BMS-based lithium ion battery equalization charging control method and system

Abstract

The present application relates to the field of battery charging control, and particularly relates to a lithium ion battery equalization charging control method and system based on BMS. The method comprises the following steps: continuously obtaining cell monitoring parameters of the battery charging state based on BMS, performing high-frequency filtering denoising processing and cell charging behavior analysis to obtain cell behavior characteristic time sequence; identifying the charging voltage response signal of each cell based on BMS, performing multi-frequency impedance change trend evolution, and performing cell charging response characteristic perception based on the cell behavior characteristic time sequence to construct multiple cell charging state atlases; performing charging energy flow affinity quantitative calculation according to the cell monitoring parameters, combining the multiple cell charging state atlases to distinguish state deviation trends, constructing a cell inter-topology connection graph, and mining weak cell and strong cell energy deviation complementary paths and dynamic deviation equalization planning. The present application realizes efficient and accurate energy equalization distribution and charging control.

Description

Technical Field

[0001] This invention relates to the field of battery charging control, and in particular to a lithium-ion battery equalization charging control method and system based on a battery management system (BMS). Background Technology

[0002] With the rapid development of new energy vehicles, energy storage systems, and portable electronic devices, lithium-ion batteries have become the mainstream electrochemical energy storage device due to their advantages such as high energy density, long cycle life, and low self-discharge rate. In various power systems and energy storage systems, Battery Management Systems (BMS) are widely used to ensure the safe, reliable, and efficient operation of batteries, enabling key functions such as real-time monitoring, state estimation, and charge / discharge control of lithium-ion batteries. Especially in large-capacity battery packs composed of multiple series-connected cells, differences in manufacturing processes and aging levels can easily lead to inconsistencies in voltage and charge levels, further exacerbating battery imbalance and potentially affecting the overall performance, safety, and lifespan of the entire battery pack. During actual operation, the voltage differences between individual cells in a lithium-ion battery pack accumulate, causing some cells to overcharge during charging or over-discharge during discharging, reducing the system's usable capacity and posing significant safety hazards. To curb this imbalance and improve the overall performance of the battery pack, battery balancing technology has become one of the core functions of a BMS system. Common balancing methods currently include passive balancing and active balancing. Passive balancing dissipates excess energy from high-voltage batteries as heat, resulting in a simple structure but low efficiency. Active balancing, on the other hand, redistributes energy among batteries through electrical transfer, which is more efficient, but also increases system complexity and cost.

[0003] However, existing lithium-ion battery equalization charging control methods still have many problems. On the one hand, traditional equalization strategies are mostly based on static threshold judgments, which are difficult to adapt to complex and ever-changing operating environments and battery state changes, resulting in unsatisfactory equalization effects and even potentially causing new battery safety risks. On the other hand, existing BMS systems typically have a delayed response to equalization control and lack the ability to comprehensively judge factors such as battery aging state and dynamic load changes, failing to achieve efficient and accurate energy distribution and charging control. Therefore, there is an urgent need to propose a more intelligent and dynamically responsive lithium-ion battery equalization charging control method and system that can fully leverage the real-time data acquisition capabilities of the BMS and achieve accurate and dynamic equalization control strategies through in-depth analysis and modeling of battery operating states. With the help of advanced algorithm design, state estimation models, and energy scheduling mechanisms, this method should effectively improve the consistency and utilization efficiency of the battery pack, extend the overall system's operating life, and significantly enhance the battery system's safety assurance capabilities under complex operating conditions. Based on this, constructing a lithium-ion battery equalization charging control system for an intelligent BMS platform has become a key direction for promoting the development of next-generation battery management technology. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a lithium-ion battery equalization charging control method and system based on a battery management system (BMS) to solve at least one of the aforementioned technical problems.

[0005] To achieve the above objectives, this invention provides a lithium-ion battery equalization charging control method based on a battery management system (BMS), comprising the following steps:

[0006] Step S1: Based on the continuous acquisition of cell monitoring parameters of battery charging status by BMS, perform high-frequency filtering and noise reduction processing and cell charging behavior analysis to obtain the time sequence of cell behavior characteristics;

[0007] Step S2: Based on the BMS, identify the charging voltage response signal of each cell, perform multi-frequency impedance change trend evolution, and based on the cell behavior characteristic time sequence, perceive the cell charging response characteristics and construct multiple cell charging state maps.

[0008] Step S3: Quantitatively calculate the charging energy flow affinity based on the cell monitoring parameters, and distinguish the state deviation trend by combining the charging state spectrum of multiple cells, thereby classifying weak cells and strong cells.

[0009] Step S4: Construct a topology connection diagram between battery cells, and mine and dynamically balance the energy offset complementary paths of weak and strong battery cells, and construct an intelligent offset path planning strategy.

[0010] Step S5: Perform asynchronous lag prediction drive adjustment based on the cell monitoring parameters, and perform real-time dynamic charging equalization control based on the intelligent offset path planning strategy to execute intelligent equalization charging control operation.

[0011] This specification provides a BMS-based lithium-ion battery equalization charging control system for executing the BMS-based lithium-ion battery equalization charging control method described above, including:

[0012] The charging behavior analysis module is used to continuously acquire cell monitoring parameters of battery charging status based on BMS, perform high-frequency filtering and noise reduction processing and cell charging behavior analysis to obtain the time sequence of cell behavior characteristics.

[0013] The state map module is used to identify the charging voltage response signal of each cell based on the BMS, perform multi-frequency impedance change trend evolution, and perceive the charging response characteristics of the cells based on the time sequence of cell behavior characteristics to construct multiple cell charging state maps.

[0014] The cell status classification module is used to perform quantitative calculation of charging energy flow affinity based on the cell monitoring parameters, and to distinguish the state deviation trend by combining the charging status maps of multiple cells, thereby classifying weak cells and strong cells.

[0015] The complementary path mining and planning module is used to construct the topology connection diagram between battery cells, and to mine and dynamically balance the energy offset of weak and strong battery cells to construct an intelligent offset path planning strategy.

[0016] The asynchronous equalization adjustment module is used to perform asynchronous lag prediction drive adjustment based on the cell monitoring parameters, and to perform real-time dynamic charging equalization control based on the intelligent offset path planning strategy, so as to execute intelligent equalization charging control operation.

[0017] The beneficial effects of this invention are as follows: By performing high-frequency filtering on cell monitoring parameters (such as voltage, current, and temperature), environmental noise and signal distortion during BMS acquisition are effectively removed, improving data stability. The dynamic behavior patterns of the cells during charging (such as voltage fluctuation trends and charging rate changes) are analyzed, establishing a time-series sequence of cell behavior characteristics, providing a high-resolution dynamic feature foundation for subsequent analysis. The early detection capability of abnormal cell behavior (such as overcharging and sudden changes in internal resistance) is enhanced, improving system safety and reliability. Multi-frequency impedance change analysis of the voltage-time curves in the charging response of each cell reveals differences in internal chemical processes and identifies potential problems such as aging and imbalance within the cell. By integrating time-series behavior characteristics, the response dynamics of each cell are comprehensively perceived, constructing a cell charging state map, achieving a spatial representation of the multi-dimensional performance state of the cell. This map can provide the input foundation for advanced algorithms such as graph neural networks for subsequent state assessment and path optimization, improving data mining accuracy. By introducing a quantitative calculation method for energy flow affinity, the "affinity relationship" between different battery cells during energy exchange and flow can be identified, thereby accurately classifying the charge-discharge compatibility between cells. Combining cell state maps with analysis of state deviation trends between cells, a systematic distinction is made between "weak cells" (degraded performance or low charging capacity) and "strong cells" (responsive and healthy performance). This provides a crucial foundation for energy scheduling, helping to implement a "weak-strong complementarity" strategy and avoiding uneven energy distribution or overloading of any particular cell. A topological connection graph between cells is constructed, systematically representing the connection structure and exchangeable paths between cells, improving the accuracy of overall system modeling. Graph analysis techniques are used to uncover energy compensation paths between weak and strong cells, enabling dynamic shifting of energy from excess cells to insufficient cells. By fitting the actual needs of the cells with path capacity, personalized energy shifting path planning strategies are developed, improving charging balancing efficiency and system adjustment flexibility. By identifying asynchronous and hysteretic effects in cell response (such as slow response and sluggish adjustment in some cells), preventative control is achieved by adjusting parameters in advance through a predictive driving mechanism. Based on the aforementioned intelligent offset path planning strategy, real-time charging equalization control is implemented, dynamically adapting to changes in cell state without relying on traditional static equalization circuits. This significantly improves charging efficiency, extends the overall battery pack lifespan, reduces energy loss, and enhances the system's energy efficiency ratio. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of the steps of a lithium-ion battery equalization charging control method based on BMS according to the present invention.

[0019] Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1.

[0020] Figure 3 This is a detailed flowchart illustrating the implementation steps of step S2;

[0021] Figure 4 This is a flowchart illustrating the detailed implementation steps of step S3. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0023] This application provides a lithium-ion battery equalization charging control method and system based on a BMS. The execution entities of the BMS-based lithium-ion battery equalization charging control method and system include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered as general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio / image management system, an information management system, and a cloud data management system.

[0024] Please see Figures 1 to 4 This invention provides a lithium-ion battery equalization charging control method based on a battery management system (BMS), comprising the following steps:

[0025] Step S1: Based on the continuous acquisition of cell monitoring parameters of battery charging status by BMS, perform high-frequency filtering and noise reduction processing and cell charging behavior analysis to obtain the time sequence of cell behavior characteristics;

[0026] Step S2: Based on the BMS, identify the charging voltage response signal of each cell, perform multi-frequency impedance change trend evolution, and based on the time sequence of cell behavior characteristics, perceive the cell charging response characteristics and construct multiple cell charging state maps.

[0027] Step S3: Quantitatively calculate the charging energy flow affinity based on the cell monitoring parameters, and distinguish the state deviation trend by combining the charging state spectrum of multiple cells, thereby classifying weak cells and strong cells.

[0028] Step S4: Construct a topology connection diagram between battery cells, and mine and dynamically balance the energy offset complementary paths of weak and strong battery cells, and construct an intelligent offset path planning strategy.

[0029] Step S5: Perform asynchronous lag prediction drive adjustment based on the cell monitoring parameters, and perform real-time dynamic charging equalization control based on the intelligent offset path planning strategy to execute intelligent equalization charging control operation.

[0030] In the embodiments of the present invention, see Figure 1 The diagram below illustrates the steps of a BMS-based lithium-ion battery equalization charging control method according to the present invention. In this example, the steps of the BMS-based lithium-ion battery equalization charging control method include:

[0031] Step S1: Based on the continuous acquisition of cell monitoring parameters of battery charging status by BMS, perform high-frequency filtering and noise reduction processing and cell charging behavior analysis to obtain the time sequence of cell behavior characteristics;

[0032] In this embodiment, continuous acquisition of battery cell charging status parameters is performed. The acquired data includes several key operating variables for each cell, such as instantaneous voltage, current, temperature, internal resistance, and SOC (State of Charge). To ensure data continuity and high accuracy, the BMS is set to a sampling frequency of once every 100ms, using a 32-bit ADC analog-to-digital converter module for analog sampling, with accuracy controlled within ±1mV (voltage) and ±0.1°C (temperature). The data is uploaded to the central processing unit via the CAN bus protocol and enters the filtering and feature analysis process. After data acquisition, the system first performs high-frequency filtering and noise reduction on the acquired battery cell monitoring data. Due to potential influences from current disturbances, temperature fluctuations, and electromagnetic interference during charging, the raw data exhibits certain high-frequency jitter and abnormal spikes. Therefore, the system employs a combination of sliding window averaging filtering (window size of 5 sampling points) and weighted median filtering to first smooth abrupt changes in data and then remove outliers with deviations greater than 3 times the standard deviation. After processing, the cell voltage curve is more stable, and the internal resistance change trend is more continuous, laying the foundation for subsequent behavior modeling. After filtering, the system enters the cell charging behavior analysis stage. The core of this part is to extract the behavioral characteristics of each cell within a specific time period by observing the interaction and change relationship between multiple monitoring parameters. In the constant current charging stage, by observing the change in voltage rise slope, SOC growth rate, and temperature rise amplitude, it is possible to identify whether the cell has a lag or abnormal heat dissipation. In the constant voltage stage, the system focuses on the voltage plateau maintenance time and internal resistance change trend to determine whether there is a charging saturation response deviation. In particular, the system introduces the "micro-perturbation voltage fluctuation analysis method", that is, under small current perturbation conditions (e.g., introducing a ±0.5A micro-perturbation signal on the basis of the original charging current), the voltage response sensitivity is monitored to evaluate the cell's electrochemical activity and interface polarization. Finally, the system constructs the above behavioral characteristic parameters into a "cell behavior characteristic time series" according to the time series. This series is recorded individually for each cell, including the SOC change rate, internal resistance micro-change amplitude, voltage slope, temperature gradient, etc. at its corresponding timestamp. Taking experimental data as an example, for a battery pack consisting of 96 cells, after a 1-hour constant current charging cycle, the system can record 57,600 sets of behavioral feature data points (sampled 10 times per second), and perform clustering and trend modeling on them, providing accurate data support for subsequent state recognition and balanced control.

[0033] Step S2: Based on the BMS, identify the charging voltage response signal of each cell, perform multi-frequency impedance change trend evolution, and based on the cell behavior characteristic time sequence, perceive the cell charging response characteristics and construct multiple cell charging state maps.

[0034] In this embodiment, the voltage response signal of each battery cell during the charging process is captured in real time using a Battery Management System (BMS). During this process, the BMS acquires the voltage response of each cell under a specific perturbation current excitation at a millisecond-level sampling frequency (e.g., 10ms / sample). The excitation current is injected intermittently into the charging circuit with a controllable amplitude (e.g., ±0.5A), and the duration is controlled between 200 and 500ms to avoid interfering with the actual charging process while stimulating the equivalent impedance characteristics inside the battery cell. The voltage response signal is captured by the BMS's high-precision ADC module (16-bit or higher) and immediately enters the feature extraction process after signal acquisition. Subsequently, the system processes the voltage response signal and enters the multi-frequency impedance change trend evolution analysis stage. The goal of this stage is to extract the voltage-current relationship of the battery cell at different frequency responses using a multi-scale time-frequency decomposition method, thereby deriving the equivalent impedance spectrum. Common methods include wavelet packet decomposition and short-time Fourier transform (STFT), which can decompose the original response signal into multiple frequency bands, such as 0.1Hz, 0.5Hz, 1Hz, and 5Hz, from which the complex impedance variation trend under different frequency bands can be derived. In the experimental environment, by superimposing perturbation current, the system can accurately observe the amplitude-frequency response of each cell under different frequency excitations. The analysis focuses on the evolution trend of impedance magnitude, the rate of change of phase angle, and the evolution path of equivalent series resistance and capacitance effect. Next, the system synchronizes and jointly analyzes the extracted impedance evolution features with the cell behavior feature time series established in the previous step S1. By constructing a one-to-one time stamp mapping model, the system fuses and analyzes "behavioral features" (such as SOC change rate, voltage slope, temperature gradient, etc.) with "impedance features" (such as impedance amplitude at a specific frequency, fitted pseudo-impedance trajectory, etc.) to identify the response characteristics of each cell in different time periods. If the impedance modulus of a battery cell suddenly increases at 0.5Hz, accompanied by a decrease in its state of charge (SOC) ramp-up rate, it may indicate a trend of lithium deposition, increased polarization, or intensified thermal diffusion resistance. Ultimately, the system visualizes these fused characteristics as time-series graphs, constructing multiple "cell charging state graphs." Each graph represents the multidimensional charging state evolution of a single cell, with the horizontal axis representing time and the vertical axis representing the combined results of various state parameters (such as frequency domain impedance, behavioral characteristic indicators, and local thermal response). Taking 96 cells as an example, the system generates 96 independent graphs, clearly presenting the consistency, deviations, and possible aging paths among the cells through heatmaps, trend lines, and threshold markers.

[0035] Step S3: Quantitatively calculate the charging energy flow affinity based on the cell monitoring parameters, and distinguish the state deviation trend by combining the charging state spectrum of multiple cells, thereby classifying weak cells and strong cells.

[0036] In this embodiment, the system performs quantitative calculations of charging energy flow affinity. The raw parameters used include real-time voltage (V_cell), internal resistance (R_cell), current (I_cell), SOC (State of Charge), and temperature for each cell during charging. To comprehensively reflect the cell's ability to accept charging energy flow, the system designs a comprehensive "energy affinity index," which integrates factors such as voltage rise rate per unit time, cell temperature rise, internal resistance change trend, and SOC growth rate. In the experiment, the unit charging cycle is 10 seconds, and the sampling period is 100ms. The system evaluates the energy absorption efficiency and reverse resistance of each cell within this cycle and generates a normalized affinity score, typically set between 0 and 1, with higher values ​​indicating better charging acceptance. Next, the system performs a corresponding analysis of the energy affinity results with the cell charging state map generated in the previous step. In the map, the changing trends of each cell in multiple dimensions (frequency response, SOC evolution, internal resistance trajectory, etc.) have been time-seriesified. By dynamically comparing the energy affinity graph with the energy affinity score, the system can effectively identify which cells exhibit consistently low affinity and negative evolution in their graph trends during the global charging process (such as continuously increasing internal resistance, expanding impedance modulus, and slow SOC increase). These cells are candidates for "weak cells." Conversely, if certain cells exhibit consistently high affinity, fast charging response, relatively stable internal resistance, and linear SOC growth across multiple dimensions, they can be classified as "strong cells." To enhance the reliability of the judgment, the system also introduces a state deviation trend analysis mechanism. This mechanism calculates the degree of deviation of each cell's state trajectory from the overall state center by performing clustering and comparative analysis on the historical state graphs of all cells. Cells with large deviations, especially those that continuously deviate from the average trajectory and show negative growth, will be marked by the system as weak cells with a high risk of performance degradation. In the measured data, if 5-8 out of 96 cells have an energy affinity below 0.35 and an average internal resistance above 10mΩ, while their temperature fluctuation exceeds 3℃, the system will automatically classify them as weak cells and treat them as priority equalization targets. Ultimately, the system will separately label weak and strong cells and record them in a unified cell state classification model, providing structured input for subsequent offset equalization control, energy path planning, and dynamic charging adjustment strategies. This classification result will be continuously updated through the BMS system, supporting online dynamic adjustment and achieving truly adaptive cell state identification and classification optimization.

[0037] Step S4: Construct a topology connection diagram between battery cells, and mine and dynamically balance the energy offset complementary paths of weak and strong battery cells, and construct an intelligent offset path planning strategy.

[0038] In this embodiment, the system first obtains the connection data within the battery modules based on the battery pack structure information stored in the BMS. This data typically originates from the battery pack's manufacturing configuration file or is identified using real-time detection methods (such as an embedded BMS diagnostic module with multi-channel sampling capabilities). Through a comprehensive scan of cell numbers, physical connection methods (series or parallel), and wiring topology sequence, the system constructs a physical topology connection diagram between the cells. In practical applications, for example, in a battery pack consisting of 96 18650 cells, the system uses 12 cells as a module unit. Each module uses series connections internally, while modules may use parallel or mixed series-parallel connections between each other. The BMS system uses each cell as a node in the graph and each connection relationship as an edge to construct the graph structure, thus forming a complete cell topology network. Next, the system marks the "weak cells" and "strong cells" identified in the previous step in this topology diagram and, based on factors such as physical path distance, reachability, thermal resistance correlation coefficient, and current path switching capability between cells, mines potential complementary energy offset paths. The system employs a weighted graph-based path search algorithm (such as the shortest path A* algorithm with heuristic factors or the weighted Dijkstra algorithm) for path mining. The weighting factors consider not only distance but also path switching costs, channel temperature rise coefficients, and bypass switch delays. In one experiment, the system limited the maximum energy transfer distance to no more than two module spans, and set the path weight threshold to 0.8 (after normalization), successfully identifying 12 pairs of weak and strong cell combinations with good energy complementarity potential. After identifying feasible energy complementarity paths, the system further performs "dynamic offset balancing planning." This planning is not statically executed but generates a dynamic scheduling sequence based on the dynamic changes in the SOC of each cell, real-time fluctuations in voltage differences, the current path switching status, and thermal distribution. The system maps the scheduling sequence to an adjustable current control matrix and defines an allowable energy transfer window (no more than 30 seconds per transfer, maximum transfer current no more than 1A, and priority given to paths with temperatures below 45°C) to ensure effective energy transfer from strong cells to weak cells, thereby achieving rapid SOC convergence. Ultimately, the system integrates all offset strategies and path information to form an "intelligent offset path planning strategy," which is then injected into the equalization control module of the BMS main control unit in real time. This strategy supports dynamic interruption, path reconstruction, and priority switching, and possesses adaptive environmental changes, fault avoidance, and high-temperature avoidance capabilities. In actual road condition simulation tests, the system's offset path planning strategy effectively shortened the average cell SOC convergence time by 18%, while reducing the local cell peak temperature rise by 3.2℃, significantly improving equalization charging efficiency and system safety.

[0039] Step S5: Perform asynchronous lag prediction drive adjustment based on the cell monitoring parameters, and perform real-time dynamic charging equalization control based on the intelligent offset path planning strategy to execute intelligent equalization charging control operation.

[0040] In this embodiment, the system first constructs the response timing trajectory of each cell during the charging process based on the cell monitoring parameters collected by the BMS, including charging voltage, charging current, temperature, SOC, internal resistance, and response delay data for each cell. Specifically, for the charging pulse response timing, the system analyzes the perturbation signals during the charging process (e.g., step current excitation within 5ms), and uses a sliding window method (100ms window width, 10ms step) to extract the delayed response interval of voltage changes, thereby estimating the response hysteresis parameters of each cell. These hysteresis parameters reflect the differences in response speed of different cells to the same current excitation and are an important basis for measuring asynchronous behavior. Subsequently, the system normalizes the extracted hysteresis parameters and uses this as a basis for predicting asynchronous behavior. Nonlinear regression modeling based on time series (such as LSTM regression or locally linearly embedded regression) is used to predict the cell response state in the next few tens of seconds. Taking actual test data as an example, in a battery pack of 96 18650 cells, 16 cells exhibited a response lag of over 50ms and showed significant SOC ramp-up lag during charging. The system, through the aforementioned modeling, predicted that these cells would exhibit further response delay risk within the next 20 seconds. After prediction, the system, combined with the "intelligent offset path planning strategy" generated in the previous step, dynamically schedules the charging adjustment mechanism. Specifically, the system adopts a "pre-load reduction" and "delayed charging strategy" for cells predicted to exhibit lag, while simultaneously increasing the energy flow input in stages for cells with strong response, and activating priority energy offset paths through the path selection module to achieve peak-shifting energy compensation. To ensure the accuracy and real-time performance of the adjustment process, the system integrates a high-resolution PWM current control unit at the hardware level, combined with a cell-side MOS control matrix, to dynamically adjust the current on / off state and amplitude according to the adjustment strategy. During the charging equalization process, each strategy execution cycle is controlled to be completed within 500ms, ensuring timely response to each predicted asynchronous state. Ultimately, the entire charging balancing process is no longer a static "charge until full" mode, but rather involves real-time adjustment and path reconstruction, intelligently distributing load and adjusting peak loads based on the dynamic state of each cell. Experimental data shows that applying this strategy under simulated road conditions can reduce the overall battery pack charging time by approximately 7.6%, while reducing the SOC difference between cells from a maximum of 8.5% to less than 1.2%, effectively achieving the goal of efficient, safe, and adaptive dynamic balancing control.

[0041] In this embodiment, see Figure 2The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0042] Based on BMS, millisecond-level micro-time sampling is performed to continuously acquire cell monitoring parameters of battery charging status;

[0043] The cell monitoring parameters are subjected to high-frequency filtering and noise reduction processing to obtain the filtered and denoised monitoring parameters;

[0044] The real-time SOC parameters, internal resistance, thermal parameters, charge / discharge efficiency, and voltage of each cell are calculated based on the filtering and noise reduction monitoring parameters to obtain the multi-dimensional state characteristics of the cell.

[0045] The multi-dimensional state characteristics of the battery cell are divided into time-series windows to obtain the multi-dimensional state characteristics of different time periods.

[0046] Based on the multi-dimensional state characteristics of different time periods, the charging behavior of the battery cell is analyzed, including changes in internal resistance perturbation, micro-voltage fluctuations, and transient thermal response, in order to obtain the time series sequence of battery cell behavior characteristics.

[0047] In this embodiment, to achieve high-precision control of the lithium-ion battery's state of charge, a high-frequency, millisecond-level micro-time-series data acquisition mechanism must first be implemented at the Battery Management System (BMS) level. The sampling system primarily collects data including cell voltage, current, temperature, and their rate of change. These parameters reflect the cell's current state of charge, temperature rise, and electrochemical stability. In the experimental design, a 1ms sampling period is used to continuously sample data from each cell, employing TI's BQ76PL455A-Q1 battery monitoring chip. This chip supports cascaded acquisition of up to 16 cells, with a sampling accuracy of ±1mV and a current sampling error within ±0.2%. The sampling system needs to be coupled with a real-time data transmission mechanism, implementing high-speed ADC conversion and CAN bus communication on an STM32 microcontroller platform to ensure stable data upload. A 1ms data acquisition period means that 1000 sets of cell data can be obtained per second, thereby capturing key transient responses such as rapid voltage drops, current pulsations, and transient temperature rises. During high-power charging (e.g., 2C rate) or rapid current switching phases (e.g., from 0.5C to 1.5C), this micro-timing sampling is crucial for capturing the dynamic response of the battery cell in the millisecond range, providing fundamental data support for subsequent behavior analysis and control.

[0048] Because high-speed data sampling is often accompanied by strong electromagnetic interference (EMI) and switching noise (such as pulsations from DC-DC converters and chargers), the raw sampled data inevitably contains high-frequency noise components. Therefore, after data acquisition, high-frequency filtering and noise suppression processing of the cell monitoring parameters are required. In the experiment, a multi-resolution analysis method based on wavelet transform was used for filtering. This method has good localization capabilities in both time and frequency dimensions and can effectively identify non-stationary noise occurring in a short time. Specifically, the db4 mother wavelet was selected for wavelet filtering, and the signal was reconstructed in a three-level decomposition to retain the main low-frequency characteristics of the cell response and remove high-frequency interference components above 50Hz. Compared with the traditional moving average method, the wavelet method can more accurately retain the edge information in the voltage change and current pulse process, avoiding the loss of details caused by excessive smoothing. In addition, in terms of hardware filtering, a second-order Butterworth low-pass filter was introduced into the sampling front-end circuit with a cutoff frequency of 500Hz to effectively reduce the interference of external coupling signals on the ADC input. Through hardware and software collaborative filtering, the peak-to-peak noise amplitude in the raw voltage data can be reduced from 30mV to approximately 5mV, significantly improving the availability and accuracy of cell state parameters and laying the foundation for subsequent state calculations. After obtaining high-quality monitoring data, key state indicators for each cell need to be calculated based on physical and data-driven models to constitute its multi-dimensional state characteristics. First, the State of Charge (SOC) is calculated using a model based on the Extended Kalman Filter (EKF). This method combines historical voltage and current data with the cell's open-circuit voltage (OCV) curve to dynamically update the SOC estimate. The initial SOC is estimated based on the charging history before power-on, and under 1C charging conditions, the EKF model error remains within ±2%. The internal resistance state calculation uses an improved version of the HPPC (Hybrid PulsePower Characterization) method. A 1C pulse current is injected at SOC levels of 60%, 40%, and 20% to analyze voltage dips and steady-state voltage differences, separating the ohmic internal resistance and polarization internal resistance. The internal resistance of each cell was measured to be 2–4 mΩ at room temperature (25°C), rising to approximately 6 mΩ at 10°C, reflecting thermal correlation. Thermal parameters (such as thermal resistance and thermal capacity) were modeled using an equivalent thermal network, derived from continuous temperature change data combined with heat flux calculations. The thermal resistance range is generally between 0.2 and 0.5°C / W, depending on the cell packaging and heat dissipation structure. Charge / discharge efficiency was calculated based on coulombic efficiency and energy efficiency, reaching over 98%. Under 1C standard conditions, some cells showed a significant efficiency decrease at high temperatures (45°C). Ultimately, the multi-dimensional characteristics of each cell include over eight dimensions: SOC, internal resistance, temperature, thermal conductivity parameters, voltage, current history, and charging efficiency. These are timestamped to form a cell state vector, providing a data foundation for time-series behavior modeling.After obtaining the multi-dimensional state parameters of the battery cell, it is necessary to structure them in the time dimension to capture the stage-by-stage behavior during the charging process. A sliding window approach is used for sequence division. Each window is set to a length of 5 seconds, with a sliding step size of 1 second, meaning overlapping windows exist. This window design maintains continuity while allowing for fine-grained segmentation of the battery cell's state change trends.

[0049] Within each window, statistics for each state characteristic within that time period are calculated, such as mean, maximum, minimum, standard deviation, and gradient rate of change. The rate of change of SOC reflects the stability of the charging current, while fluctuations in internal resistance indicate changes in material structure or local thermal disturbances. Experiments showed that voltage fluctuations were significant in the initial stage of CC (constant current) charging, with a standard deviation reaching 15mV, while gradually stabilizing towards the end of CV (constant voltage) charging, with fluctuations decreasing to around 3mV. Furthermore, to enhance the characterization of dynamic characteristics, Fourier transform was introduced to extract the distribution of each state characteristic across various frequency bands. The spectral energy distribution can help determine whether abnormal periodic oscillations exist, such as high-frequency interference or load instability. Finally, each window generates a set of time-series feature vectors, containing indicators such as state trends, fluctuation amplitudes, and frequency components, which are converted into a high-dimensional sequence data format, providing continuous input for subsequent cell behavior modeling and balancing strategy formulation. After obtaining the structured time window features, it is necessary to conduct in-depth analysis of cell behavior from multiple dimensions to uncover the underlying electrochemical dynamic characteristics. Multivariate time series analysis, combined with principal component analysis (PCA) and dynamic time warping (DTW), was used to identify microscopic behavioral changes such as internal resistance perturbations, voltage fluctuations, and transient thermal responses. Internal resistance perturbations were primarily identified by analyzing the minute upward trend of internal resistance within a short time window and its coupling relationship with temperature and current changes. If, while maintaining a constant state of charge (SOC) of 80%, the internal resistance slowly increased from 2.5 mΩ to 2.8 mΩ within 5 minutes, accompanied by a slight temperature increase (<1°C), it could be preliminarily determined that the cell had entered a region of enhanced thermal coupling. Micro-voltage fluctuations were analyzed by extracting energy changes in the 1–10 Hz range of the voltage signal using FFT transformation. A sudden increase in spectral energy within this range might reflect controller oscillations or electrochemical imbalances caused by material particle expansion during charging. Transient thermal response characteristics were extracted primarily based on a temperature-current hysteresis response model. By analyzing the temperature change delay time and maximum temperature rise rate before and after a sudden current change (e.g., 1C→2C), the thermal diffusion path and the effectiveness of the thermal management system could be estimated. Experimental data shows that a shorter response time indicates a stronger thermal coupling capability of the battery cell, which is beneficial for maintaining thermal stability. Using the above methods, a time-series sequence of battery cell behavior characteristics was finally constructed. This sequence not only includes numerical trends but also incorporates physically significant "behavioral segments," such as "impedance surge segment," "stable charging segment," and "thermosensitive transition segment." This time-series sequence will serve as the basis for subsequent equalization charging strategy decisions, guiding when to bypass a particular battery cell, adjust the charging current, or trigger a temperature control mechanism.

[0050] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0051] Based on BMS, a perturbation-type excitation signal is introduced into the charging current, and the charging voltage response signal of each cell is identified.

[0052] Multi-time-point response state fitting is performed on the charging voltage response signal to construct a steady-state voltage response waveform;

[0053] Multi-scale time-frequency decomposition of the steady-state voltage response waveform is performed to extract the voltage response impedance spectrum at different frequencies;

[0054] The voltage response impedance spectrum is subjected to multi-frequency impedance variation trend evolution to obtain impedance variation trend characteristics at different frequencies;

[0055] Based on the impedance change trend characteristics and the time sequence of cell behavior characteristics, the cell charging response characteristics are perceived, and multiple cell charging state maps are constructed.

[0056] In this embodiment, during the equalization charging control of lithium-ion batteries, traditional charging strategies typically employ a constant current-constant voltage (CC-CV) mode, without actively perturbing system parameters. However, to obtain the impedance characteristics and response sensitivity of the cell during charging, this step innovatively introduces a perturbation-type excitation signal into the charging current through a BMS control strategy, inducing a response from the cell and thus identifying its dynamic voltage change pattern. This perturbation signal generally uses a low-amplitude sine wave or step signal with an amplitude controlled within ±0.2C, and a frequency set between 0.1Hz and 5Hz, covering the low-frequency and mid-frequency bands of the electrochemical impedance spectroscopy. In the experiment, a 1Hz sine wave with an amplitude of ±200mA is superimposed on the standard 1C charging current, with the application time controlled within 5 seconds to ensure that the perturbation's impact on the overall SOC does not exceed 0.2%. By using the BMS embedded controller (such as the STM32F4 series) to sample the cell voltage at a frequency of 1ms during charging, the transient response changes of the voltage before and after the excitation signal injection can be clearly captured.

[0057] The voltage response signal is uploaded to the main control platform via the CAN bus. The system automatically identifies key characteristics such as instantaneous voltage fluctuations, response delays, and amplitude changes of each cell. By superimposing multiple perturbations and statistically averaging the response, the signal-to-noise ratio can be effectively improved, and random errors caused by ambient temperature and current fluctuations can be reduced. This technology provides a key experimental foundation for subsequent impedance modeling and state judgment. After acquiring the charging voltage response signal of the cell under perturbation current, multi-time-point response analysis is required to extract its stable response characteristics. To achieve this goal, this step uses time-domain curve fitting and signal steady-state response modeling methods to process the voltage response curve of each cell and reconstruct the steady-state voltage response waveform of the cell under perturbation. Specifically, the corresponding voltage sequence is acquired within each perturbation signal duration (e.g., 5 seconds) and divided into three time periods: rising edge, steady-state segment, and falling edge. In the experiment, the voltage change range is approximately ±8mV, and the typical response time is around 200ms. During the fitting process, the least squares method was used to fit the voltage response curve, and a combined model of a Gaussian function and a first-order exponentially decreasing function was selected to effectively simulate the inertial delay and steady-state stability of the voltage response. Furthermore, for the noise interference, a sliding window midpoint filter and a Savitzky-Golay smoothing filter were used in conjunction to preserve the key inflection points of the response waveform. Through repeated measurements and averaging, a stable and reproducible voltage response waveform curve was finally obtained for each cell, and the data was standardized based on frequency and amplitude normalization. This steady-state voltage waveform serves as the input for subsequent time-frequency analysis, laying the foundation for a deeper understanding of the cell's response mechanism to different disturbance frequencies. To reveal the cell's response capability to micro-perturbation current signals at different frequencies, the steady-state voltage waveform needs to be mapped from the time domain to the time-frequency domain to obtain the frequency response characteristics. The continuous wavelet transform (CWT) method was used to perform multi-scale time-frequency decomposition of the voltage response waveform, extracting the impedance characteristics of the cell at different frequency scales.

[0058] CWT uses the Morlet mother wavelet, which has good time-frequency local resolution capabilities. In the experiment, the scale was divided into 20 levels, with corresponding frequency distributions ranging from 0.1Hz to 5Hz. For each scale, the phase delay and amplitude ratio between the voltage signal and the input perturbation current signal were analyzed to construct a complex impedance spectrum (Z(f) = V(f) / I(f)). After synchronous sampling and Fourier transform processing of the voltage and current signals, amplitude-frequency response and phase-frequency response curves were obtained. Experimental data show that in the low-frequency region (<1Hz), the impedance is dominated by polarization internal resistance, exhibiting a large phase lag (approximately 60° or more); while in the high-frequency region (3~5Hz), the impedance amplitude tends to be stable, mainly affected by ohmic internal resistance, and has a small phase difference. By comparing and analyzing the impedance spectra of multiple cells at different frequency points, the frequency sensitivity characteristics, internal diffusion dynamics differences, and active material response delay characteristics of individual cells can be identified. Finally, a multi-frequency voltage response impedance spectrum containing 20 frequency points was established, providing detailed data support for subsequent multi-frequency trend analysis and state map construction. To further understand the dynamic change trend of cell impedance at different operating stages, time series modeling and trend evolution analysis of the multi-frequency impedance spectrum obtained in the previous step are required. Specifically, the multi-frequency impedance spectra collected at different time periods are stacked into a three-dimensional time-frequency-impedance matrix, and the trend of impedance change over time at each frequency point is modeled. A combination of weighted moving average (WMA) and local weighted regression (LOWESS) is used to extract the smooth trend of impedance at each frequency point. In the experimental setup, impedance spectrum sampling is performed every 10 seconds, with a total duration of 300 seconds, resulting in 30 time series segments for each frequency point. In the low-frequency region (0.2Hz), the impedance increases slowly during the SOC increase, reflecting an increase in polarization impedance; while in the high-frequency region (>3Hz), it remains basically stable, indicating that the ohmic internal resistance remains relatively unchanged.

[0059] Through quantitative analysis of the slope and amplitude of trend curves, the "Frequency Sensitivity Factor" (FSI) and the "Impedance Variation Index" (IZI) were defined to characterize the behavioral activity and health status of a battery cell at a specific frequency. Experimental results show that when the FSI exceeds 0.5 at 0.2Hz, the cell is more likely to have internal problems such as severe polarization and structural diffusion barriers. This trend characteristic can serve as a core indicator for cell anomaly warning, state screening, and equalization strategy triggering, effectively identifying cells whose response capability gradually deteriorates or becomes mismatched during charging. After obtaining the cell impedance change trend characteristics and the time series of cell behavior characteristics, the two types of information need to be fused to construct a cell state map reflecting the entire charging process, achieving a panoramic perception of the entire battery pack's state. This map uses the cell as a node, integrating its time-series behavioral characteristics (such as SOC change rate, internal resistance dynamics, temperature rise response, etc.) and frequency domain impedance characteristics (such as impedance curves at different frequencies, change trends, frequency response delay, etc.) to construct a multi-dimensional state space. In the specific implementation, an autoencoder is used to reduce the dimensionality of high-dimensional state features, compressing them into three-dimensional feature vectors (representing charging dynamics, impedance stability, and thermal response activity). The distribution of multiple cells in the state space is then visualized using t-SNE visualization technology. In the graph, close proximity indicates similar cell characteristics, while nodes with significant distances suggest state deviation or degradation trends. Furthermore, overlaying the time-dimensional evolution trajectory onto the state graph creates a "cell state trajectory diagram," visually displaying the state transition process of a cell at different charging stages. Experiments revealed that when some cells reach a SOC of over 80%, their state trajectories shift rapidly, indicating an imbalance trend that requires active balancing strategies for current bypass or compensation control.

[0060] In this embodiment, reference Figure 4 The above is a detailed implementation flowchart of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0061] Based on the cell monitoring parameters, the global maximum voltage of the battery, the real-time charging voltage of each cell, and the reverse internal resistance value are extracted.

[0062] Based on the battery's global highest voltage and the real-time charging voltage of each cell, the voltage difference of each cell is calculated, and the remaining voltage of each cell is extracted.

[0063] Based on the remaining voltage and reverse internal resistance value, the charging energy flow affinity is quantitatively calculated to generate the charging energy flow affinity of each cell.

[0064] Based on the charging energy flow affinity and the charging state spectrum of multiple cells, the average state of the cells is evaluated, and the state deviation trend is distinguished to classify weak cells and strong cells.

[0065] In this embodiment, key parameters, including voltage, current, temperature, and internal resistance, are extracted from the real-time cell monitoring data uploaded by the BMS. Every 100ms, the system acquires the current voltage value of each cell in the entire battery pack from the cell sampling module and automatically identifies the cell with the highest global voltage in each round of sampling. This parameter is used to calibrate the optimal reference point for the current charging state of the battery pack. Furthermore, the reverse internal resistance value, as an indicator of a cell's energy absorption potential, is typically obtained by inversely calculating the slope of the cell's voltage-current curve. A larger reverse value indicates that the cell can withstand a greater charging power per unit voltage. In a specific experiment, a charging test was conducted on a battery pack containing 96 18650 lithium-ion cells. The sampling period was set to 100ms. During a certain sampling period, the highest cell voltage was measured to be 4.187V, the lowest cell voltage to be 4.108V, and the reverse internal resistance value ranged from 0.35 to 0.62. The data from this stage provides a foundation for subsequent voltage difference calculations and energy affinity quantification. To quantify the "voltage space" of each cell in the current charging state, the system compares the real-time voltage of each cell with the global highest voltage during the current detection cycle and calculates the voltage difference. This difference is defined as "voltage margin," which reflects the cell's ability to continue accepting energy input under the current charging environment. The larger the voltage margin, the greater the cell's charging potential.

[0066] Under the aforementioned experimental conditions, if the highest voltage cell is cell 5 (4.187V) and cell 37 has a voltage of 4.135V, then its remaining voltage is 0.052V. This calculation is performed sequentially throughout the entire battery pack to form a set of cell voltage remaining vectors. This step is crucial for assessing the degree of charging imbalance between cells and predicting potential deviation behavior; it is an indispensable foundational data in the equalization control strategy. After obtaining the remaining voltage and internal resistance reverse value, the system performs a composite quantitative analysis to construct the cell's "charging energy flow affinity" index. This index, calculated by combining the remaining voltage with the cell's response to current changes (i.e., internal resistance reverse value), reflects the actual affinity of each cell in absorbing current and completing charge transfer under current conditions. To enhance recognition accuracy, the system uses a normalized vector superposition method to fuse the two sets of parameters, ensuring that numerical differences do not cause index deviation. Experiments revealed that cells with high remaining voltage but high internal resistance exhibit moderate affinity, while cells with low remaining voltage but extremely low internal resistance may also show high affinity. Therefore, this index can more comprehensively reflect actual charging absorption characteristics, providing a scientific basis for further cell state classification. Finally, the system integrates the cell charging energy affinity index with the previously constructed cell charging state map for average state assessment among cells. By calculating the average affinity of all cells in the current period as a group benchmark, the affinity of each cell is compared with this average to classify cells deviating from their state. If a cell's affinity is consistently more than 10% below the average for multiple consecutive periods and exhibits characteristics such as low remaining voltage and high internal resistance, it is labeled as a "weak cell." Conversely, if a cell's affinity is consistently more than 15% above the average and shows stable thermal response and positive SOC changes in the state map, it is defined as a "strong cell." The results of this classification will serve as the input point for the BMS equalization control strategy: prioritize suppressing the charging rate of strong cells and extend the charging time of weak cells, thereby achieving energy balance of the entire battery pack.

[0067] In this embodiment, step S4 includes the following steps:

[0068] Obtain a diagram of the battery's internal structure; identify cell connections from the battery's internal structure diagram and extract cell connection information;

[0069] Physical topology association mining is performed based on cell connection information to identify the topological connection relationships between cells;

[0070] The topological connection relationships between battery cells are fitted with the spatial distribution of topological connection paths to construct a topological connection map between battery cells.

[0071] We will mine energy complementary cells for both weak and strong cells and extract the energy complementary cell matrix.

[0072] Based on the topological connection graph between cells, energy offset complementary path mining is performed on the energy complementary cell matrix to extract the nearest energy offset priority complementary path.

[0073] Dynamic offset balancing planning is performed on the energy offset priority complementary path to construct an intelligent offset path planning strategy.

[0074] In this embodiment, the lithium-ion battery pack, as a multi-cell integrated system, directly affects the energy transfer path and equalization control strategy due to the connection method (series, parallel, or hybrid) between its internal cells. To achieve refined control based on the physical connection structure, it is first necessary to acquire and analyze the internal structure diagram of the battery. This step uses battery module structure drawings or point cloud data obtained from production data or 3D CT imaging, and extracts cell connection information through image recognition and structural analysis methods. In actual operation, a 3D structural scanner is used to acquire the internal structure of the battery module, and the spatial layout and tab connection method of each cell in the diagram are analyzed. Through edge detection algorithms and connected component analysis, the spatial coordinates of each cell are extracted, and their electrical connection status is determined by combining color identification and polarity features. In a 96S battery pack, the position and polarity of all 96 cells are identified using an automatic image segmentation method. After obtaining the cell connection information matrix, further topology mining and structural association identification are required for the physical connection relationships between cells. This step uses graph theory modeling to transform the battery cells and their connections into a topological graph, where nodes represent battery cells and edges represent electrical connections. Specifically, an undirected graph G=(V,E) is constructed, where V is the set of battery cells and E is the set of edges connecting the cells. Edge weights are assigned based on physical distance or resistance. In the experiment, Euclidean distance is used to calculate the relative positions of the battery cells in physical space, and the edge weights are then combined with measured connection resistance values ​​(ranging from approximately 0.5 to 2 mΩ) to reflect the actual conductivity and energy transfer cost. Subsequently, graph traversal and adjacency matrix analysis methods are used to identify key topological relationships, such as series path chains, parallel distribution clusters, and branch connection areas, and cluster analysis is used to divide the topological subgraph regions. In a PHEV battery pack, 12 sets of series units were identified, each containing 8 parallel battery cells connected via busbars. This topological relationship not only reveals the regularity of the physical connections between battery cells but also provides a graph structure basis for subsequent power scheduling and path planning. The mining results are standardized and stored as a topology vector map, forming an intermediate layer for cell network relationship modeling, realizing cross-layer data transformation from structure identification to control decision-making. After identifying the topology connections, the connection paths between cells need to be fitted and mapped in a spatial dimension to construct a three-dimensional cell topology connection map. This map can intuitively represent the physical proximity and conduction paths between cells, providing a basic spatial model for energy flow simulation and equilibrium path optimization. The fitting method uses multidimensional scaling (MDS) and force-directed graph drawing algorithms.MDS is used to project the high-dimensional cell connection relationship into three-dimensional space, preserving the relative distance between nodes; while the force-directed algorithm simulates the "spring-repulsion" behavior between nodes to optimize the graph layout and avoid node overlap. In the modeling, the connection strength (edge weight) is used as the attraction parameter, and the spatial distance is used as the repulsion parameter to achieve the natural distribution fitting of the graph. In a certain battery pack structure, after generating the topological graph, it is found that the cells are connected in a "hierarchical network", with parallel connections on the periphery and series connections in the middle, forming a typical fan-shaped conduction structure. This graph is standardized into a set of connection maps under a three-dimensional coordinate system, which can be visually operated and displayed in联动 with the real-time BMS status data.

[0075] The final output is a structured spatial topology graph of the cells, including information such as node coordinates, side lengths, and connection weights. This graph not only supports subsequent modeling of the energy flow path but can also be used in engineering applications such as thermal management simulation and electrical fault path prediction, which is the core data basis for implementing the spatial perception equalization control strategy. In the previous steps, the identification and classification of weak cells and strong cells have been completed, and a cell topology connection graph has been constructed. Based on this, the goal of this step is to discover cell pairs with potential energy complementarity possibilities in the topological structure and construct an "energy complementary cell matrix" as the input for the offset equalization path planning. The specific method is for each weak cell , and the topological path length (the number of shortest path hops or the minimum weight path) between all strong cells . Combining the affinity difference between the two (the matching degree between the energy overflow potential of the strong cell and the energy gap of the weak cell) to define the complementary index: ; A_j and A_i are the affinities of the cells respectively, D_ij is the topological path weight distance, is the adjustment coefficient, and ϵ is a small constant to prevent division by zero. Based on the value of C, cell pairs with strong complementarity are selected to construct the energy complementary matrix, C ∈ R^(m×n), where m is the number of weak cells and n is the number of strong cells. In the experiment, 5 weak cells and 8 strong cells were identified in a 32-cell system, and finally 40 effective complementary pairs were constructed, with an average path length of 2.5 hops and an average voltage difference of more than 40 mV, showing significant complementary adjustment potential. After completing the energy complementary matrix, this step focuses on extracting the "energy offset priority path" from the cell topology connection graph, that is, the shortest path required to transfer energy from strong cells to weak cells under actual physical connection conditions. The goal is to identify candidate lines with the lowest transmission loss and the smallest intervention path to achieve energy flow optimization at the path level. The path extraction is based on the graph traversal algorithm, using the Dijkstra shortest path method or the A* heuristic search. Starting from each strong cell node, the minimum edge weight path to the weak cell is found, and the weight considers factors such as physical distance, connection impedance, and control response delay for weighted combination. In the experiment, the edge weight is set as: Where R_ij is the resistance, L_ij is the path length, τ_ij is the control delay, and the coefficients are set to... Through algorithmic execution, an optimal energy offset path is generated for each pair of cells, and these paths are aggregated into a "priority complementary path set". In a certain structure, it was found that the path from cell #2 (strong) to cell #27 (weak) requires only two hops, transmitted via relay cell #15, with a total path impedance of only 1.2 mΩ, showing significant transmission potential. These paths will serve as the candidate set of control routes for the final equalization planning, ensuring that energy is optimally compensated in physical space with the shortest link. This matrix will be used for subsequent energy offset path analysis and is the fundamental data structure for planning the equalization energy transfer direction and path strength. After all priority complementary paths are determined, a dynamic offset control strategy needs to be formulated to automatically adjust the energy scheduling amount of each path according to the real-time system status, achieving intelligent and low-loss offset equalization scheduling. This step constructs the "offset path planning strategy", the core of which is to dynamically allocate energy transfer tasks by combining path load capacity, cell state evolution trend, and scheduling frequency constraints. The strategy adopts a multi-objective optimization method, and the objective functions include maximizing the energy compensation rate, minimizing the path load, and accelerating cell aging control. During the scheduling process, the BMS monitors parameters such as cell voltage, internal resistance, and temperature in real time. Each sampling period (e.g., 10 seconds) reassesses the schedulability of all current paths and dynamically activates the optimal sub-path. For cell pairs with the largest affinity difference and the smallest path weight, priority is given to bypass discharge (for strong cells) and weak charge (for very weak cells) control actions in the current cycle, with the scheduling ratio limited by the path's transmission capacity. In experiments, after the strategy was implemented, the very weak cells with an initial voltage difference of 50mV converged to ±5mV with the group's average level within 20 minutes, improving the balancing efficiency by nearly 40% compared to the traditional fixed balancing method. The planning strategy is highly adaptable, automatically adjusting the control strategy under different discharge rates, ambient temperatures, or cell aging conditions, making it a key element in achieving intelligent balancing control and improving battery pack consistency and lifespan.

[0076] In this embodiment, the specific steps of step S5 are as follows:

[0077] The charging pulse frequency is determined based on the cell monitoring parameters, and the pulse frequency timestamp is extracted.

[0078] Based on the pulse frequency timestamp, a sliding window regression analysis is performed on the steady-state voltage response waveform to extract the charging response hysteresis parameters for each cell.

[0079] The charging response hysteresis parameters are identified on a cell-by-cell basis to identify differentiated hysteresis characteristics, and peak-shifting response time-series analysis is performed to generate a global cell peak-shifting response sequence.

[0080] A asynchronous hysteresis prediction-driven adjustment is performed on the global cell peak shaving response sequence to construct an asynchronous charging equalization adjustment strategy.

[0081] Real-time dynamic charging equalization control is performed based on asynchronous charging equalization adjustment strategy and intelligent offset path planning strategy to execute intelligent equalization charging control operation.

[0082] In this embodiment, to more accurately analyze the response characteristics of each cell during charging, the system embeds slightly disturbed charging pulse signals during the standard constant current and constant voltage charging phase in actual BMS operation. These pulses are superimposed on the base charging current in the form of micro-amplitude steps, and voltage response is acquired using a high sampling rate. The system identifies the position of each valid pulse by detecting the instantaneous trigger points of the rate of change of current (ΔI) and the rate of change of voltage (ΔV), thereby accurately extracting the timestamp information of each pulse. In experimental testing, the system applied a low-amplitude (approximately 100mA) charging pulse with a frequency of 0.5Hz every 10 seconds, continuously monitoring 96 cells for 15 minutes, extracting 900 sets of pulse timestamps, which were then marked on the charging voltage response curve of each cell. These timestamps provide accurate reference points for the subsequent identification of voltage hysteresis response. After obtaining the pulse frequency timestamp, the system sets a sliding time window of fixed width (e.g., ±0.5 seconds) before and after each pulse time point, and performs local regression fitting on the voltage change curve within this time period to extract the delay time of the voltage response start and the lag time of the maximum response amplitude. This method uses linear or nonlinear regression (such as Loess or Savitzky-Golay regression) to fit the curve trend and identifies the start and peak delay of the actual response based on residual analysis.

[0083] For 96 battery cells, the system established a database of charging response hysteresis parameters. The average hysteresis time of each cell under different pulses ranged from 40ms to 110ms. The inconsistency in hysteresis time reveals subtle differences in charge transport, interface reaction, and internal impedance among different cells, and is an important indicator for measuring cell consistency and health status. After extracting the hysteresis parameters of each cell, the system began to perform lateral differential analysis. Specifically, the system clustered the hysteresis time series of all cells (e.g., based on DBSCAN or K-Means), grouping cells with similar response hysteresis characteristics into one category, thereby identifying groups of cells with higher hysteresis and slower response, as well as clusters of cells with faster response. Based on these classifications, the system sorted all cells according to their time response capability, generating a set of "off-peak response time series" from fastest to slowest. In the experimental model, hysteresis characteristics were divided into 5 levels, from level 1 (fastest response) to level 5 (slowest response). By rearranging the time sequence to avoid peaks, the system generates a global time scheduling table for cell charging responses. This ensures that the response rhythms of different cells can be logically staggered during subsequent charging control, reducing the risk of concentrated current fluctuations and improving the precision of balanced scheduling. Based on the staggered response time sequence, the system further designs an "asynchronous charging equalization adjustment strategy." This strategy aims to achieve compensation delay control for lagging cells and excitation control for fast-responding cells by nonlinearly adjusting the charging drive start-up time of different cells. This strategy mainly relies on a lag trend prediction model (such as LSTM or a moving regression predictor) to determine the lag change trend of the cells at the next moment and dynamically allocates the adjustment time window accordingly.

[0084] In actual testing, for cells with a response lag exceeding 100ms, the system delays their charging trigger window by 50ms to reduce impact. For cells with a response faster than 50ms, low-amplitude pre-charging is performed earlier. The adjustment strategy sends control commands to the BMS main control chip via serial port, adjusting the PWM duty cycle and current channel switching logic in real time to achieve asynchronous control of each cell, constructing a response balancing mechanism that can dynamically adapt to changes in cell state. The aforementioned asynchronous lag adjustment strategy is integrated with the previously established energy offset path planning strategy to form a complete real-time balancing control system. In this system architecture, the asynchronous strategy is responsible for adjusting the response time and rhythm of each cell, while the offset path strategy is responsible for scheduling the spatial transfer path of energy between cells, achieving coordinated control in both time and space dimensions. In testing, the system uses 1 second as the minimum time adjustment unit, calculating the current charging state, lag trend, and energy demand of each cell in real time, and automatically selecting to execute asynchronous delay, advance, or skip processing according to the adjustment strategy, while simultaneously controlling the direction and rate of energy offset. During continuous charging cycles, the control system effectively suppressed the SOC deviation rate between cells, reducing the maximum deviation from the initial 8.7% to 2.1%, and significantly reduced the thermal load difference between cells, thereby improving the overall system stability and safety.

[0085] In this embodiment, the specific steps for performing real-time dynamic charging balancing control based on asynchronous charging balancing adjustment strategy and intelligent offset path planning strategy to execute intelligent balancing charging control operation are as follows:

[0086] Real-time dynamic charging balance control is performed based on asynchronous charging balance adjustment strategy and intelligent offset path planning strategy, and the temperature detection value of each cell is identified based on the thermistor.

[0087] The cell temperature detection values ​​are analyzed for cell temperature differences, and a distribution fit is performed to obtain the cell temperature difference gradient.

[0088] Identify the cell internal resistance during the real-time dynamic charging balancing process;

[0089] Nonlinear correlation depth analysis was performed on the internal resistance of the battery cell and the temperature difference gradient of the battery cell to obtain the nonlinear correlation law between temperature and internal resistance;

[0090] Based on the temperature-internal resistance correlation law, local thermal excitation decision is made, and current bias path is driven to construct a dynamic local thermal excitation collaborative engine.

[0091] Intelligent equalization charging control is performed based on a dynamic local thermal excitation collaborative engine.

[0092] In this embodiment, real-time acquisition of the cell temperature status is a crucial link in the closed-loop control strategy while implementing asynchronous adjustment and intelligent path offset strategies. This step uses thermistors (NTCs) placed outside or near the heat source of each cell to detect temperature, forming a real-time temperature sampling channel for each cell, ensuring timely response of the control strategy to changes in thermal state. Specifically, at least one 10kΩ negative temperature coefficient thermistor is configured for each cell in the cell module, with a sampling frequency set to 1Hz to 10Hz, and parallel acquisition is performed according to the number of ADC channels of the BMS main control chip. The sampled temperature value is converted into the actual temperature value using the thermistor voltage divider calculation formula, with the error controlled within ±0.5°C. Temperature data is collected synchronously with charging current, voltage, SOC, internal resistance, etc., and summarized in real time in the BMS main control logic module. This temperature acquisition system provides a "thermal feedback" channel for equalization adjustment, especially when executing high-frequency offset paths, where some paths may cause local overheating risks. Temperature monitoring enables adjustment of the adjustment rhythm and current amplitude limitation. When the temperature rise rate of cell #12 exceeds 0.2°C / min, the BMS automatically delays its offset reception path scheduling priority. This step realizes a closed-loop equalization strategy under thermal sensing, laying a precise monitoring foundation for subsequent temperature difference analysis and thermal excitation regulation. After acquiring the temperature detection values ​​of all cells, the system first performs uniform standardization preprocessing on each set of sampled data, removing occasional data jitter and applying sliding median filtering (window size 3) to eliminate temperature measurement errors caused by environmental disturbances. Subsequently, the temperature difference between each cell and the average temperature within its battery module is calculated, constructing a 96-dimensional cell temperature difference vector. To comprehensively evaluate the spatial distribution structure of cell temperature, the system uses a combination of multinomial regression and kernel density estimation to fit the spatial distribution of cell temperature difference data, outputting a cell temperature difference gradient map. In the experiment, the graph revealed that some cells (such as those located near the heat dissipation end) consistently exceeded the average temperature by approximately 3-5°C, while other cells consistently remained below the average temperature by approximately 2°C or more, forming a clear high-low temperature gradient distribution, providing a basis for temperature-driven energy regulation. Simultaneously with temperature data processing, the system utilizes embedded pulse perturbation technology to sample and calculate the response ratio of the voltage difference before and after the application of a pulse current, achieving online identification of the cell's dynamic internal resistance. To improve accuracy, the system selects multiple perturbation windows (once every 10 seconds) within each charging cycle to collect voltage-current response values, using the difference moving average method to obtain the cell's instantaneous internal resistance parameters. During internal resistance identification, the system pays particular attention to the trend of internal resistance changes with temperature, SOC, and current. In the test model, the internal resistance difference between different cells ranged from 0.7mΩ to 2.1mΩ, with significantly higher internal resistance cells corresponding to weaker charging capabilities.Meanwhile, all internal resistance values ​​are dynamically bound to temperature fluctuations and are updated synchronously with the temperature difference gradient map in the data structure, providing accurate support for the subsequent establishment of a temperature-internal resistance relationship model. To gain a deeper understanding of the coupling relationship between the thermal characteristics of the battery cells and the efficiency of the electrochemical reaction, the system performs bivariate nonlinear modeling of the internal resistance of all battery cells and their corresponding temperature differences. Specifically, the system uses methods such as random forest regression and support vector regression (SVR) to identify the variation patterns of temperature and internal resistance under different SOC stages and charging rates. The model training data comes from measured results under five sets of environmental temperature gradients (10°C, 20°C, 30°C, 40°C, 50°C), with a sample size exceeding 20,000 sets.

[0093] Modeling results show that the battery cell exhibits the lowest internal resistance response at around 30°C. Below 20°C or above 40°C, the internal resistance increases significantly, indicating that the nonlinear effect of temperature on internal resistance presents a typical "U"-shaped structure. Extracting this correlation allows the system to determine the charging potential and load capacity of the battery cell based on its current temperature during charging regulation, thereby assisting in more precise current path scheduling and energy matching. Based on the identified temperature-internal resistance nonlinear correlation, the system further constructs a "dynamic local thermal excitation collaborative engine." Its core mechanism, while ensuring thermal safety, intelligently adjusts the current path and charging intensity to implement targeted excitation charging for battery cells with low temperatures and high internal resistance, rapidly raising their temperature to the optimal internal resistance range (approximately 30°C), thereby improving overall charging efficiency. Based on path planning strategies, the system appropriately reduces the current flux to high-heat-capacity (or high-temperature) battery cells, while increasing the current input to low-temperature battery cells, achieving "thermal energy guidance" in the spatial dimension. In testing, this strategy successfully raised the temperature of five groups of low-temperature cells by approximately 2°C, while simultaneously reducing their internal resistance by 0.6 mΩ, providing conditions for releasing more charging power and demonstrating excellent local thermal regulation capabilities. Ultimately, the system integrates the "Dynamic Local Thermal Excitation Collaborative Engine" into the overall BMS equalization control strategy, forming a three-dimensional intelligent equalization control model with spatial energy regulation, temporal asynchronous response, and thermal excitation guidance compensation. This model can output the optimal current bias path in real time based on the current SOC, temperature, and internal resistance of each cell, achieving coordinated execution of multiple objectives such as local thermal regulation, response timing shifting, and energy flow distribution. During long-cycle operation (3 hours of continuous charging), the system can control the maximum temperature deviation between cells from 6.3°C to 2.8°C, and the maximum internal resistance difference from 1.4 mΩ to 0.6 mΩ, while increasing the SOC convergence speed between cells by more than 30%, effectively extending the effective period of the system's equalization charging window. This provides a solid technical guarantee for the safe and efficient operation of large-scale lithium battery packs in new energy vehicles.

[0094] In this embodiment, a BMS-based lithium-ion battery equalization charging control system is provided to execute the BMS-based lithium-ion battery equalization charging control method described above, including:

[0095] The charging behavior analysis module is used to continuously acquire cell monitoring parameters of battery charging status based on BMS, perform high-frequency filtering and noise reduction processing and cell charging behavior analysis to obtain the time sequence of cell behavior characteristics.

[0096] The state map module is used to identify the charging voltage response signal of each cell based on the BMS, perform multi-frequency impedance change trend evolution, and perceive the charging response characteristics of the cells based on the time sequence of cell behavior characteristics to construct multiple cell charging state maps.

[0097] The cell status classification module is used to perform quantitative calculation of charging energy flow affinity based on the cell monitoring parameters, and to distinguish the state deviation trend by combining the charging status maps of multiple cells, thereby classifying weak cells and strong cells.

[0098] The complementary path mining and planning module is used to construct the topology connection diagram between battery cells, and to mine and dynamically balance the energy offset of weak and strong battery cells to construct an intelligent offset path planning strategy.

[0099] The asynchronous equalization adjustment module is used to perform asynchronous lag prediction drive adjustment based on the cell monitoring parameters, and to perform real-time dynamic charging equalization control based on the intelligent offset path planning strategy, so as to execute intelligent equalization charging control operation.

[0100] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, 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 the equivalents of the application are intended to be included within the invention.

[0101] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A BMS-based lithium-ion battery equalization charging control method, characterized in that, The method comprises the following steps: Step S1: Based on the BMS, continuously acquire the cell monitoring parameters of the battery charging state, perform high-frequency filtering denoising processing and cell charging behavior analysis to obtain the cell behavior characteristic time sequence; Step S2: Based on the BMS, identify the charging voltage response signal of each cell, perform multi-frequency impedance change trend evolution, and based on the cell behavior characteristic time sequence, perform cell charging response feature perception to construct multiple cell charging state atlases; Step S3: According to the cell monitoring parameters, perform charging energy flow affinity quantitative calculation, combine the multiple cell charging state atlases to distinguish state deviation trends, and thus divide out weak cells and strong cells; Step S4: Construct a cell interconnection topology graph, and perform weak cell and strong cell energy deviation complementary path mining and dynamic deviation balance planning to construct an intelligent deviation path planning strategy; Step S5: According to the cell monitoring parameters, perform non-synchronous lag prediction driving adjustment, and based on the intelligent deviation path planning strategy, perform real-time dynamic charging balance control to execute intelligent balance charging control work; The specific steps of step S5 are: According to the cell monitoring parameters, perform charging pulse frequency, and extract the pulse frequency timestamp; According to the pulse frequency timestamp, perform sliding window regression analysis on the steady-state voltage response waveform to extract the charging response lag parameter of each cell; Perform individual cell differential lag feature recognition on the charging response lag parameter, and perform staggered peak response time sequence analysis to generate a global cell staggered peak response sequence; Perform non-synchronous lag prediction driving adjustment on the global cell staggered peak response sequence to construct a non-synchronous charging balance adjustment strategy; Based on the non-synchronous charging balance adjustment strategy and the intelligent deviation path planning strategy, perform real-time dynamic charging balance control to execute intelligent balance charging control work; The specific steps of performing real-time dynamic charging balance control based on the non-synchronous charging balance adjustment strategy and the intelligent deviation path planning strategy to execute intelligent balance charging control work are: Based on the non-synchronous charging balance adjustment strategy and the intelligent deviation path planning strategy, perform real-time dynamic charging balance control, and based on the thermistor, identify the cell temperature detection value; Perform cell temperature difference analysis on the cell temperature detection value, and perform distribution fitting to obtain the cell temperature difference gradient; Identify the cell internal resistance of the real-time dynamic charging balance process; Perform non-linear correlation deep analysis on the cell internal resistance and the cell temperature difference gradient to obtain the temperature-internal resistance non-linear correlation law; According to the temperature-internal resistance correlation law, perform local thermal excitation decision-making to perform current bias path driving, and thus construct a dynamic local thermal excitation cooperative engine; Based on the dynamic local thermal excitation cooperative engine, perform intelligent balance charging control work.

2. The BMS-based Li-ion battery equalization charging control method according to claim 1, wherein, The specific steps of step S1 are: Based on the BMS, perform millisecond-level micro-time sequence sampling to continuously acquire the cell monitoring parameters of the battery charging state; Perform high-frequency filtering denoising processing on the cell monitoring parameters to obtain filtered and denoised monitoring parameters; According to the filter denoising monitoring parameter, the real-time SOC parameter, the internal resistance state, the thermal parameter, the charging and discharging efficiency and the voltage of each battery cell are calculated to obtain the multi-dimensional state characteristics of the battery cell; The multi-dimensional state characteristics of the battery cell are divided into time sequence windows to obtain the multi-dimensional state characteristics of different time periods; Based on the multi-dimensional state characteristics of different time periods, the charging behavior of the battery cell is analyzed, the internal resistance disturbance change, the micro-voltage fluctuation and the transient thermal response are analyzed to obtain the time sequence of the behavior characteristics of the battery cell.

3. The BMS-based Li-ion battery equalization charging control method according to claim 1, wherein, The specific steps of step S2 are: Based on the BMS, a perturbation excitation signal is introduced into the charging current, and the charging voltage response signal of each battery cell is identified; The multi-time point response trend of the charging voltage response signal is fitted to construct a steady-state voltage response waveform; The steady-state voltage response waveform is decomposed into multiple scales and frequencies to extract the voltage response impedance spectrum of different frequencies; The voltage response impedance spectrum is subjected to multi-frequency impedance change trend evolution to obtain the impedance change trend characteristics of different frequencies; Based on the impedance change trend characteristics and the time sequence of the behavior characteristics of the battery cell, the charging response characteristics of the battery cell are perceived to construct a plurality of battery cell charging state maps.

4. The BMS-based Li-ion battery equalization charging control method of claim 1, wherein, The specific steps of step S3 are: According to the battery cell monitoring parameters, the global maximum voltage of the battery, the real-time charging voltage of each battery cell and the internal resistance reverse value are extracted; Based on the global maximum voltage of the battery and the real-time charging voltage of each battery cell, the voltage difference of each battery cell is calculated to extract the voltage remaining amount of each battery cell; Based on the voltage remaining amount and the internal resistance reverse value, the charging energy flow affinity is quantitatively calculated to generate the charging energy flow affinity of each battery cell; Based on the charging energy flow affinity and the plurality of battery cell charging state maps, the average state of the battery cell is evaluated, and the state deviation trend is distinguished to divide the weak battery cell and the strong battery cell.

5. The BMS-based Li-ion battery equalization charging control method of claim 1, wherein, The specific steps of step S4 are: An internal structure diagram of the battery is obtained, the internal structure diagram of the battery is subjected to battery cell connection identification, and the battery cell connection information is extracted; Based on the battery cell connection information, the physical topology association is mined to identify the topological connection relationship between the battery cells; The topological connection relationship between the battery cells is subjected to topological connection path space distribution fitting to construct a topological connection diagram between the battery cells; The energy complementary battery cells are mined for the weak battery cell and the strong battery cell to extract an energy complementary battery cell matrix; Based on the topological connection diagram between the battery cells, the energy complementary battery cell matrix is subjected to energy offset complementary path mining to extract the nearest energy offset priority complementary path; The energy offset priority complementary path is subjected to dynamic offset balancing planning to construct an intelligent offset path planning strategy.

6. A BMS-based lithium-ion battery equalization charging control system, characterized in that, The BMS-based lithium ion battery equalization charging control method according to claim 1 comprises: A charging behavior analysis module is configured to continuously acquire the battery cell monitoring parameters of the battery charging state based on the BMS, perform high-frequency filter denoising processing and battery cell charging behavior analysis to obtain the time sequence of the behavior characteristics of the battery cell; A state map module is configured to identify the charging voltage response signal of each battery cell based on the BMS, perform multi-frequency impedance change trend evolution, perceive the charging response characteristics of the battery cell based on the time sequence of the behavior characteristics of the battery cell, and construct a plurality of battery cell charging state maps. The battery cell state trend division module is configured to perform charging energy flow affinity quantitative calculation according to the battery cell monitoring parameters, and combine the plurality of battery cell charging state maps to divide the state deviation trend, thereby dividing out weak battery cells and strong battery cells. The complementary path mining planning module is configured to construct a topological connection graph among the battery cells, and mine and dynamically deviate equalization planning for the energy deviation complementary paths of the weak battery cells and the strong battery cells, thereby constructing an intelligent deviation path planning strategy. The non-synchronous equalization adjustment module is configured to perform non-synchronous lag prediction driving adjustment according to the battery cell monitoring parameters, and perform real-time dynamic charging equalization control based on the intelligent deviation path planning strategy, so as to perform intelligent equalization charging control work.

Images