on the distribution of these sample datasets and selects frequently occurring feature values ​​that meet the cell voltage stability requirements within the historical operating cycle as optimized thresholds. In one specific implementation, the system analyzes the current sample set to identify the current level that occurs most frequently. For example, if the analyzed sample set... If cases where the absolute value of the current is no greater than 2A occur frequently, then the optimized resting current threshold will be determined through self-learning. Subsequently, the system selected all static current samples whose values ​​did not exceed the optimized threshold. The corresponding time sample, for example, to obtain a time subset. The system selects a typical value from the time subset as the optimized settling time threshold based on preset rules, such as balancing frequency of occurrence with engineering effectiveness. For example, the median value might be chosen. As a typical value. Through this process, the static determination parameter is adjusted from a preset fixed value (such as...). The adaptive adjustment is made to an optimized value that better reflects the actual load characteristics of the product (e.g., ).

[0063] For the equalization voltage difference threshold, the system analyzes the recorded paired initial and final voltage difference samples, and combines this with the corresponding charging and discharging current direction information, to optimize the equalization start-up voltage difference threshold. ) and equalization shutdown voltage difference threshold ( Traditional solutions using fixed thresholds, such as an on threshold of 50mV and an off threshold of 30mV, have limitations. They cannot adapt to the differences in various cell systems, such as LFP cells with a distinct voltage plateau and NMC cells with a less pronounced plateau. Furthermore, they struggle to handle diverse actual operating conditions and different balancing strategies, such as non-full charge / discharge, top-side balancing, and bottom-side balancing. Therefore, this system continuously records system current data. Operating condition initial pressure difference dataset and the differential pressure dataset at the end of the operating condition The data is then aggregated and analyzed to achieve self-learning optimization of the threshold.

[0064] The system first identifies the natural evolution trend of cell inconsistency under typical charge and discharge conditions. For example, by calculating the final voltage difference under specific operating conditions (… ) and initial pressure difference ( The difference between () By combining this with the current direction, it can be determined whether the voltage difference tends to increase at the end of charging or decrease at the end of discharging. Based on this trend and the product's target balancing strategy, such as top-end balancing or bottom-end balancing, the system calculates an optimized threshold that effectively promotes uniformity of cell voltage within the target state of charge range. For example, to emphasize the bottom-end balancing strategy for uniformity at the end of discharging, the system can, based on statistical analysis of discharge data, self-learn and set the balancing off voltage difference threshold to the expected voltage difference value required to achieve uniformity at the end of discharging, and add a hysteresis margin to determine the balancing on voltage difference threshold. In this way, the balancing voltage difference threshold can be personalized and adaptively adjusted according to the characteristics of the cell system and the actual usage mode.

[0065] In this way, the battery management system can free key equalization control parameters from the limitations of fixed preset values ​​and transform them into adaptive variables that are dynamically generated and continuously optimized based on the battery pack's own historical operating performance and specific application requirements.

[0066] In an exemplary embodiment, the step of dynamically determining the optimized value of at least one equilibrium control parameter based on the change characteristics using a self-learning algorithm specifically includes:

[0067] Based on the aforementioned variation characteristics, a compensation voltage difference is calculated to compensate for the inconsistency between the target cell and the reference cell at the target state of charge point; and based on the compensation voltage difference, optimized values ​​for the equalization turn-on voltage difference threshold and the equalization turn-off voltage difference threshold are determined.

[0068] The calculation of the compensation differential pressure is based on statistical analysis of paired initial and final differential pressure sample sequences recorded in historical operating data, combined with corresponding charging and discharging current direction information. The system analyzes these changing characteristics, i.e. By analyzing statistical patterns under different typical operating conditions, the natural evolution trend of cell inconsistencies toward the target SOC point can be identified. For example, in a system configured in bottom-end equalization mode (target SOC point is the discharge end), by analyzing data under discharge conditions, the natural convergence of the voltage difference from the charging end to the discharging end can be calculated. This convergence constitutes the key calculation basis for the compensation voltage difference.

[0069] Specifically, in one application example, the system analyzes paired initial and final voltage difference sample sequences recorded in historical operating data. It finds that the typical voltage difference of a certain battery cell at the end of charging is 45mV, while after a complete discharge to the end, its typical voltage difference converges to 27mV. Therefore, it can be inferred that to achieve consistency at the end of discharge, the voltage difference that needs to be pre-compensated at the end of charging through equalization is approximately 45mV - 27mV = 18mV. This 18mV is the compensation voltage difference calculated for this battery cell and this usage mode.

[0070] Subsequently, an optimized value for the equalization voltage difference threshold is determined based on the compensated voltage difference. In one specific implementation, the optimized value of the equalization shutdown voltage difference threshold is set to a value equal to or related to the compensated voltage difference, such as 18mV. The purpose is to stop equalization when the real-time voltage difference decreases to this optimized threshold through equalization, so that the cell voltage tends to be consistent when it subsequently reaches the target SOC point. To prevent frequent start-up and shutdown of the equalization circuit, the optimized value of the equalization start voltage difference threshold is obtained by adding a preset hysteresis margin to the optimized value of the equalization shutdown threshold, for example, by adding 10mV, i.e., 28mV. Through the above process, the equalization voltage difference threshold is optimized from a fixed preset value to an adaptive value closely bound to the actual inconsistency evolution law of the battery pack and the specific application goals.

[0071] In an exemplary embodiment, the step of calculating the compensation voltage difference for compensating the inconsistency between the target cell and the reference cell at the target state of charge point further includes:

[0072] Based on the preset equalization target mode of the battery pack, select the corresponding calculation strategy;

[0073] The equilibrium target mode includes a top equilibrium mode, a bottom equilibrium mode, or a middle equilibrium mode. Different modes define different target SOC points, and thus determine the core data characteristics and analysis logic upon which the calculation of compensation pressure difference depends.

[0074] Specifically, when configured in top-balancing mode, the goal is to ensure voltage consistency of the battery pack at the end of charging, where the charging end represents a high SOC point, such as 100% SOC. Top-balancing mode is suitable for scenarios such as backup power supplies that are in a long-term float charging state. In this mode, the self-learning algorithm focuses on analyzing typical charging segments from a lower SOC point to a higher SOC point in historical data. The algorithm calculates this by statistically analyzing the increase in the voltage difference between the target cell and the reference cell from the beginning (first voltage difference) to the end (second voltage difference) during such charging processes. The typical value of ) is used as the core basis for calculating the compensation pressure difference required to achieve top uniformity.

[0075] When configured in bottom-level balancing mode, the goal is to ensure voltage consistency of the battery pack at the end of discharge, where the discharge end represents a low SOC point, such as 20% SOC. Bottom-level balancing mode is suitable for scenarios such as energy storage systems requiring deep cycling to maximize discharge capacity. In this mode, the self-learning algorithm focuses on analyzing typical discharge segments from high SOC to low SOC in historical data. The algorithm calculates this by statistically analyzing the change in voltage difference from the start to the end of such discharge processes. The typical value of ) is negative, indicating convergence, and is used as the core basis for calculating the compensation pressure difference required to achieve bottom uniformity.

[0076] When configured in the mid-range equalization mode, the goal is to maintain voltage consistency of the battery pack within a common SOC plateau region, such as 30% to 70% SOC. In this mode, since the voltage is not sensitive to changes in SOC, the algorithm instead analyzes the statistical distribution characteristics of cell voltage difference samples recorded by the system during quiescent or low-current steady-state operation, such as the fluctuation range, as an important basis for determining the compensation voltage difference or directly setting relevant thresholds.

[0077] By introducing the equilibrium target mode and selecting the corresponding calculation strategy accordingly, the self-learning algorithm enables the calculation process of the compensation pressure difference and the final optimized equilibrium pressure difference threshold to be accurately matched with the actual functional requirements of the product and the preset optimization target, thereby significantly improving the pertinence and overall effectiveness of the equilibrium control strategy.

[0078] In an exemplary embodiment, the calculation strategy for the compensation pressure difference is closely related to the equilibrium target mode and is implemented based on statistical analysis of historical data of the corresponding type of operating condition. The system intelligently selects the most relevant data source and statistical method to calculate the compensation pressure difference according to the selected equilibrium target mode.

[0079] If the target equalization mode is the top equalization mode, the compensation voltage difference is derived based on the statistical value of the difference between the second voltage difference and the first voltage difference recorded multiple times under charging conditions. Specifically, the self-learning algorithm extracts the first voltage difference and the second voltage difference recorded in multiple complete charging processes from historical data, and calculates a sequence of the differences between the two in each charging process. By performing statistical analysis on this difference sequence, such as calculating the average, median, or a specific percentile value considering distribution characteristics, a statistical value characterizing the typical degree of voltage difference expansion during charging is obtained. This statistical value or its derivative (such as multiplied by an empirical coefficient) is used as the core basis for calculating the compensation voltage difference.

[0080] If the target equalization mode is the bottom-end equalization mode, the compensation voltage difference is derived based on the statistical value of the difference between the first voltage difference and the second voltage difference recorded multiple times under discharge conditions. Specifically, the self-learning algorithm extracts the first voltage difference and the second voltage difference recorded in multiple complete discharge processes from historical data, and calculates a sequence of the differences between the two in each discharge process. By performing statistical analysis on this difference sequence, a statistical value characterizing the typical convergence or expansion of the voltage difference during the discharge process is obtained. This statistical value or its derived value is used as the core basis for calculating the compensation voltage difference.

[0081] If the target equilibrium mode is the central equilibrium mode, the compensation voltage difference is derived based on the statistical value of the absolute value of the voltage difference recorded multiple times under static or low-current steady-state conditions. In this mode, since the voltage in the plateau region is insensitive to changes in SOC, the calculation of the compensation voltage difference instead relies on voltage difference data collected when the system is in a relatively stable state, such as static or low-current steady-state operation. The self-learning algorithm extracts a sequence of absolute values ​​of the voltage difference between each cell and the reference cell, recorded multiple times under static or low-current steady-state conditions. By performing statistical analysis on this absolute value sequence—for example, observing its distribution, calculating its standard deviation, or the range at a specific confidence level—a statistical value characterizing the typical fluctuation level of the voltage difference under this steady-state condition is obtained. This statistical value or its derivative is used as an important basis for determining the compensation voltage difference or directly setting the relevant equilibrium threshold.

[0082] In this way, the self-learning algorithm can intelligently select the most relevant historical data sources and statistical analysis methods according to different equilibrium objectives, thereby calculating the compensation pressure difference that best matches a specific application scenario, laying the foundation for the accurate optimization of the subsequent equilibrium threshold.

[0083] In an exemplary embodiment, the step of dynamically determining the optimized value of at least one equilibrium control parameter using a self-learning algorithm specifically includes:

[0084] The system management system performs statistical analysis on the duration distribution of system current below multiple candidate thresholds during the historical operating cycle. Specifically, the system management system adaptively determines the static current threshold by statistically analyzing the system current data collected during the historical operating cycle.

[0085] The battery management system is configured to: first, based on historical current data, identify and extract a series of stable events where the system current value is lower than multiple different candidate current thresholds, and record the duration of each event. This data constitutes a set of information regarding the distribution of current level and duration in a quasi-static state.

[0086] Based on the duration distribution, candidate thresholds that satisfy preset frequency and / or duration conditions are selected as optimized resting current thresholds. The self-learning algorithm performs statistical analysis on the above distribution information. The algorithm analyzes the frequency of stable events corresponding to each candidate current threshold, as well as the distribution characteristics of the durations of these events, such as the median duration, common value range, or cumulative probability. The optimized resting current threshold is not fixed in advance, but is dynamically selected based on the statistical analysis results. The selection criterion is to select a candidate current threshold that satisfies the preset frequency and / or preset duration conditions. The preset frequency condition may require that the current threshold be frequently touched or exceeded to ensure that the resting criterion has practical significance and sufficient triggering opportunities; the preset duration condition may require that the system can remain stable for a sufficiently long time under the current threshold to ensure that the cell polarization is effectively relaxed and the voltage tends to stabilize.

[0087] For example, in a specific learning process, the system might discover that when 2A is used as a candidate threshold, events with system current below this value occur significantly more frequently than with other candidate values, and most of these events last for more than 0.5 hours, thus meeting the voltage stability requirements. Therefore, the self-learning algorithm determines the optimized resting current threshold to be 2A, replacing a more lenient or stringent value that might have been set empirically in the initial design. In this way, the resting current determination condition is precisely matched to the load characteristics and usage habits of the product in actual operation.

[0088] In an exemplary embodiment, after determining the optimized value of the resting current threshold, the optimized value of the resting time threshold is dynamically determined, specifically including:

[0089] From a sample of multiple durations corresponding to the settling current threshold, a typical value is selected as the optimized settling time threshold according to a preset rule. Specifically, the battery management system filters out all operating segments with system currents lower than or equal to the optimized settling current threshold from historical operating data, and extracts the durations corresponding to these segments to form a duration sample set. The self-learning algorithm then determines a typical value from this set as the optimized settling time threshold according to a preset selection rule.

[0090] The preset rule aims to select a representative time that is reasonable in engineering and effective in actual operation. For example, the rule could be: selecting the median of the duration sample set; or, after excluding samples with excessively short durations, such as those shorter than the preset minimum effective time, selecting the duration value with the highest frequency; or, selecting the minimum duration that can cover a preset proportion of static events in historical operation.

[0091] In this way, the resting time threshold is not a fixed preset, but is dynamically learned by analyzing the stable duration that the battery pack typically maintains after reaching the resting current condition in actual use. This ensures that the system's determination of effective resting not only meets the actual requirements of cell voltage stability but also matches the actual usage mode of the product, thereby improving the accuracy and reliability of determining the timing of the equalization action.

[0092] In an exemplary embodiment, the method further includes:

[0093] After running an evaluation cycle with the optimized equalization control parameters, the equalization effect evaluation score is calculated based on at least two of the following: voltage consistency improvement, equalization energy consumption, and number of equalization actions. The evaluation cycle can be a preset fixed duration, such as 72 hours, or a complete equalization event from start to stop. At the end of the cycle, the system calculates the equalization effect evaluation score for this parameter combination based on one or more preset quantitative indicators. These quantitative indicators include, but are not limited to: voltage consistency improvement, total energy consumption of the equalization process, and the number of equalization actions triggered. Voltage consistency improvement, for example, is the reduction in the voltage difference between key cells at the end of the evaluation cycle compared to the beginning of the cycle. The system calculates the evaluation score by weighting or regularizing at least two of the above indicators.

[0094] Based on historical evaluation scores, multiple sets of equalization control parameters are screened, optimized iteratively, or locked. The battery management system stores and analyzes the parameter combinations and their evaluation scores corresponding to multiple evaluation cycles as historical records. Based on this historical evaluation data, the system performs subsequent processing on multiple sets of equalization control parameters, including: screening and eliminating parameter combinations with poor performance; optimizing and iterating parameter values ​​or their combinations according to scores and parameter distribution patterns to explore better solutions; and locking specific parameter combinations that have been repeatedly verified and whose evaluation scores are stable at a high level, storing them in the knowledge base as the preferred solution under specific operating conditions.

[0095] By introducing the aforementioned effect evaluation and iterative optimization mechanism, this method achieves a closed loop from parameter learning and application to effect feedback optimization. This enables the balancing strategy not only to undergo initial self-learning adaptation but also to continuously self-correct and evolve based on feedback from actual usage throughout the entire battery system's lifecycle, thereby ensuring the long-term optimality of balancing performance.

[0096] For example, in a specific learning process, the system records multiple sets of parameter and performance data after a complete learning cycle, such as:

[0097] {Quiet current, settling time, on-threshold threshold, off-threshold threshold, remaining differential pressure}

[0098] Record 1: {3A, 0.5H, 45mV, 27mV, 18mV}

[0099] Record 2: {3A, 0.7H, 66mV, 22mV, 13mV}

[0100] Record 3: {3A, 1H, 72mV, 24mV, 11mV}

[0101] Record 4: {2A, 0.6H, 102mV, 32mV, 16mV}

[0102] Record 5: {2A, 1H, 45mV, 18mV, 7mV}

[0103] Record 6: {2A, 0.5H, 32mV, 22mV, 12mV}

[0104] By comparing and analyzing this data, the system can comprehensively deduce the optimal parameter range for achieving the minimum residual voltage difference under different quiescent current conditions. For example, the analysis may show that: under the condition of a quiescent current of 3A, the balancing effect is optimal when the quiescent time is approximately 0.7 hours, the equalization start-up voltage difference threshold is approximately 66mV, and the equalization stop-up voltage difference threshold is between 9mV and 22mV; under the condition of a quiescent current of 2A, the balancing effect is optimal when the quiescent time is approximately 0.5 hours, the equalization start-up voltage difference threshold is approximately 32mV, and the equalization stop-up voltage difference threshold is between 12mV and 22mV.

[0105] Based on the results of this in-depth analysis, the system performs filtering, optimization iterations, or locking operations on parameter combinations. For specific parameter combinations that have been verified multiple times and have consistently excellent evaluation scores, the system locks them as the optimal solution for the corresponding operating conditions.

[0106] Parameter iteration and storage are reflected in the fact that the system records the optimal parameter combination obtained from the self-learning optimization in non-volatile memory, forming a self-learning parameter knowledge base that continuously grows and updates throughout the product's lifetime operation. When the system runs subsequently, it can first match the current operating conditions and directly call the historically optimal parameter combination from the knowledge base for application, thereby achieving rapid adaptation and performance optimization. This process is continuously online, enabling the balancing strategy to continuously evolve with the customer's actual usage conditions, achieving continuous self-learning and adaptive optimization of balancing parameters throughout the entire battery system lifecycle.

[0107] It is understandable that the above methods and control logic achieve self-learning by recording and analyzing a large amount of operating data during long-term use by the battery management system. The self-learning control concept and logic presented in this invention are applicable to the passive equalization control of battery cells with different chemical systems, and can be adapted to different equalization target modes such as top equalization, bottom equalization, or middle equalization according to the needs of actual application scenarios.

[0108] In an exemplary embodiment, the framework of the self-learning algorithm is further refined into a process including data preprocessing, pattern decision-making, and parameter calculation. First, in the data clustering and operating condition labeling stage, the battery management system automatically analyzes and classifies the collected operating data sequences, including current, initial differential pressure, and final differential pressure, based on the characteristics of current magnitude, direction, and duration, labeling data segments with operating condition tags such as high-current charging, low-current resting, and typical discharge. Subsequently, in the pattern recognition and target selection stage, the system determines the current application's balancing target mode based on the distribution characteristics of various tags in historical data or preset instructions. Finally, in the parameter calculation stage, the system calls the corresponding algorithm to calculate the core compensation differential pressure according to the selected mode.

[0109] Specifically, for the bottom-end balancing mode, the compensation pressure difference The calculation was performed by filtering out all data pairs marked as discharge conditions. ), calculate the decrease in initial and final voltage differences during each discharge ( - Then take the median of these decreases as... .

[0110] For the top-level equilibrium mode, compensation pressure difference The calculation is performed as follows: All data pairs marked as charging conditions are selected, and the increase in the final voltage difference compared to the initial voltage difference is calculated for each charging cycle. - Then take the median of these increases as... .

[0111] For the central equilibrium model, the parameter characterizing the average inconsistency is... The voltage difference data was calculated by filtering out all voltage difference data marked as static or low-current steady-state conditions. Or, the stable voltage difference under this operating condition, take the median of the absolute values ​​of these voltage differences as... .

[0112] To evaluate the self-learning effect and drive continuous parameter optimization, the system implements an effect evaluation and parameter iterative optimization algorithm. After running a complete evaluation cycle with a set of optimized parameters, the system calculates a comprehensive equalization effect evaluation score. This score is calculated using a scoring function that comprehensively considers voltage consistency improvement, equalization energy loss, and equalization operation frequency. In a specific embodiment, the scoring function is defined as:

[0113]

[0114] in, and These represent the statistical pressure difference of the battery pack at the beginning and end of the evaluation period, such as the maximum pressure difference or the average pressure difference. This represents the total energy consumed by all equilibrium actions within the evaluation period, and it is calculated as follows: , For the first The average voltage of the cells being balanced during the secondary equalization process. To balance the current, To balance the duration; This represents the total number of balancing actions triggered during the evaluation period. These are preset weighting coefficients used to balance the contributions of different optimization objectives, such as... Version 1.0 is acceptable. A value of 0.1 to 0.5 is acceptable. A value of 0.01 to 0.05 is acceptable.

[0115] The system stores multiple sets of equilibrium control parameters and their corresponding evaluation scores. Iterative parameter optimization is performed periodically, for example, every 30 learning cycles. The rules include: retaining the top N sets of parameters with the highest historical evaluation scores as high-quality samples; and applying small-range random perturbations to the vicinity of these high-quality parameters, such as... (Perturbation of ±5mV) generates a new set of candidate parameters for trial to explore a better solution; when the optimal evaluation score no longer improves significantly in multiple consecutive evaluation cycles and the parameter fluctuation range is less than the preset threshold, the system determines that it has learned the approximate optimal parameter solution under the current operating condition, and can lock the set of parameters and apply it to subsequent control until the system detects a significant change in the operating condition.

[0116] Through the complete self-learning algorithm framework described above, including target pattern adaptation, statistical parameter calculation, and closed-loop iterative optimization based on effect feedback, this invention achieves comprehensive self-adaptation of the balanced control strategy, which can accurately match diverse application scenarios and the state changes of the battery pack throughout its entire life cycle.

[0117] In an exemplary embodiment, the present invention also provides a parameter self-learning system for cell balancing, applied to a battery management system, the parameter self-learning system for cell balancing comprising:

[0118] The data monitoring module 100 is used to monitor and acquire operating data, including the voltage of each cell and the charging and discharging current, during the historical operating cycle of the battery pack. The data includes at least the terminal voltage of each individual cell in the series battery pack and the charging and discharging current flowing through the battery pack, providing the raw data basis for subsequent analysis.

[0119] The feature analysis module 200 is used to analyze the operating data and identify the variation characteristics between the first voltage difference at the start time and the second voltage difference at the end time of a typical operating condition between the target cell and the reference cell. The feature analysis module 200 is communicatively connected to the data monitoring module 100 and configured to analyze and process the historical operating data. Its core function is to identify key features from the data that clearly reflect the dynamic evolution of inconsistencies between cells. This module identifies typical operating condition segments, calculates and compares the first voltage difference at the start time and the second voltage difference at the end time of the target cell and the reference cell under that operating condition, thereby extracting the variation characteristics representing the trend of inconsistency.

[0120] The parameter learning module 300 is used to dynamically determine the optimized value of at least one equalization control parameter based on the changing characteristics using a self-learning algorithm. This equalization control parameter is used to control the activation and deactivation of the passive equalization circuit. The parameter learning module 300 is communicatively connected to the feature analysis module 200 and is configured to receive the changing characteristics and execute the self-learning algorithm. The parameter learning module 300 dynamically calculates and determines the optimized value of at least one equalization control parameter based on the inconsistency evolution pattern revealed by the changing characteristics. The equalization control parameter includes, but is not limited to, a voltage difference threshold for controlling the activation and deactivation of the passive equalization circuit, and current and time thresholds for determining whether the battery pack has entered a state suitable for equalization and resting. This module enables the parameter values ​​to adapt to the actual state and aging process of the battery pack.

[0121] The equalization execution module 400 is used to perform cell equalization control on the battery pack using the optimized equalization control parameters. The equalization execution module 400 is communicatively connected to the parameter learning module 300 and is configured to receive and use the optimized equalization control parameters output by the parameter learning module 300 to perform real-time cell equalization control on the battery pack. Based on optimized threshold judgment conditions, the equalization execution module 400 controls the on / off state of corresponding switching devices in the passive equalization circuit, thereby initiating or stopping the equalization discharge operation on a specific target cell.

[0122] The aforementioned modules work collaboratively on the hardware and software platform of the battery management system, forming a complete closed loop from data acquisition to feature extraction, parameter learning, and finally strategy execution. Through continuous online learning and parameter updates, the system achieves the technical effect of adaptive optimization of the balancing strategy throughout the entire lifecycle, adapting to the product's own operating conditions.

[0123] In an exemplary embodiment, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a parameter self-learning method for cell equalization as described in any of the above embodiments. Optionally, the computer-readable storage medium may be a ROM (Read-Only Memory), RAM (Random Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic tape, floppy disk, and optical data storage device, etc.

[0124] In an exemplary embodiment, a computer program product is also provided, which includes computer program code stored in a computer-readable storage medium. A processor of a computer device reads the computer program code from the computer-readable storage medium and executes the computer program code, causing the computer device to perform the above-described parameter self-learning method for cell balancing.

[0125] Any aspects of this invention not described in detail are well-known to those skilled in the art.

[0126] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A parameter self-learning method for battery cell equalization, characterized in that, The method is applied to a battery management system, and the method includes: Throughout the battery pack's historical operating cycle, monitor and acquire operating data, including the voltage and charging / discharging current of each cell. Analyze the operational data to identify the variation characteristics between the first voltage difference at the start time and the second voltage difference at the end time of the typical operating condition between the target cell and the reference cell; Based on the aforementioned change characteristics, an optimized value for at least one equalization control parameter is dynamically determined using a self-learning algorithm, wherein the equalization control parameter is used to control the start and stop of the passive equalization circuit. The optimized equalization control parameters are used to perform cell equalization control on the battery pack.

2. The parameter self-learning method for cell equalization according to claim 1, characterized in that, The equalization control parameters include at least one of the following: equalization on-state voltage difference threshold, equalization off-state voltage difference threshold, resting current threshold for determining the resting state of the battery pack, and resting time threshold.

3. The parameter self-learning method for cell equalization according to claim 1, characterized in that, The step of dynamically determining the optimized value of at least one equilibrium control parameter based on the change characteristics using a self-learning algorithm specifically includes: Based on the aforementioned variation characteristics, calculate the compensation voltage difference used to compensate for the inconsistency between the target cell and the reference cell at the target state of charge point; Based on the compensation voltage difference, the optimized values ​​of the equalization start voltage difference threshold and the equalization stop voltage difference threshold are determined.

4. The parameter self-learning method for cell equalization according to claim 3, characterized in that, The step of calculating the compensation voltage difference used to compensate for the inconsistency between the target cell and the reference cell at the target state of charge point further includes: Based on the preset equalization target mode of the battery pack, select the corresponding calculation strategy; The equilibrium target mode includes a top equilibrium mode, a bottom equilibrium mode, or a middle equilibrium mode.

5. The parameter self-learning method for cell equalization according to claim 4, characterized in that: If the target equalization mode is the top equalization mode, the compensation voltage difference is calculated based on the statistical value of the difference between the second voltage difference and the first voltage difference recorded multiple times under the charging condition; If the target equalization mode is the bottom equalization mode, the compensation voltage difference is calculated based on the statistical value of the difference between the first voltage difference and the second voltage difference recorded multiple times under the discharge condition; If the target balancing mode is the central balancing mode, the compensation voltage difference is calculated based on the statistical value of the absolute value of the voltage difference recorded multiple times under static or low-current steady-state conditions.

6. The parameter self-learning method for cell equalization according to claim 1 or 2, characterized in that, The step of dynamically determining the optimized value of at least one equilibrium control parameter through a self-learning algorithm specifically includes: Statistical analysis was performed on the distribution of durations during which the system current was below multiple candidate thresholds within the historical operating cycle. Based on the duration distribution, candidate thresholds that satisfy preset frequency conditions and / or duration conditions are selected as optimized resting current thresholds.

7. The parameter self-learning method for cell equalization according to claim 6, characterized in that, After determining the optimized value of the resting current threshold, the optimized value of the resting time threshold is dynamically determined, specifically including: From a sample of multiple durations corresponding to the static current threshold, a typical value is selected as the optimized static time threshold according to a preset rule.

8. The parameter self-learning method for cell equalization according to claim 1, characterized in that, The method further includes: After running an evaluation cycle with the optimized equalization control parameters, the equalization effect evaluation score is calculated based on at least two of the following: voltage consistency improvement, equalization energy consumption, and number of equalization actions. Based on historical evaluation scores, multiple sets of equilibrium control parameters are screened, optimized iteratively, or locked.

9. A parameter self-learning system for battery cell equalization, characterized in that, The parameter self-learning system for cell balancing, applied to a battery management system, includes: The data monitoring module is used to monitor and acquire operating data, including the voltage and charging / discharging current of each cell, during the historical operating cycle of the battery pack. The feature analysis module is used to analyze the operating data and identify the variation characteristics between the first voltage difference at the start time and the second voltage difference at the end time of the target cell and the reference cell under typical operating conditions. The parameter learning module is used to dynamically determine the optimized value of at least one equalization control parameter based on the change characteristics through a self-learning algorithm, wherein the equalization control parameter is used to control the start and stop of the passive equalization circuit. The equalization execution module is used to perform cell equalization control on the battery pack using the optimized equalization control parameters.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the parameter self-learning method for cell balancing as described in any one of claims 1 to 8.

Images