Intelligent charging and discharging protection method and system for energy storage battery pack

Abstract

The invention discloses an intelligent charging and discharging protection method and system for an energy storage battery pack, and relates to the technical field of energy storage system management, and the method comprises the steps: collecting the state data of a single battery in real time to construct a data set, carrying out the risk assessment through a single abnormal analysis model, and actively triggering the pre-protection; the battery pack analysis model generates initial state data, and the health state value of the battery pack is estimated and predicted in real time after comprehensive correction of the correction model; and after the risk is judged through comparison with a threshold value, the protection strategy generation model generates a charging and discharging protection strategy. A machine learning self-learning algorithm is adopted to realize monomer-module-system three-level progressive intelligent analysis, the capability of active prediction, pre-protection and strategy adaptive optimization is achieved, the safety, economy and reliability of an energy storage battery pack are remarkably improved, and an energy storage system is promoted to be transformed from passive protection to active risk management.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent protection for energy storage batteries, and in particular to an intelligent charging and discharging protection method and system for energy storage battery packs. Background Technology

[0002] In energy storage power stations and industrial and commercial microgrids, the safe and stable operation of battery packs is the core of ensuring energy reliability. Current mainstream technologies mainly rely on the passive protection mechanism of the Battery Management System (BMS), which triggers alarms or disconnections through preset fixed thresholds such as overvoltage, undervoltage, and overtemperature. This lacks the ability to proactively predict the health status of the battery pack, often responding only after a fault occurs, making it difficult to promptly prevent safety accidents such as thermal runaway. Existing technologies limit battery status assessment to the individual cell or module level, failing to effectively integrate individual cell risks, module synergies, and system-level operating conditions. This leads to a one-sided identification of systemic risks, rigid and inflexible protection strategies, and an inability to adapt to dynamic factors such as battery aging, ambient temperature, and load fluctuations, thus reducing the safety of energy storage battery packs. Therefore, improvements are needed. Summary of the Invention

[0003] To improve the safety of energy storage battery packs, this application provides an intelligent charging and discharging protection method and system for energy storage battery packs.

[0004] Firstly, the above-mentioned inventive objective of this application is achieved through the following technical solution: A method for intelligent charging and discharging protection of an energy storage battery pack, the method comprising the following steps: Real-time status data of individual cells is collected, and all collected real-time status data of individual cells are associated with a pre-set common time axis to construct an individual cell status dataset. A pre-set single-cell abnormality analysis model analyzes the real-time status data of the single cell to generate single-cell risk assessment data, and actively triggers pre-protection measures when a risk is predicted to exist in the single cell. A pre-set battery pack analysis model analyzes the individual cell state dataset to generate initial battery pack state analysis data. The pre-set battery pack correction model comprehensively analyzes the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. Based on the corrected battery pack status analysis data, the current health status value of the battery pack is estimated in real time, and the health status value of the battery pack is predicted according to the preset compensation rules to generate the predicted battery pack health value, which is then compared with the preset initial warning threshold to determine whether there is a risk to battery use. If there is a risk to battery use, a pre-set battery protection strategy generation model will analyze the battery pack to generate a battery charge and discharge protection strategy.

[0005] By adopting the above technical solutions, a three-tiered progressive intelligent protection system of "cell-module-system" is constructed. This system systematically addresses the core pain points of traditional energy storage BMS, such as passive alarms, one-sided risk assessment, and rigid and inefficient strategies. It proactively predicts and protects against risks in individual cells, advancing safety responses; it provides refined correction and assessment of group-level status to avoid underestimating systemic risks; and it achieves closed-loop protection from prediction to proactive intervention, generating tiered compensation strategies rather than simple shutdowns. The system reduces the incidence of battery usage risk events in energy storage power stations. At the same time, it extends the overall lifespan of battery clusters through proactive balancing and power limiting, reducing unplanned downtime losses. This promotes the transformation of energy storage systems from passive safety protection to an intelligent paradigm of proactive risk management, improving the safety of energy storage battery packs.

[0006] In a preferred example, this application can be further configured to include the following steps before the step of collecting real-time status data of individual cells and associating all collected real-time status data of individual cells with a pre-set common time axis to construct an individual cell status dataset: By synchronously acquiring the surface temperature field distribution, electrolyte pressure micro-vibration signal, and electrode potential transient waveform of individual battery cells through edge computing nodes, a multimodal state vector of a single battery cell is constructed. By utilizing a preset synchronous triggering mechanism, the multimodal state vector is aligned with the common time axis at the nanosecond level to eliminate sensor sampling delay differences; The aligned data stream is then subjected to sliding window filtering and outlier removal to generate a preprocessed single-cell state dataset.

[0007] In a preferred embodiment, this application can be further configured such that the step of analyzing the real-time status data of the single cell using a pre-set single cell anomaly analysis model includes the following steps: The first channel deploys a lightweight convolutional neural network to extract spectral features of the single-unit voltage ripple and identify high-frequency noise components in the early stage of a micro-short circuit. The second channel deploys a long short-term memory network to perform time-series modeling of the temperature transient response and predict the accelerated temperature rise trend before thermal runaway. The dual-channel output results are fused and input into a random forest classifier to generate a single-cell battery risk assessment level.

[0008] In a preferred embodiment, this application can be further configured as follows: in the step of comprehensively analyzing the individual cell risk assessment data and the initial battery pack state analysis data in a pre-set battery pack correction model to generate corrected battery pack state analysis data, a three-level progressive correction strategy is executed, wherein the execution of the three-level progressive correction strategy includes: The first stage uses individual cell risk assessment data to weight and correct the estimated state of charge in the initial battery pack state analysis data, reducing the interference of high-risk cells on the overall estimate. The second stage, based on the corrected state of charge value, recalculates the power allocation coefficient of each module in the battery pack and generates a power redistribution command. The third stage couples the power redistribution command with the temperature distribution field in the initial battery pack state analysis data, corrects the thermal equalization control parameters, and forms the final corrected battery pack state analysis data.

[0009] In a preferred embodiment, this application can be further configured as follows: The step of estimating the current health status value of the battery pack in real time based on corrected battery pack status analysis data, predicting the health status value of the battery pack according to a preset compensation rule to generate a predicted battery pack health value, and comparing it with a preset initial warning threshold to determine whether there is a risk to battery use, includes the following steps: Based on the degradation trajectory of the same model of battery pack in the historical database, a nonlinear regression model of health status value and cycle number is established. Based on the current number of battery cycles used, the initial warning threshold is dynamically lowered so that the warning is triggered when the aging battery pack is at a lower health status value. The adjusted dynamic threshold is compared with the predicted battery pack health value in real time. If the health value is lower than the dynamic threshold, it is determined that there is a risk to battery use.

[0010] In a preferred embodiment, this application can be further configured as follows: in the step of analyzing the battery pack using a pre-set battery protection strategy generation model to generate a battery charge / discharge protection strategy, a multi-objective optimization framework is constructed, the multi-objective optimization framework including: Define sub-indicators for the objective function and assign adjustable weight coefficients to them; A reinforcement learning algorithm is used to simulate the strategy in a digital twin battery pack, traversing combinations of current limit values, voltage cutoff values, and temperature derating curves. The strategy that maximizes the comprehensive objective function is selected as the optimal battery charge and discharge protection strategy, and is then sent to the battery management system for execution via the northbound interface.

[0011] In a preferred example, this application can be further configured to include the following steps after analyzing the battery pack using a pre-set battery protection strategy generation model to generate a battery charge / discharge protection strategy: Acquire the current working scenario data of the battery pack, which includes the time-of-use electricity price period identifier in the home energy storage scenario, the dynamic curve of the production line load in the industrial energy storage scenario, and the power dispatch instruction sequence in the power grid frequency regulation scenario. Calculate the residual between the predicted battery pack health value and the subsequently collected actual health status value, mark the residual as the prediction deviation, and store the prediction deviation in the scenario-deviation association database according to the working scenario. Attribution analysis is performed on the scenario-deviation association database to identify key influencing factors that lead to prediction deviations in different scenarios and generate scenario correction coefficients. The scenario correction coefficients are used to perform weighted corrections on the health state prediction branch in the battery protection strategy generation model, generating a corrected protection strategy suitable for a specific scenario, thereby improving prediction accuracy and making the strategy more in line with actual working conditions.

[0012] Secondly, the above-mentioned inventive objective of this application is achieved through the following technical solutions: A smart charging and discharging protection device for energy storage battery packs, the device comprising: a single-cell battery status dataset construction unit, used to collect real-time status data of single cells in real time, and associate all collected real-time status data of single cells with a preset common time axis to construct a single-cell battery status dataset; The single-cell battery risk assessment data generation unit is used to pre-set a single-cell battery anomaly analysis model to analyze the real-time status data of the single-cell battery, generate single-cell battery risk assessment data, and actively trigger pre-protection measures when a risk is predicted to exist in a single-cell battery. The initial battery pack state analysis data generation unit is used to pre-set the battery pack analysis model to analyze the individual battery state dataset based on a machine learning algorithm in order to generate initial battery pack state analysis data. A battery pack state analysis data generation unit is used to pre-set a battery pack correction model to comprehensively analyze the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. The battery usage risk assessment unit is used to estimate the current health status value of the battery pack in real time based on the corrected battery pack status analysis data, and predict the health status value of the battery pack according to the preset compensation rules to generate a predicted battery pack health value, and compare it with the preset initial warning threshold to determine whether there is a battery usage risk. The battery charge / discharge protection strategy generation unit is used to analyze the battery pack using a pre-set battery protection strategy generation model when there is a risk to battery use, in order to generate a battery charge / discharge protection strategy.

[0013] Thirdly, the above-mentioned objectives of this application are achieved through the following technical solutions: An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described intelligent charging and discharging protection method for an energy storage battery pack.

[0014] Fourthly, the above-mentioned objectives of this application are achieved through the following technical solutions: A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described intelligent charging and discharging protection method for an energy storage battery pack. Attached Figure Description

[0015] Figure 1 This is a flowchart of an intelligent charging and discharging protection method for an energy storage battery pack according to one embodiment of this application; Figure 2 This is a schematic block diagram of an intelligent charging and discharging protection device for an energy storage battery pack according to one embodiment of this application; Figure 3 This is a schematic diagram of an electronic device according to an embodiment of this application.

[0016] Icon labels: 1. Single cell state dataset construction unit; 2. Single cell risk assessment data generation unit; 3. Initial battery pack state analysis data generation unit; 4. Corrected battery pack state analysis data generation unit; 5. Battery usage risk judgment unit; 6. Battery charge and discharge protection strategy generation unit. Detailed Implementation

[0017] The present application will be further described in detail below with reference to the accompanying drawings.

[0018] In one embodiment, such as Figure 1 As shown, this application discloses a smart charging and discharging protection method for energy storage battery packs, which specifically includes the following steps: S10: Collect real-time status data of individual cells in real time, and associate all collected real-time status data of individual cells with a pre-set common time axis to construct a single cell status dataset. Specifically, taking a 10MWh commercial and industrial energy storage power station as an example in this application embodiment, the system collects real-time status data such as voltage, current, surface temperature, and internal resistance of each individual lithium iron phosphate battery through a battery management system (BMS). Using a GPS clock as a common time axis reference, the voltage drop signal of battery cell number 128 (instantly dropping from 3.2V to 2.8V and recovering) is precisely aligned with the corresponding charging and discharging current pulse (jumping from 50A to 150A) to construct a single-cell battery status dataset. This dataset accurately records the timestamps of abnormal events and the correlation of multiple parameters, providing a high-temporal-resolution data foundation for subsequent anomaly analysis and avoiding misjudgments caused by asynchronous data acquisition.

[0019] S20: The pre-set single-cell abnormality analysis model analyzes the real-time status data of the single cell to generate single-cell risk assessment data, and actively triggers pre-protection measures when a risk is predicted to exist in the single cell. Specifically, the time-series state data of a specific battery is input into a single-cell anomaly analysis model. This model, based on a Long Short-Term Memory (LSTM) network, learns from 100,000 historical normal charge-discharge curves and identifies a pattern of voltage drop accompanied by a slight temperature rise (from 25°C to 27°C), which matches the micro-short circuit feature library with an 85% match rate. The model generates risk assessment data as "High Risk - Micro-Short Circuit Warning" and proactively triggers pre-protection measures: immediately reducing the upper limit of the charge-discharge current of the module containing the single battery from 100A to 30A, and simultaneously pushing a warning message to the operation and maintenance platform. Compared to the traditional BMS's "overvoltage / undervoltage cut-off" mode, this proactive prediction and protection mechanism provides a 30-minute advance warning of battery anomalies, preventing short-circuit deterioration from leading to thermal runaway and significantly improving the safety of the energy storage system.

[0020] S30: The pre-set battery pack analysis model analyzes the individual battery state dataset to generate initial battery pack state analysis data; Specifically, the battery pack analysis model performs cluster analysis on the state dataset of the entire battery cluster (16 modules and 192 cells), including the 128th anomalous cell. The model identifies three other cells in the cluster whose internal resistance growth trends are similar to those of cell 128, but have not yet reached the warning threshold, generating initial battery pack state analysis data: the overall health of the cluster shows a downward trend, but it can still maintain basic operation. This analysis provides a macroscopic view of the battery pack's state, avoiding focusing only on a single anomaly while ignoring group risks, and providing a global judgment basis for subsequent refined corrections.

[0021] S40: The pre-set battery pack correction model performs a comprehensive analysis of the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. Specifically, the battery pack correction model comprehensively analyzes the individual cell risk assessment of S20 (high-risk micro-short circuit No. 128) and the preliminary group status of S30 (3 cells with increasing internal resistance), correcting and generating more accurate battery pack status analysis data: determining that the battery cluster has an "early micro-short circuit propagation risk," downgrading the health status assessment from the initial "good" to "needs attention," and upgrading the risk level from "module level" to "cluster level." This comprehensive analysis and correction avoids the one-sided judgment of underestimating systemic risk based solely on individual cell risk or overestimating overall health based solely on average status, making the group-level status assessment closer to the true safety level and providing an accurate baseline for subsequent health prediction.

[0022] S50: Based on the corrected battery pack status analysis data, estimate the current health status value of the battery pack in real time, and predict the health status value of the battery pack according to the preset compensation rules to generate the predicted battery pack health value, and compare it with the preset initial warning threshold to determine whether there is a risk to battery use. Specifically, based on the corrected state analysis of S40, the system estimates the current state of health (SOH) of the battery cluster in real time. According to preset compensation rules, combined with the deterioration rate of the three cells with increasing internal resistance, it predicts that the health status will accelerate its decline to below the warning threshold within the next 30 days. After comparing with the initial warning threshold, the system determines that there is a "battery usage risk" and triggers a yellow warning. Compared with traditional threshold alarms based solely on the current SOH, this predictive judgment reveals the health degradation trend in advance, allowing sufficient response time for operation and maintenance decisions and avoiding power outages in the energy storage system due to sudden failures.

[0023] It should be noted that, in this application, "pre-set compensation rules" refers to a set of multi-dimensional dynamic adjustment rules built into the system to correct health status prediction biases, based on the battery's physical degradation mechanism and operating condition characteristics. This set of rules is not a static threshold, but rather consists of a configurable logical judgment matrix, specifically including the following core compensation rules: 1. Nonlinear compensation rules for aging trajectory: This rule establishes a nonlinear regression model between the state of health (SOH) value and the number of battery cycles based on the degradation trajectory of battery packs of the same model in a historical database. When the current number of cycles of the battery pack exceeds the inflection point threshold (e.g., 8000 cycles), the quadratic compensation coefficient is automatically activated, transforming the SOH prediction formula from linear SOH=100-a×N to nonlinear SOH=100-a×N - b×N², accurately reflecting the aging acceleration effect and avoiding optimistic bias in mid-to-late stage predictions. 'a' is the basic degradation coefficient, representing the slope of the linear degradation stage, and 'b' is the nonlinear acceleration coefficient, reflecting the degree of deterioration in the degradation rate in the later stages of aging, set based on practical experience.

[0024] 2. Dynamic compensation rules for working conditions: This rule dynamically generates a working condition intensity coefficient based on real-time collected data on charging / discharging rates, deep discharge frequency, and the proportion of idle time. When continuous high-rate discharge (e.g., >1C) is detected for more than a set duration, an accelerated attenuation compensation factor is triggered, which adjusts the predicted health status value downward according to the degree of wear and tear, ensuring that the prediction results closely match the actual usage intensity.

[0025] 3. Temperature field non-uniformity compensation rules: This rule is based on the temperature distribution field data of each module within the battery pack, calculating the temperature standard deviation and maximum temperature difference characteristics. When the temperature difference between modules exceeds a preset threshold, it is determined that there is a risk of local overheating, and the thermal imbalance compensation item is activated. This penalizes the estimated health status of the battery in the high-temperature area by lowering it, and increases the weight of the low-temperature area, so that the overall health prediction reflects the impact of thermal environment degradation.

[0026] 4. Individual consistency drift compensation rules: This rule continuously monitors the differences in internal resistance and voltage dispersion among individual cells within the battery pack. When inconsistency indicators (such as range and variance) exceed the dynamic threshold, a consistency compensation mechanism is triggered, which corrects the predicted health status of the pack according to the "weakest link effect" principle. That is, the pack health value is dominated by the health value of the worst individual cell, avoiding the "average estimation" from masking the risk of individual cell deterioration.

[0027] In other embodiments, corresponding compensation rules can be set according to requirements.

[0028] S60: If there is a risk to battery use, the pre-set battery protection strategy generation model analyzes the battery pack to generate a battery charge and discharge protection strategy. Specifically, the battery protection strategy generation model automatically generates a three-level protection strategy targeting the primary cause of "micro-short circuit propagation risk": The first level limits the charging and discharging power of the problematic battery cluster to 40% of its rated value to alleviate individual cell voltage stress; the second level initiates active balancing to supplement the charging of three cells with increasing internal resistance to delay deterioration; the third level generates maintenance recommendations, suggesting offline testing within 15 days. The strategy is translated into control commands and issued to the power conversion system (PCS) and battery management system (BMS) for execution, actively reducing the cluster power from 500kW to 200kW, while the balancing system initiates a 1A small current compensation for the abnormal cells. This protection strategy represents a leap from a single alarm to proactive intervention, containing systemic risks in their initial stages.

[0029] In summary, the construction of a three-tiered, progressive intelligent protection system—comprising individual cells, modules, and the entire system—systematically addresses the core pain points of traditional energy storage BMS, such as passive alarms, incomplete risk assessments, and rigid, inefficient strategies. It proactively predicts and protects against risks in individual cells, advancing safety responses; it provides refined correction and assessment of group-level status to avoid underestimating systemic risks; and it achieves closed-loop protection from prediction to proactive intervention, generating tiered compensation strategies rather than simple shutdowns. This system reduces the incidence of battery usage risk events in energy storage power stations, improves the safety of energy storage battery packs, and extends the overall lifespan of battery clusters through proactive balancing and power limiting, reducing unplanned downtime losses and driving the intelligent transformation of energy storage systems from passive safety protection to proactive risk management.

[0030] Before S10: Real-time acquisition of individual battery status data and association of all acquired individual battery status data with a pre-set common time axis to construct an individual battery status dataset, the following steps are included: S01: Simultaneously collect the surface temperature field distribution, electrolyte pressure micro-vibration signal and electrode potential transient waveform of the battery cell through edge computing nodes to construct the multi-mode state vector of the battery cell. Specifically, in a 10MWh lithium iron phosphate energy storage power station, edge computing nodes are deployed inside each battery module to synchronously collect multi-source data from a single cell: the surface temperature field distribution of the battery is acquired through an infrared thermal imaging array (identifying that the temperature at the top of the cell is 3°C higher than at the bottom); micro-vibration signals of electrolytic hydraulic pressure are captured through fiber optic sensing probes (detecting abnormal pressure pulse frequency); and transient waveforms of electrode potential are acquired through a high-frequency potential sensor (recording voltage spikes and glitches). These three types of heterogeneous data are fused to construct a multimodal state vector, breaking through the traditional reliance on a single dimension of voltage, current, and temperature. This enables subsequent anomaly analysis to possess a three-dimensional perception capability encompassing "temperature field - mechanical vibration - potential transients," significantly improving the sensitivity to capturing early characteristics of micro-short circuits.

[0031] S02: Using a preset synchronous triggering mechanism, the multimodal state vector is aligned with the common time axis at the nanosecond level to eliminate sensor sampling delay differences; Specifically, the aforementioned multimodal state vectors were acquired by three different sensors, with original timestamps exhibiting microsecond-level deviations. Edge nodes utilized a pre-defined GPS synchronization trigger mechanism, using the PPS pulse as a reference alignment signal to unify temperature field data, pressure vibration signals, and potential waveforms onto a common time axis, eliminating sensor sampling delay differences. After alignment, the system accurately identified voltage spikes and pressure pulses occurring synchronously in time (with an interval of less than 5μs), determining it as a mechanical-electrical coupling phenomenon caused by micro-short-circuit discharge. This alignment accuracy ensures the accuracy of causal relationship determination, avoiding misjudgments caused by traditional time asynchrony, and laying a data foundation for subsequent accurate attribution.

[0032] S03: Perform sliding window filtering and abnormal pulse removal on the aligned data stream to generate a preprocessed single-cell state dataset; Specifically, the aligned data stream contains noise pulses caused by electromagnetic interference. The system employs a sliding window filtering algorithm (covering 10 sampling points) to smooth the multimodal data, removing high-frequency noise that deviates from the historical average by more than a threshold; it also identifies and removes transient abnormal pulses caused by equipment start-up and shutdown (such as voltage surges during PCS startup). The signal-to-noise ratio of the preprocessed state dataset for cell number 128 generated after filtering is significantly improved, and abnormal pulse interference is reduced by 90%, making the subsequent cell battery state dataset constructed by S10 highly reliable, avoiding false alarms triggered by noise, and ensuring the stability and accuracy of the anomaly detection system.

[0033] This application systematically improves the depth, accuracy, and timeliness of battery state perception through a multimodal data acquisition, nanosecond-level alignment, and filtering preprocessing mechanism constructed in S01-S03. The multimodal vector in S01 breaks through the limitations of traditional three-parameter methods, capturing the early mechanical-electrical coupling characteristics of micro-short circuits; the alignment in S02 eliminates time delay errors and accurately determines causal relationships; and the filtering in S03 removes abnormal pulses, ensuring data reliability. These three mechanisms provide a high-quality data foundation for subsequent intelligent analysis in S10-S60, improving the energy storage power station's ability to identify early battery pack faults from the minute level to the second level, significantly reducing the false alarm rate, and substantially enhancing overall safety and availability.

[0034] The step of analyzing the real-time status data of the single cell using the pre-set single cell anomaly analysis model includes the following steps: The first channel deploys a lightweight convolutional neural network (CNN) to extract spectral features of the single-unit voltage ripple and identify high-frequency noise components in the early stage of a micro-short circuit. The second channel deploys a long short-term memory network to perform time-series modeling of the temperature transient response and predict the accelerated temperature rise trend before thermal runaway. The dual-channel output results are fused and input into a random forest classifier to generate a single-cell battery risk assessment level. In this embodiment, the level is divided into three levels: normal, warning, and alarm.

[0035] Specifically, in this embodiment of the application, battery number 128 is used as an example as follows: First channel: CNN spectral feature extraction: When cell #128 exhibits early signs of a micro-short circuit, its voltage ripple is mixed with high-frequency noise components in the second-level data. The lightweight convolutional neural network in the first channel performs a Fast Fourier Transform on the cell's voltage signal, extracts spectral features, and identifies high-frequency noise components in the 100-500Hz frequency range whose energy values ​​exceed the normal baseline by three times. The CNN quantifies this feature as a "high-frequency noise index," which serves as strong evidence of micro-short circuit discharge and is input to the fusion layer.

[0036] Second channel: LSTM temperature timing modeling: Meanwhile, the temperature sensor of cell 128 showed that the temperature had been accelerating from 25.0°C to 0.3°C every 2 minutes over the past 10 minutes. The Long Short-Term Memory (LSTM) network of the second channel performed a time-series modeling of this temperature transient response, predicting that the temperature would reach 27.5°C within the next 20 minutes, exhibiting an exponential upward trend before thermal runaway. The LSTM outputs two time-series features: the "temperature rise rate" and the "predicted peak value," which serve as early warning signals of thermal runaway risk and are input into the fusion layer.

[0037] Dual-channel feature fusion and random forest classification: The fusion layer concatenates the "high-frequency noise index" output by the CNN with the "temperature rise rate" and "predicted peak value" output by the LSTM to form a three-dimensional feature vector. A random forest classifier uses a voting decision based on 100,000 historically labeled normal, warning, and alarm samples: it comprehensively determines that the noise index and temperature rate of the unit both exceed the warning threshold but not the alarm threshold, ultimately outputting a risk assessment level of "warning". The system immediately marks the unit as "yellow warning - early stage of micro-short circuit", triggering pre-protection measures (reducing the current limit of the module from 100A to 30A) and pushing this level information to the operation and maintenance platform. This dual-channel architecture improves the accuracy of micro-short circuit identification to over 95%, providing a warning 30 minutes earlier than a single voltage threshold alarm, significantly reducing the risk of thermal runaway.

[0038] In step S40: the pre-set battery pack correction model comprehensively analyzes the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data, a three-level progressive correction strategy is executed. The execution of the three-level progressive correction strategy includes: S401: The first stage uses the individual cell risk assessment data to perform a weighted correction on the estimated state of charge in the initial battery pack state analysis data, reducing the interference of high-risk cells on the overall estimate. Specifically, in the 10MWh energy storage power station, cell number 128 was assessed as "high-risk - micro-short circuit warning" by S20. Its SOC estimate was artificially high due to an abnormally increased internal resistance (displayed as 68%, but actually only 60%). The first-level correction strategy reduced the SOC estimation weight of this cell from an equal weight (1 / 192) to a weight of 0.3, reducing its interference with the overall SOC calculation of the module (16 cells). After correction, the module SOC decreased from the artificially high 65% to the true 62%, avoiding the abnormal data of high-risk cells "polluting" the overall estimate, improving the accuracy of the battery cluster-level SOC estimation, and providing a reliable benchmark for subsequent power allocation.

[0039] S402: The second stage recalculates the power allocation coefficient of each module in the battery pack based on the corrected state of charge value and generates a power redistribution command. Specifically, based on the module SOC (62%) corrected by S401, the second-level strategy recalculates the power allocation coefficients of the four modules within the battery cluster. The original equalization allocation coefficient was 25% for each module. Now, the allocation coefficient for the third module (with the lowest SOC and containing a micro-short-circuited cell), which contains a high-risk cell, is reduced to 15%, while the allocation coefficients for the remaining three healthy modules are increased to 28.3%. The generated power redistribution command is sent to the power storage converter (PCS), which prioritizes power injection into healthy modules during charging and prioritizes power extraction from healthy modules during discharging. This isolates risky modules at the system level, preventing them from being subjected to large current surges that could lead to short-circuit deterioration, and significantly improving the operational safety of the battery pack.

[0040] S403: The third stage couples the power redistribution command with the temperature distribution field in the initial battery pack state analysis data, corrects the thermal equalization control parameters, and forms the final corrected battery pack state analysis data. In this embodiment, the third-level strategy couples the power redistribution command of S402 with the battery cluster temperature distribution field: infrared thermal imaging shows that the third module generates heat due to abnormal internal resistance, and its temperature is 3°C higher than that of the other modules. The thermal balance control parameters are modified to "activate forced air cooling for the third module and further limit its allowable charge and discharge power to 10kW (originally 15kW)", while increasing power allocation to neighboring modules to compensate for the total output power. This coupling correction suppresses the risk of thermal runaway through physical cooling and reduces heat sources through power limitation, forming a "temperature-power" dual protection. The final corrected battery pack state analysis data shows that the cluster can still operate safely under reduced power operation, avoiding the sudden drop in energy storage capacity caused by directly cutting off the entire cluster, and ensuring the continuity of power supply in the industrial park.

[0041] In this application, a three-level progressive correction strategy (S401-S403) is employed to achieve refined battery pack management, progressing from "individual cell risk isolation" to "pack-level power redistribution" and finally to "thermal-electric coupling protection." Compared to the crude protection of traditional BMS that only disconnects faulty modules, the three-level correction strategy maximizes available capacity while ensuring safety, increasing the power utilization rate of risky modules from 0% (disconnection) to 40% (derating operation). Simultaneously, thermal runaway risk is reduced through thermal equilibrium coupling control, significantly improving the economy, safety, and power supply reliability of the energy storage system. This promotes the transformation of energy storage protection from a "binary switch" to a "continuously adjustable" intelligent management paradigm.

[0042] In S50: the step of estimating the current health status of the battery pack in real time based on the corrected battery pack status analysis data, predicting the health status of the battery pack according to the preset compensation rules to generate a predicted battery pack health value, and comparing it with the preset initial warning threshold to determine whether there is a risk to battery use, includes the following steps: S501: Based on the degradation trajectory of the same model of battery pack in the historical database, a nonlinear regression model of health status value and cycle number is established. Specifically, the system retrieves degradation trajectory data of the same model of lithium iron phosphate battery clusters from the historical database. Analysis shows that before the battery cluster reaches 8000 cycles, the State of Health (SOH) decreases by an average of about 2% per thousand cycles; however, after exceeding 8000 cycles, the degradation rate accelerates non-linearly, decreasing by up to 5% per thousand cycles. Based on this trajectory, a non-linear regression model of SOH and cycle count is established: SOH = 100 - a×N - b×N² (where N is the cycle count, and a and b are fitting parameters). This model accurately characterizes the aging inflection point, providing a scientific benchmark for subsequent dynamic threshold adjustment and avoiding the shortcomings of linear estimation in reflecting accelerated aging characteristics.

[0043] S502: Based on the current number of cycles used by the battery pack, dynamically lower the initial warning threshold so that the warning is triggered when the aging battery pack is at a lower health status value. Specifically, a certain battery cluster has been used 8500 times. Based on the S501 regression model, the system dynamically adjusts the initial warning threshold: the threshold for new batteries is set at 70% SOH, while the threshold for this cluster is lowered to 62% SOH. This adjustment rule allows the aging battery cluster to trigger warnings at an earlier stage (more lenient than new batteries but with lower absolute values), reflecting the objective reality of its capacity decay and internal resistance increase. Compared to the traditional fixed threshold (70%), the dynamic threshold avoids premature and frequent alarms triggered by the normal decline in SOH of aging batteries, while ensuring timely warnings before the actual risk threshold, balancing warning sensitivity and false alarm rate.

[0044] S503: The adjusted dynamic threshold is compared with the predicted battery pack health value in real time. If the health value is lower than the dynamic threshold, it is determined that there is a risk to battery use. Specifically, in this embodiment, the system estimates the current SOH of the battery cluster to be 61.5% in real time and compares it with the dynamic threshold of 62%. Since 61.5% < 62%, a "battery usage risk" is identified, triggering a yellow warning and indicating to maintenance personnel that the cluster's performance has fallen below the aging safety line, requiring preventative maintenance. If a fixed threshold of 70% is still used, the cluster's SOH being far below the threshold will result in long-term false alarms; while the dynamic threshold accurately matches its aging stage, triggering only at the true risk threshold, avoiding false alarm interference, and improving the reliability of the warning and the targeted nature of the maintenance response.

[0045] In step S60: the pre-set battery protection strategy generation model analyzes the battery pack to generate a battery charge / discharge protection strategy, a multi-objective optimization framework is constructed, which includes: S601: Define sub-indicators for the objective function and assign adjustable weight coefficients to the sub-indicators; Specifically, in this embodiment, for a 10MWh lithium iron phosphate energy storage power station with micro-short circuit risk, the system constructs a multi-objective optimization framework with the following objective function: "avoiding thermal runaway" is set as a safety sub-indicator, "maximizing peak-valley arbitrage revenue" is set as an economic sub-indicator, and "minimizing battery capacity degradation" is set as a lifetime sub-indicator. Based on user preferences (commercial and industrial users prioritize safety), the safety sub-indicator is assigned a high weight, economic sub-indicator a medium weight, and lifetime a low weight. This customized objective function clarifies the priority of subsequent strategy optimization, avoids safety risks caused by a single objective (such as pursuing only revenue), and ensures the maximization of the comprehensive value generated by the strategy. In other embodiments, different sub-indicators can be defined according to different rules.

[0046] S602: A reinforcement learning algorithm is used to perform strategy simulation in a digital twin battery pack, traversing combinations of current limit values, voltage cutoff values, and temperature derating curves. Specifically, the reinforcement learning algorithm performs strategy simulation in a digital twin battery pack model: the agent traverses various current limiting values ​​(such as gradually reducing the current limit of the problem module from 100A to 30A in multiple stages), voltage cutoff values ​​(adjusting the charging cutoff voltage from 3.6V to 3.55V per cell), and temperature derating curves (such as various curve slopes where the derating is reduced by 5% for every 1℃ increase above 50℃). The simulation evaluates the comprehensive objective function value of each combination under multiple time periods and operating conditions. After thousands of iterations, the strategy combination that maximizes the function value of "safety + 0.7 × economy + 0.3 × lifespan" is selected. This traversal mechanism overcomes the limitations of human experience and autonomously discovers that "current limit of 30A + voltage reduction of 3.55V + temperature derating curve B" is the optimal solution, achieving a comprehensive exploration of the strategy space.

[0047] S603: Select the strategy that maximizes the comprehensive objective function as the optimal battery charge and discharge protection strategy, and send it to the battery management system for execution via the northbound interface; Specifically, simulation results show that "current limiting of 30A + voltage reduction of 3.55V + temperature derating curve B" maximizes the overall objective function. The system selects this strategy as the optimal battery charge / discharge protection strategy, converts it into a control command set conforming to the communication protocol through the northbound interface, and sends it to the battery management system for execution: the PCS receives the current limiting command, and the BMS receives the voltage cutoff and temperature derating curve commands. After execution, the current of the micro-short-circuit module is limited, the temperature rise rate is significantly slowed down, the safety indicators meet the standards, and the power of other healthy modules is increased to compensate for the total capacity, minimizing economic losses. This selection mechanism ensures that the issued strategy is fully verified by the digital twin, avoids the trial-and-error risk of the physical system, and improves the reliability and execution effect of the protection strategy.

[0048] The multi-objective optimization and reinforcement learning strategy generation framework constructed through S601-S603 systematically addresses the core pain points of traditional BMS protection strategies: singularity, rigidity, and inability to balance multiple objectives. S601's custom weights allow strategies to match user value preferences; S602's reinforcement learning simulation traversal breaks through the boundaries of human experience, autonomously discovering the globally optimal strategy; and S603's digital twin pre-simulation verification ensures the strategy's safety and reliability. The synergy of these three elements shifts energy storage protection from a "fixed threshold" to "global optimization," maximizing economic benefits while ensuring safety. It curbs micro-short circuit risks at their inception, reduces capacity loss due to overly stringent protection strategies, and improves the overall operational efficiency and long-term economic viability of energy storage systems.

[0049] After the step of analyzing the battery pack using the pre-set battery protection strategy generation model in S60 to generate a battery charge and discharge protection strategy, the following steps are included: S61: Obtain the current working scenario data of the battery pack. The working scenario data includes the time-of-use electricity price period identifier in the home energy storage scenario, the dynamic curve of the production line load in the industrial energy storage scenario, and the power dispatch instruction sequence in the power grid frequency regulation scenario. Specifically, the system acquires data on the current operating scenarios of the battery pack: In industrial energy storage scenarios, the dynamic load curve of the production line is collected through the park's EMS system. The data shows that the stamping workshop equipment is concentratedly started from 10:00 to 11:30 every day, with the load jumping from 2MW to 5MW; in frequency regulation ancillary service scenarios, the system receives a sequence of grid dispatch AGC commands, which require the energy storage system to provide a rapid frequency regulation response of ±1MW between 09:45 and 10:15. This scenario data is tagged and input into subsequent analysis stages, enabling the protection strategy to perceive external load impacts and changes in dispatch commands. This avoids the limitations of traditional BMS, which only focus on internal battery parameters and ignore changes in external operating conditions, thus improving the strategy's foresight and scenario adaptability.

[0050] S62: Calculate the residual between the predicted battery pack health value and the subsequently collected actual health status value, mark the residual as the prediction deviation, and store the prediction deviation in the scenario-deviation association database according to the working scenario. Specifically, during the 10:00-11:30 production line load surge period, the system predicted a health status of 85% based on S50. However, the actual collected battery pack health status value dropped to 82% due to accelerated internal temperature rise caused by high-current discharge, resulting in a 3% prediction deviation. The system categorized this deviation as "Industrial Load Surge - Prediction Lag Deviation" and stored the deviation value, load curve, temperature rise rate, and other data in the "Industrial Energy Storage Scenarios" category of the scenario-deviation association database. This categorized storage mechanism accumulates deviation patterns under different scenarios, providing a scenario-tagged sample library for subsequent attribution analysis. This enables the analysis to distinguish between "deviation caused by load surge" and "deviation caused by normal aging," improving attribution accuracy.

[0051] S63: Perform attribution analysis on the scenario-deviation association database to identify key influencing factors that lead to prediction deviations in different scenarios and generate scenario correction coefficients; Specifically, the system performs attribution analysis on the scenario-deviation correlation database: through machine learning algorithms, it was found that in the "industrial load impact" scenario, the key influencing factors of prediction deviation are "load ramp-up rate" and "uniformity of the internal temperature field of the battery cluster," which together account for 78%. Based on this, a scenario correction coefficient is generated: for periods when the load ramp-up rate exceeds a threshold, the weight of the temperature rise coefficient in the health status prediction model is increased by 30% to predict the negative impact of high current impact on health in advance. This correction coefficient improves the model's prediction accuracy from 85% to 92% in subsequent similar load impact scenarios, avoiding risk misjudgment caused by prediction lag.

[0052] S64: Use the scenario correction coefficient to perform weighted correction on the health state prediction branch in the battery protection strategy generation model, generate a corrected protection strategy suitable for a specific scenario, improve the prediction accuracy and make the strategy more in line with the actual working conditions. Using scenario correction coefficients generated by S63, the health state prediction branch in the battery protection strategy generation model is weighted and corrected: Under industrial load impact scenarios, the confidence weight of the SOC estimation is lowered, while the weight of the temperature field uniformity feature is increased, generating a corrected protection strategy suitable for this scenario—reducing the energy storage discharge power limit to 60% of the rated value 30 minutes before the daily load peak from 10:00 to 11:30, and initiating active balancing to optimize temperature distribution. This corrected strategy successfully avoids health state prediction bias in subsequent load impact cycles, while controlling the frequency regulation response deviation rate to within 1%, meeting the requirements of grid ancillary services. This mechanism transforms the protection strategy from scenario-general to scenario-specific, accurately matching the actual operating conditions of industrial parks and improving system safety and economy.

[0053] The scenario correction coefficients are not only used to correct the health state prediction branch of S64, but also fed back into S602 as input constraints to dynamically adjust the objective function weights and simulation parameters. For example, in an industrial load impact scenario, after the scenario correction coefficients increase the weight of the temperature rise coefficient, reinforcement learning reduces the simulation traversal frequency of high-current strategies and prioritizes optimizing thermal equilibrium control to prevent thermal runaway. Conversely, the strategy execution results generated by reinforcement learning (such as the actual thermal runaway mitigation rate) are used to update the scenario-deviation association database, improving the accuracy of subsequent corrections. Together, they improve the physical safety state of the battery pack and optimize economics and lifespan.

[0054] This application systematically addresses the core pain points of traditional BMS protection strategies—the inability to adapt to multiple scenario switching and the inability to learn and optimize from deviations—through a closed loop constructed by S61-S63: scenario perception, deviation learning, and strategy correction. S61's scenario data acquisition enables the system to perceive external operating conditions; S62's deviation classification and storage accumulates scenario-based knowledge; and S63's attribution analysis and coefficient correction achieve adaptive model optimization. The synergy of these three elements allows the energy storage protection system to "remember" and "adapt" to different scenario characteristics, significantly improving prediction accuracy, strategy matching, and reducing false alarm rates during scenario switching. This drives the autonomous evolution of energy storage systems from "fixed parameters" to "scenario intelligence," providing highly reliable and economical energy storage security for complex scenarios such as industrial parks.

[0055] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0056] In one embodiment, a smart charge / discharge protection device for an energy storage battery pack is provided, which corresponds one-to-one with the smart charge / discharge protection method for the energy storage battery pack described in the above embodiments. For example... Figure 2 As shown, the intelligent charging and discharging protection device for the energy storage battery pack includes: a single cell state data set construction unit 1, which is used to collect real-time state data of single cells in real time and associate all collected real-time state data of single cells with a preset common time axis to construct a single cell state data set. The single-cell battery risk assessment data generation unit 2 is used to pre-set a single-cell battery anomaly analysis model to analyze the real-time status data of the single-cell battery, generate single-cell battery risk assessment data, and actively trigger pre-protection measures when a risk is predicted to exist in the single-cell battery. The initial battery pack state analysis data generation unit 3 is used to pre-set the battery pack analysis model to analyze the single cell state dataset based on a machine learning algorithm in order to generate initial battery pack state analysis data. The battery pack state analysis data generation unit 4 is used to pre-set a battery pack correction model to comprehensively analyze the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. The battery usage risk judgment unit 5 is used to estimate the current health status value of the battery pack in real time based on the corrected battery pack status analysis data, and predict the health status value of the battery pack according to the preset compensation rules to generate a predicted battery pack health value, and compare it with the preset initial warning threshold to determine whether there is a battery usage risk. The battery charge / discharge protection strategy generation unit 6 is used to analyze the battery pack using a pre-set battery protection strategy generation model when there is a risk to battery use, in order to generate a battery charge / discharge protection strategy.

[0057] Specific limitations regarding the intelligent charge-discharge protection device for energy storage battery packs can be found in the limitations of the intelligent charge-discharge protection method for energy storage battery packs mentioned above, and will not be repeated here. Each module in the aforementioned intelligent charge-discharge protection device for energy storage battery packs can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the electronic device in hardware form or independent of it, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0058] In one embodiment, an electronic device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3As shown, this electronic device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores the database. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a smart charging and discharging protection method for an energy storage battery pack.

[0059] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Real-time status data of individual cells is collected, and all collected real-time status data of individual cells are associated with a pre-set common time axis to construct an individual cell status dataset. A pre-set single-cell abnormality analysis model analyzes the real-time status data of the single cell to generate single-cell risk assessment data, and actively triggers pre-protection measures when a risk is predicted to exist in the single cell. A pre-set battery pack analysis model analyzes the individual cell state dataset to generate initial battery pack state analysis data. The pre-set battery pack correction model comprehensively analyzes the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. Based on the corrected battery pack status analysis data, the current health status value of the battery pack is estimated in real time, and the health status value of the battery pack is predicted according to the preset compensation rules to generate the predicted battery pack health value, which is then compared with the preset initial warning threshold to determine whether there is a risk to battery use. If there is a risk to battery use, a pre-set battery protection strategy generation model will analyze the battery pack to generate a battery charge and discharge protection strategy.

[0060] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Real-time status data of individual cells is collected, and all collected real-time status data of individual cells are associated with a pre-set common time axis to construct an individual cell status dataset. A pre-set single-cell abnormality analysis model analyzes the real-time status data of the single cell to generate single-cell risk assessment data, and actively triggers pre-protection measures when a risk is predicted to exist in the single cell. A pre-set battery pack analysis model analyzes the individual cell state dataset to generate initial battery pack state analysis data. The pre-set battery pack correction model comprehensively analyzes the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. Based on the corrected battery pack status analysis data, the current health status value of the battery pack is estimated in real time, and the health status value of the battery pack is predicted according to the preset compensation rules to generate the predicted battery pack health value, which is then compared with the preset initial warning threshold to determine whether there is a risk to battery use. If there is a risk to battery use, a pre-set battery protection strategy generation model will analyze the battery pack to generate a battery charge and discharge protection strategy.

[0061] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0062] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0063] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for intelligent charging and discharging protection of an energy storage battery pack, characterized in that, The method includes the following steps: real-time acquisition of real-time status data of individual cells, and association of all acquired real-time status data of individual cells with a pre-set common time axis to construct an individual cell status dataset. A pre-set single-cell abnormality analysis model analyzes the real-time status data of the single cell to generate single-cell risk assessment data, and actively triggers pre-protection measures when a risk is predicted to exist in the single cell. A pre-set battery pack analysis model analyzes the individual cell state dataset to generate initial battery pack state analysis data. The pre-set battery pack correction model comprehensively analyzes the individual cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. Based on the corrected battery pack status analysis data, the current health status value of the battery pack is estimated in real time, and the health status value of the battery pack is predicted according to the preset compensation rules to generate the predicted battery pack health value, which is then compared with the preset initial warning threshold to determine whether there is a risk to battery use. If there is a risk to battery use, a pre-set battery protection strategy generation model will analyze the battery pack to generate a battery charge and discharge protection strategy.

2. The intelligent charging and discharging protection method for an energy storage battery pack according to claim 1, characterized in that, Before collecting real-time status data of individual cells and associating all collected real-time status data of individual cells with a pre-set common time axis to construct an individual cell status dataset, the following steps are included: By synchronously acquiring the surface temperature field distribution, electrolyte pressure micro-vibration signal, and electrode potential transient waveform of individual battery cells through edge computing nodes, a multimodal state vector of a single battery cell is constructed. By utilizing a preset synchronous triggering mechanism, the multimodal state vector is aligned with the common time axis at the nanosecond level to eliminate sensor sampling delay differences; The aligned data stream is then subjected to sliding window filtering and outlier removal to generate a preprocessed single-cell state dataset.

3. The intelligent charging and discharging protection method for an energy storage battery pack according to claim 1, characterized in that, The step of analyzing the real-time status data of the single cell using the pre-set single cell anomaly analysis model includes the following steps: The first channel deploys a lightweight convolutional neural network to extract spectral features of the single-unit voltage ripple and identify high-frequency noise components in the early stage of a micro-short circuit. The second channel deploys a long short-term memory network to perform time-series modeling of the temperature transient response and predict the accelerated temperature rise trend before thermal runaway. The dual-channel output results are fused and input into a random forest classifier to generate a single-cell battery risk assessment level.

4. The intelligent charging and discharging protection method for an energy storage battery pack according to claim 1, characterized in that, In the step of generating corrected battery pack state analysis data by comprehensively analyzing the individual cell risk assessment data and the initial battery pack state analysis data using a pre-set battery pack correction model, a three-level progressive correction strategy is executed. This three-level progressive correction strategy includes: The first stage uses individual cell risk assessment data to weight and correct the estimated state of charge in the initial battery pack state analysis data, reducing the interference of high-risk cells on the overall estimate. The second stage, based on the corrected state of charge value, recalculates the power allocation coefficient of each module in the battery pack and generates a power redistribution command. The third stage couples the power redistribution command with the temperature distribution field in the initial battery pack state analysis data, corrects the thermal equalization control parameters, and forms the final corrected battery pack state analysis data.

5. The intelligent charging and discharging protection method for an energy storage battery pack according to claim 1, characterized in that, The step of estimating the current health status of the battery pack in real time based on corrected battery pack status analysis data, predicting the battery pack health status according to preset compensation rules to generate a predicted battery pack health value, and comparing it with a preset initial warning threshold to determine whether there is a risk to battery use includes the following steps: Based on the degradation trajectory of the same model of battery pack in the historical database, a nonlinear regression model of health status value and cycle number is established. Based on the current number of battery cycles used, the initial warning threshold is dynamically lowered to form a dynamic threshold, so that the aging battery pack triggers a warning when it is at a lower health status value. The dynamic threshold is compared with the predicted battery pack health value in real time. If the health value is lower than the dynamic threshold, it is determined that there is a risk to the use of the battery.

6. The intelligent charging and discharging protection method for an energy storage battery pack according to claim 1, characterized in that, In the step of analyzing the battery pack using a pre-set battery protection strategy generation model to generate a battery charge / discharge protection strategy, a multi-objective optimization framework is constructed, which includes: Define sub-indicators for the objective function and assign adjustable weight coefficients to them; A reinforcement learning algorithm is used to simulate the strategy in a digital twin battery pack, traversing combinations of current limit values, voltage cutoff values, and temperature derating curves. The strategy that maximizes the comprehensive objective function is selected as the optimal battery charge and discharge protection strategy, and is then sent to the battery management system for execution via the northbound interface.

7. The intelligent charging and discharging protection method for an energy storage battery pack according to claim 6, characterized in that, After analyzing the battery pack using a pre-set battery protection strategy generation model to generate a battery charge / discharge protection strategy, the following steps are included: Acquire the current working scenario data of the battery pack, which includes the time-of-use electricity price period identifier in the home energy storage scenario, the dynamic curve of the production line load in the industrial energy storage scenario, and the power dispatch instruction sequence in the power grid frequency regulation scenario. Calculate the residual between the predicted battery pack health value and the subsequently collected actual health status value, mark the residual as the prediction deviation, and store the prediction deviation in the scenario-deviation association database according to the working scenario. Attribution analysis is performed on the scenario-deviation association database to identify key influencing factors that lead to prediction deviations in different scenarios and generate scenario correction coefficients. The scenario correction coefficients are used to perform weighted corrections on the health state prediction branch in the battery protection strategy generation model, generating a corrected protection strategy suitable for a specific scenario, thereby improving prediction accuracy and making the strategy more in line with actual working conditions.

8. An intelligent charge-discharge protection device for an energy storage battery pack, applied to the intelligent charge-discharge protection method for an energy storage battery pack as described in any one of claims 1 to 7, characterized in that, The device includes: The single cell state dataset construction unit (1) is used to collect real-time state data of single cells in real time and associate all collected real-time state data of single cells with a pre-set common time axis to construct a single cell state dataset. The single cell risk assessment data generation unit (2) is used to pre-set the single cell abnormality analysis model to analyze the real-time status data of the single cell, generate single cell risk assessment data, and actively trigger pre-protection measures when the single cell is predicted to have a risk. The initial battery pack state analysis data generation unit (3) is used to pre-set the battery pack analysis model to analyze the single cell state dataset based on the machine using a self-learning algorithm to generate initial battery pack state analysis data. The battery pack state analysis data generation unit (4) is used to pre-set the battery pack correction model to comprehensively analyze the single cell risk assessment data and the initial battery pack state analysis data to generate corrected battery pack state analysis data. The battery use risk judgment unit (5) is used to estimate the current health status value of the battery pack in real time based on the corrected battery pack status analysis data, and predict the health status value of the battery pack according to the preset compensation rules to generate the predicted battery pack health value, and compare it with the preset initial warning threshold to determine whether there is a battery use risk. The battery charge and discharge protection strategy generation unit (6) is used to analyze the battery pack using a pre-set battery protection strategy generation model when there is a risk of battery use, so as to generate a battery charge and discharge protection strategy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent charging and discharging protection method for an energy storage battery pack as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent charging and discharging protection method for an energy storage battery pack as described in any one of claims 1 to 7.

Images