Multi-stage early warning linkage control method and system for energy storage system

Abstract

The invention discloses a multistage early warning linkage control method and system for an energy storage system, and relates to the technical field of safety of battery energy storage systems, and the method comprises the steps: obtaining multi-modal real-time data streams, including voltage, current, temperature, specific gas concentration and internal pressure of a box body, of a single battery / battery box layer in the energy storage system; performing time sequence alignment and feature level fusion on the multi-modal real-time data stream to generate a fusion feature vector; inputting the fusion feature vector into a pre-trained dynamic risk assessment model, and outputting a comprehensive risk score and a corresponding early warning level of the battery; and according to the early warning grade, executing a matched linkage control strategy by following a grading progressive principle. According to the method, the basic statistical characteristics, the related characteristics and the model driving characteristics are fused, the coupling rule among the parameters is comprehensively identified, the pre-trained dynamic risk assessment model is matched, the comprehensive risk score and the early warning level are output, and the safety protection requirements of different risk levels are accurately matched.

Description

Technical Field

[0001] This invention relates to the field of battery energy storage system safety technology, and in particular to a multi-level early warning linkage control method and system for energy storage systems. Background Technology

[0002] With the large-scale application of new energy power generation and the increasing demand for grid peak shaving, energy storage systems have become core supporting equipment in the energy system, widely used on the generation side, grid side, and user side. However, energy storage systems use lithium batteries as the core energy storage unit, and their operation is susceptible to factors such as charge / discharge rate, ambient temperature, and battery aging, which can lead to safety hazards such as thermal runaway, gas leakage, and abnormal pressure, and in turn cause major accidents such as fires and explosions.

[0003] Existing early warning and control methods for energy storage systems mostly rely on single-dimensional data, directly triggering an alarm when battery parameters (such as temperature, voltage, and gas concentration) exceed a threshold. This approach cannot differentiate the severity of faults; for example, minor temperature fluctuations and high temperatures indicating impending thermal runaway are treated with the same alarm level. While methods utilizing multimodal data have emerged, they only involve simple data stitching, resulting in an inability to effectively identify the coupling relationships between parameters such as voltage, current, temperature, gas concentration, and pressure, making it difficult to accurately match the safety protection requirements of different risk levels. Summary of the Invention

[0004] To address the problem that existing technologies cannot effectively identify the coupling relationships between parameters and are difficult to accurately match the safety protection requirements of different risk levels, this invention provides a multi-level early warning and linkage control method and system for energy storage systems. This method can identify the coupling relationships between various parameters, and, in conjunction with a pre-trained dynamic risk assessment model, accurately output a comprehensive risk score and early warning level, precisely matching the safety protection requirements of different risk levels. This achieves early identification of safety hazards. The specific technical solution is as follows: This invention provides a multi-level early warning and linkage control method for energy storage systems, comprising: Step S1: Obtain multimodal real-time data streams at the battery cell / battery box level in the energy storage system, including voltage, current, temperature, specific gas concentration, and internal pressure of the box; Step S2: Perform time-series alignment and feature-level fusion on the multimodal real-time data stream to generate a fused feature vector, including: A sliding time window is used to truncate and align the multimodal real-time data stream. Within each aligned window, the following features are extracted synchronously: The rise gradient of the temperature series, the fluctuation variance of the voltage series, and the number of abrupt changes in the current series are calculated as basic statistical features. The peak value of the cross-correlation function between the gas concentration series and the pressure series is calculated as the correlation feature; Based on the battery electrothermal model, temperature and current data are coupled and analyzed to generate electrothermal coupling features, and gas-pressure correlation features are generated based on the gas diffusion model and gas concentration and pressure data, which serve as model-driven features. The basic statistical features, the correlation features, and the model-driven features are weighted and combined to generate the fused feature vector. Step S3: Input the fused feature vector into the pre-trained dynamic risk assessment model and output the battery's comprehensive risk score and corresponding warning level; Step S4: Based on the warning level, execute the corresponding linkage control strategy according to the hierarchical progression principle.

[0005] Preferably, the weighted combination of the basic statistical features, the correlation features, and the model-driven features includes: Obtain the current health status and operating conditions of the batteries in the energy storage system; Based on the current health status of the battery, select an appropriate subset of features for computation from a pre-configured feature extraction strategy library; Based on the current operating conditions of the battery, the fusion weights of each feature in the selected feature calculation subset are dynamically adjusted. The pre-configured feature extraction strategy library includes at least: a first type of strategy focusing on battery performance and life degradation assessment, and a second type of strategy focusing on battery safety and sudden fault response.

[0006] Preferably, selecting a suitable feature calculation subset from a pre-configured feature extraction strategy library based on the current health status of the battery includes: Based on the current health status of the battery, the battery health status is divided into multiple intervals, and an optimized feature subset is preset for each health status interval. The optimized feature subset is obtained by analyzing the data before the fault occurred in the corresponding interval in the historical fault case library.

[0007] Preferably, the step of dynamically adjusting the fusion weights of each feature in the selected feature calculation subset based on the current operating condition of the battery includes: Based on the current operating conditions of the battery, determine the dominant risk type most relevant to the current operating conditions; Based on the risk-dominant type, the weight ratio of the corresponding feature in the fused feature vector is dynamically adjusted; The operating conditions include at least ambient temperature and charge / discharge rate, and the risk-dominant types include thermal runaway risk-dominant, electrical overstress risk-dominant, or gas leakage risk-dominant.

[0008] Preferably, the step of generating gas-pressure correlation features by performing correlation analysis on gas concentration and chamber pressure data based on a gas diffusion model includes: Calculate the cross-correlation function between the gas concentration sequence and the pressure sequence within a sliding window to obtain the corresponding peak value; When the peak value exceeds a first preset threshold, a correlation analysis based on a gas diffusion model is triggered to calculate the model conformity between the concentration change and the pressure change, which is used as the gas-pressure correlation feature.

[0009] Preferably, the step of generating electrothermal coupling features by performing coupled analysis of temperature and current data based on the battery electrothermal model includes: Based on the current sequence and battery parameters within the sliding window, the theoretical temperature rise curve of the battery is calculated using an electrochemical-thermal coupling model. The theoretical temperature rise curve is compared with the measured temperature sequence within the same window, and the sum of squared residuals is calculated to generate an electrothermal coupling characteristic that characterizes the balance between heat generation and heat dissipation.

[0010] Preferably, the dynamic risk assessment model is a time-series prediction model built on long short-term memory network and attention mechanism. The time-series prediction model is trained with historical normal data and multi-level fault data, and is used to output a comprehensive risk score of the battery system according to the time-series evolution trend of the fused feature vector, and to determine the corresponding warning level based on the comprehensive risk score.

[0011] Preferably, the step of inputting the fused feature vector into a pre-trained dynamic risk assessment model and outputting a comprehensive risk score and corresponding warning level for the battery system includes: The temporal dependency pattern of the fused feature vector sequence is extracted using a long short-term memory network, and the hidden state sequence is output. The hidden state sequence is analyzed using an attention mechanism to calculate the importance weight of features at different time steps to the current risk assessment and to identify key abnormal segments. The final output layer of the model is based on the weighted context vector, and generates a continuous comprehensive risk score and a multidimensional risk vector. The multidimensional risk vector represents the sub-scores of thermal runaway risk, electrical fault risk and gas leakage risk, respectively.

[0012] The present invention also provides a multi-level early warning and linkage control system for an energy storage system, which applies the aforementioned method and includes: The data acquisition unit is used to acquire multimodal real-time data streams at the battery cell / battery box level in the energy storage system, including voltage, current, temperature, specific gas concentration, and internal pressure of the box; The feature fusion unit is used to perform time-series alignment and feature-level fusion on the multimodal real-time data stream to generate a fused feature vector, including: A sliding time window is used to truncate and align the multimodal real-time data stream. Within each aligned window, the following features are extracted synchronously: The rise gradient of the temperature series, the fluctuation variance of the voltage series, and the number of abrupt changes in the current series are calculated as basic statistical features. The peak value of the cross-correlation function between the gas concentration series and the pressure series is calculated as the correlation feature; Based on the battery electrothermal model, temperature and current data are coupled and analyzed to generate electrothermal coupling features, and gas-pressure correlation features are generated based on the gas diffusion model and gas concentration and pressure data, which serve as model-driven features. The basic statistical features, the correlation features, and the model-driven features are weighted and combined to generate the fused feature vector. The risk warning output unit is used to input the fused feature vector into the pre-trained dynamic risk assessment model and output the battery's comprehensive risk score and corresponding warning level. The linkage control unit is used to execute a matching linkage control strategy according to the warning level and following the principle of hierarchical progression.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention discloses a multi-level early warning linkage control method and system for energy storage systems. By using a sliding time window to achieve time-series alignment of multimodal data streams, it integrates basic statistical features, correlation features, and model-driven features to comprehensively identify the coupling patterns between various parameters. In conjunction with a pre-trained dynamic risk assessment model, it accurately outputs a comprehensive risk score and early warning level, effectively avoiding missed reports caused by single data monitoring, accurately matching the safety protection requirements of different risk levels, and achieving early identification of safety hazards. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0015] Figure 1 This is a flowchart of a multi-level early warning and linkage control method for an energy storage system according to the present invention.

[0016] Figure 2 This is a flowchart of another embodiment of the multi-level early warning and linkage control method for an energy storage system according to the present invention.

[0017] Figure 3 This is a schematic diagram of a multi-level early warning and linkage control system for an energy storage system according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that, when used in this specification, the terms “comprising” and “including” indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes such combinations.

[0022] Please refer to the following examples. Figures 1 to 3 .

[0023] This invention provides a multi-level early warning and linkage control method for an energy storage system, comprising: Step S1: Obtain multimodal real-time data streams at the battery cell / battery box level in the energy storage system, including voltage, current, temperature, specific gas concentration, and internal pressure of the box; Hall voltage sensors and closed-loop Hall current sensors are used, which are directly installed on the terminals of individual battery cells or the busbars of battery clusters to monitor voltage and current.

[0024] Digital temperature sensors are used and are attached to the surface of individual battery cells and key areas inside the battery box in a multi-point arrangement to monitor the temperature field distribution and hot spots.

[0025] Gas sensors employing electrochemical or semiconductor principles are used to detect characteristic gases such as carbon monoxide, hydrogen, and electrolyte vapor that may be released in the early stages of battery thermal runaway. The sensors are deployed at the top of the battery compartment where gases tend to accumulate and near ventilation openings.

[0026] A miniature piezoresistive pressure sensor is installed inside the sealed cavity of the battery box to monitor internal pressure changes caused by gas generation or thermal expansion.

[0027] All sensors are connected to the local data acquisition unit via a CAN bus or distributed I / O module. The data acquisition unit synchronously or quasi-synchronously acquires data from each sensor at a fixed frequency, and performs preliminary AD conversion, dimension normalization, and invalid value filtering to form a timestamp-aligned raw data packet.

[0028] Step S2: Perform time-series alignment and feature-level fusion on the multimodal real-time data stream to generate a fused feature vector, including: A sliding time window is used to truncate and align the multimodal real-time data stream. Within each aligned window, the following features are extracted synchronously: Linear interpolation is used to align all channel data to a unified time series. Let the target time series be... The time interval is the sensor's highest sampling interval. For any sensor, if at the target time... If no sampling point is available, then the nearest sampling point before or after it is used. and ( ) data and ,calculate Interpolation at time: Thus, all sensors are obtained. Real-time synchronized data.

[0029] Set window length and sliding step size For example, setting s, s. If the data sampling interval after alignment is ms, then each window contains There are [number] data points. Starting from the initial moment, the window captures the latest data point with each slide. Each point is treated as a processing unit.

[0030] The rise gradient of the temperature series, the fluctuation variance of the voltage series, and the number of abrupt changes in the current series are calculated as basic statistical features. Since there may be slight delays in sampling from different sensors, this embodiment employs a timestamp-based interpolation algorithm to align the data from all channels in time, ensuring that parameter values ​​at the same moment correspond to the same physical state. A configurable sliding time window is maintained. This window continuously slides forward, capturing all aligned time-series data within the window as a processing unit each time. Within the window, statistics are directly calculated from the time-series data.

[0031] Temperature sequence gradient: Calculate the linear regression slope of the temperature time series within the window, or calculate the temperature difference between the end and the beginning time divided by time to characterize the rate of temperature rise.

[0032] Let the temperature sequence within the window be... Temperature sequence ascending gradient Linear regression was used. The time points were used as the basis for the analysis. Temperature is the independent variable. As the dependent variable, fit a straight line. The slope is solved using the least squares method. : Then the ascending gradient .

[0033] Voltage series fluctuation variance: Calculates the variance of voltage values ​​within a window relative to their mean, characterizing voltage stability.

[0034] Let the voltage sequence within the window be Voltage series fluctuation variance Calculate the variance of the voltage sequence: Number of current sequence abrupt changes: Set a threshold for the rate of change of current, and count the number of times the rate of change of current exceeds the threshold within a window, representing the possibility of load abrupt changes or internal short circuits.

[0035] Let the current sequence within the window be... Number of abrupt changes in the current sequence First, calculate the current change rate sequence: , Set a threshold for the rate of change. According to the battery's rated current The percentage is determined, for example A / s. Statistically satisfied Number of: in, This is an indicator function.

[0036] The peak value of the cross-correlation function between the gas concentration series and the pressure series is calculated as the correlation feature; Calculate the cross-correlation function between the gas concentration series and the pressure series within the window, and find its maximum peak value and corresponding time lag. This peak value reflects the temporal correlation strength between gas production and pressure changes.

[0037] The calculation assumes the gas concentration sequence within the window is as follows: The pressure sequence is Calculate the normalized cross-correlation function (NCCF) of the two sequences: in, The maximum time delay is determined based on the physical time of gas diffusion. The maximum absolute value is taken as the cross-correlation peak value. Simultaneously record the time delay when the peak value is reached. .

[0038] Based on the battery electrothermal model, temperature and current data are coupled and analyzed to generate electrothermal coupling features, and gas-pressure correlation features are generated based on the gas diffusion model and gas concentration and pressure data, which serve as model-driven features. The gas-pressure correlation features generated by the correlation analysis of gas concentration and chamber pressure data based on the gas diffusion model include: Calculate the cross-correlation function between the gas concentration sequence and the pressure sequence within a sliding window to obtain the corresponding peak value; When the peak value exceeds a first preset threshold, a correlation analysis based on a gas diffusion model is triggered to calculate the model conformity between the concentration change and the pressure change, which is used as the gas-pressure correlation feature.

[0039] when At a threshold of (first preset threshold), gas diffusion model analysis is triggered. A simplified model using the ideal gas law and Fick's diffusion law is employed: in, For pressure, The gas constant is For battery temperature, take the average value within a window. For the free volume of the box, This represents the number of moles of gas. Gas concentration. and satisfy Within the window, calculate using the central difference method. and The sequence is then linearly fitted: Calculate the coefficient of determination of the fit As a model fit: By defining gas-pressure correlation characteristics . The closer the value is to 1, the higher the fit of the gas diffusion model.

[0040] The generation of electrothermal coupling features by coupling analysis of temperature and current data based on the battery electrothermal model includes: Based on the current sequence and battery parameters within the sliding window, the theoretical temperature rise curve of the battery is calculated using an electrochemical-thermal coupling model. The theoretical temperature rise curve is compared with the measured temperature sequence within the same window, and the sum of squared residuals is calculated to generate an electrothermal coupling characteristic that characterizes the balance between heat generation and heat dissipation.

[0041] A battery electrothermal coupling model is adopted. The model consists of an electrical model and a thermal model: The electrical model uses a second-order RC equivalent circuit model to calculate the battery's internal resistance. and polarization heat .

[0042] The thermal equilibrium equation for the battery in the thermal model is: in, For heat capacity, For heat dissipation coefficient, It is a reversible reaction heat. The ambient temperature can be extracted from the temperature series or measured separately. The current series within the window is utilized. and initial temperature The theoretical temperature series can be calculated recursively using the Euler method or the Runge-Kutta method. Calculate the sum of squared residuals (RSS) between the theoretical and measured sequences: By defining electrothermal coupling characteristics ,in Let V be the variance of the temperature series. The larger the value, the more unbalanced the heat production and dissipation.

[0043] The basic statistical features, the correlation features, and the model-driven features are weighted and combined to generate the fused feature vector. The quantization process of weighted feature combination constructs the original feature vector from the above five features: It should be noted that in practice, there may be more or fewer features; this embodiment uses six as an example.

[0044] Each feature accumulates historical data over long-term operation, and its mean is calculated. and standard deviation ( z-score normalization was used. Alternatively, minimum-maximum normalization can be used to normalize to the [0,1] interval.

[0045] Weight vector Initially, uniform weights The system dynamically adjusts based on the battery's state of health (SOH) and real-time operating conditions. SOH Impact: When SOH falls below 80% of the threshold, the battery's internal resistance increases, leading to increased thermal risk. Increase the weighting of temperature-related characteristics: Set an adjustment factor. Then the temperature gradient and electrothermal coupling Updated to , .

[0046] Influence of ambient temperature: Ambient temperature Above 35℃, increase the weight of temperature-related features; below 0℃, appropriately decrease it. Define the temperature weighting factor. ,but and Updated to , .

[0047] Effect of charge / discharge rate: rate .when At the same time, increase the weight of electrical characteristics: define the multiplier factor. Then the voltage variance and sudden changes in current Updated to , .

[0048] Feature vector generation: This vector is the fusion feature vector of the input dynamic risk assessment model.

[0049] The extracted basic statistical features, correlation features, and model-driven features are combined into a multi-dimensional fusion feature vector.

[0050] A weighted summation method can be used during fusion. The weights can be dynamically adjusted based on the battery's state of health (SOH) and real-time operating conditions (such as ambient temperature and charge / discharge rate). For example, under high-temperature conditions, the weights of temperature gradient and electrothermal coupling characteristics are increased; during high-rate charge / discharge, the weights of current surges and voltage fluctuations are increased. The weight configuration strategy can be determined in advance through experiments or simulations and stored in a strategy library.

[0051] Step S3: Input the fused feature vector into the pre-trained dynamic risk assessment model and output the battery's comprehensive risk score and corresponding warning level; The dynamic risk assessment model is selected as a time-series prediction model built on an attention-enhanced long short-term memory network (LSTM). The time-series prediction model is trained with historical normal data and multi-level fault data. It is used to output a comprehensive risk score of the battery system based on the time-series evolution trend of the fused feature vector, and to determine the corresponding warning level based on the comprehensive risk score.

[0052] The temporal dependency pattern of the fused feature vector sequence is extracted using a long short-term memory network, and the hidden state sequence is output. The hidden state sequence is analyzed using an attention mechanism to calculate the importance weight of features at different time steps to the current risk assessment and to identify key abnormal segments. The final output layer of the model is based on the weighted context vector, and generates a continuous comprehensive risk score and a multidimensional risk vector. The multidimensional risk vector represents the sub-scores of thermal runaway risk, electrical fault risk and gas leakage risk, respectively.

[0053] In this dynamic risk assessment model, the last two layers are as follows: the first layer is a fully connected layer that outputs a scalar score representing the overall risk level; the second layer is a softmax layer or a threshold comparison layer that maps the scalar score to a specific warning level. Alternatively, the model output layer has two branches: one branch outputs a continuous overall risk score, for example, a value between 0 and 1, through a fully connected layer and an activation function; the other branch can output a multi-dimensional risk vector, representing the probability or intensity of sub-categories such as thermal runaway risk, electrical overstress risk, and gas leakage risk.

[0054] During online operation, the real-time generated fusion feature vector sequence is input into the trained model. The model outputs the current comprehensive risk score. The system presets three risk thresholds: Level 1 warning threshold, Level 2 warning threshold, and Level 3 warning threshold, with Level 1 warning threshold < Level 2 warning threshold < Level 3 warning threshold.

[0055] Logic for determining warning levels: If the overall risk score is less than the Level 1 warning threshold: the status is normal and there is no warning.

[0056] If the Level 1 warning threshold is less than or equal to the comprehensive risk score but less than the Level 2 warning threshold, a Level 1 warning is triggered, indicating a minor anomaly or potential risk trend that requires attention from operations and maintenance personnel.

[0057] If the Level 2 warning threshold is less than or equal to the comprehensive risk score but less than the Level 3 warning threshold, a Level 2 warning is triggered, indicating that a moderate fault has occurred or a risk is developing, requiring proactive intervention.

[0058] If the comprehensive risk score is greater than or equal to the Level 3 warning threshold, a Level 3 warning is triggered, indicating that a serious fault has occurred and major safety risks such as thermal runaway are imminent, requiring immediate emergency response.

[0059] The model training process involves collecting a large amount of historical operational data, including data on normal operation, various progressive failures, and sudden failures. This data is labeled, for example, normal = 0, level 1 risk = 0.3, level 2 risk = 0.6, and level 3 risk = 0.9. The corresponding fused feature vector sequences are used as input to supervised training of the neural network until the model can accurately distinguish between different risk states.

[0060] Step S4: Based on the warning level, execute the corresponding linkage control strategy according to the hierarchical progression principle.

[0061] The linkage control strategy includes one or more combinations of local alarms, remote notifications, battery management system commands, circuit breaker actions, and fire protection system activation.

[0062] The principle behind the hierarchical and progressive execution of the corresponding linkage control strategy is as follows: The system has pre-set control strategies that are strictly tied to each warning level, following a progressive principle of alarm, flow restriction, isolation, and fire suppression.

[0063] If a Level 1 warning is triggered, a local alarm will be activated, controlling the audible and visual alarms inside the battery box or compartment to issue a warning in a low-frequency, intermittent mode, such as slow LED flashing and intermittent buzzer sounding. A pop-up message will be pushed to the monitoring center, or a warning SMS / email will be sent to relevant maintenance personnel. The report will include a risk score, key abnormal parameters, and the battery number. The energy storage system will continue to operate normally, but the background system will record the event and increase the monitoring frequency of the affected battery cell.

[0064] If a Level 2 warning is triggered, an enhanced alarm will be activated, escalating the local audible and visual alarms to a high-frequency continuous mode, such as flashing LEDs and a continuous buzzer. A command will also be sent to the Battery Management System (BMS) via a high-speed communication interface to immediately limit the charging / discharging current of the faulty battery cluster or related unit to reduce heat generation and prevent further deterioration. Simultaneously, the system may switch to a reduced-power operation mode. A report containing more detailed diagnostic information will also be sent to the monitoring center.

[0065] If a Level 3 warning is triggered, the highest-level alarm will be activated, controlling the local audible and visual alarms to operate at their strongest, such as LED flashing and a high-frequency continuous buzzer. A trip command will be immediately sent to the circuit breaker or contactor to disconnect the faulty battery unit from the DC bus and the power grid, achieving physical isolation. Simultaneously, an activation signal will be sent to the fire extinguishing system to suppress and extinguish the fire in the target battery box or compartment. At the same time, the highest-priority alarm will be immediately sent to the monitoring center, and pre-set emergency numbers will be automatically dialed.

[0066] In this step, a hierarchical and progressive linkage control strategy is designed. This strategy covers multiple levels, including local and remote control, electrical control and fire intervention, forming a hierarchical and progressive safety protection system. Control measures can be upgraded step by step according to the risk level, and the fire protection system can be activated quickly in extreme situations. This avoids operational interruptions caused by excessive control in low-risk situations, while enabling the rapid activation of high-intensity protective measures in high-risk situations, minimizing accident losses and improving the overall safety and reliability of the energy storage system.

[0067] The multi-level early warning linkage control method for energy storage systems of the present invention is applicable to various large-scale battery energy storage systems such as sodium-ion battery energy storage power stations. It realizes the time sequence alignment of multimodal data streams through a sliding time window, integrates basic statistical features, correlation features, and model-driven features, comprehensively identifies the coupling rules between various parameters, and, in conjunction with a pre-trained dynamic risk assessment model, accurately outputs a comprehensive risk score and early warning level. This effectively avoids missed reports caused by single data monitoring, accurately matches the safety protection requirements of different risk levels, and achieves early identification of safety hazards.

[0068] In a preferred embodiment of this application, the weighted combination of the basic statistical features, the correlation features, and the model-driven features includes: Step S201: Obtain the current health status and operating conditions of the batteries in the energy storage system; The SOH value of the current battery cell or battery cluster is obtained through capacity calibration, internal resistance identification algorithms or model-based SOH estimators that are periodically executed by the battery management system (BMS). In the current operating conditions, the ambient temperature is read directly from the ambient temperature sensor deployed in the battery compartment, and the charge / discharge rate is calculated based on the real-time current value and the battery's rated capacity.

[0069] Step S202: Based on the current health status of the battery, select an appropriate feature calculation subset from the pre-configured feature extraction strategy library; Based on the current health status of the battery, the battery health status is divided into multiple intervals, and an optimized feature subset is preset for each health status interval. The optimized feature subset is obtained by analyzing the data before the fault occurred in the corresponding interval in the historical fault case library.

[0070] By dividing the SOH range into several discrete intervals, including: Interval A (SOH≥90%): Excellent health status.

[0071] Interval B (80%≤SOH<90%): Slight attenuation.

[0072] Interval C (SOH < 80%): Significant decay.

[0073] Each interval is associated with a predefined subset of optimized features. The process of generating and configuring the optimized feature subset involves: establishing a historical fault case library, recording the battery's multimodal data stream and corresponding feature sequences for a period before each fault occurs. For each SOH interval, all fault cases within that interval are analyzed. Feature importance analysis algorithms, such as feature importance ranking using the XGBoost model or filtering methods like ReliefF, are used to identify the set of features with the highest discriminative power and the earliest abnormalities before the fault occurs. For example, in the excellent SOH interval (A), faults are often caused by external impacts or manufacturing defects; analysis may reveal that abrupt changes in current frequency and peak values ​​of gas-pressure cross-correlation are more valuable for early warning. In the significant SOH degradation interval (C), the battery itself ages, and faults are often related to abnormal heat generation and increased internal resistance; analysis may reveal that features related to thermal management, such as temperature rise gradients and electrothermal coupling characteristics, are more valuable for early warning. The top-N key features selected for each SOH interval are used to form the optimized feature subset for that interval and stored in the feature extraction strategy library.

[0074] When the energy storage system is running, it determines the range to which it belongs based on the real-time acquired SOH value and automatically loads the corresponding optimized feature subset from the strategy library.

[0075] Step S203: Based on the current operating conditions of the battery, dynamically adjust the fusion weights of each feature in the selected feature calculation subset.

[0076] Based on the current operating conditions of the battery, determine the dominant risk type most relevant to the current operating conditions; Based on the risk-dominant type, the weight ratio of the corresponding feature in the fused feature vector is dynamically adjusted; The operating conditions include at least ambient temperature and charge / discharge rate, and the risk-dominant types include thermal runaway risk-dominant, electrical overstress risk-dominant, or gas leakage risk-dominant.

[0077] A condition-risk type mapping table is established, with rules based on electrochemical knowledge and experimental data. For example, if the ambient temperature is greater than 45°C and the charge / discharge rate is greater than 0.8, the risk is determined to be primarily thermal runaway. If the charge / discharge rate is greater than 1.2, the risk is determined to be primarily electrical overstress. If the ambient temperature is normal but the baseline gas concentration continues to rise slowly, the risk is determined to be primarily gas leakage.

[0078] Based on real-time operating conditions, this table is consulted to determine the current risk dominance type. A weight configuration template is preset for each risk dominance type. This template defines the relative importance weights of each feature (from the subset selected in step S202) under the current risk type. For example: Thermal runaway risk-dominant template: significantly increase the weights of temperature rise gradient and electrothermal coupling characteristics, such as setting them to 0.4 and 0.3, and appropriately reduce the weight of voltage fluctuation variance, such as setting it to 0.1.

[0079] Electrical overstress risk-dominant template: significantly increase the weight of current surge frequency and voltage fluctuation variance, such as setting them to 0.35 and 0.35, and reduce the weight of gas-related features.

[0080] Gas leak risk dominant template: significantly increase the weight of gas-pressure cross-correlation peak value and gas-pressure correlation characteristics, such as setting them to 0.5 and 0.3.

[0081] Based on the determined risk-dominant type, the system calls the corresponding weight configuration template to weight the selected feature subset.

[0082] The pre-configured feature extraction strategy library includes at least: a first type of strategy focusing on battery performance and life degradation assessment, and a second type of strategy focusing on battery safety and sudden fault response.

[0083] This preferred embodiment upgrades the feature weighting combination process from a static, fixed mode to a dynamic, adaptive mode. Its core lies in intelligently selecting the most relevant features and adjusting their importance based on the battery's real-time internal health status (SOH) and external operating conditions. For aging batteries, the focus is on monitoring their thermal characteristics; for new batteries, greater attention is paid to the impact of electrical shocks, making the warning signals closer to the actual weak points of the battery, thus identifying potential faults specific to that battery state at an earlier stage. Simultaneously, the condition-based weight adjustment gives the system scenario-based intelligence. During high-rate operation, electrical safety is automatically prioritized; in high-temperature environments, thermal safety is given full attention. This dynamic mechanism ensures that the system allocates its limited attention resources to the most likely risks under any operating condition, greatly reducing false alarms and missed alarms.

[0084] This invention also provides a multi-level early warning and linkage control system for an energy storage system, which applies the aforementioned method and includes: The data acquisition unit is used to acquire multimodal real-time data streams at the battery cell / battery box level in the energy storage system, including voltage, current, temperature, specific gas concentration, and internal pressure of the box; The feature fusion unit is used to perform time-series alignment and feature-level fusion on the multimodal real-time data stream to generate a fused feature vector, including: A sliding time window is used to truncate and align the multimodal real-time data stream. Within each aligned window, the following features are extracted synchronously: The rise gradient of the temperature series, the fluctuation variance of the voltage series, and the number of abrupt changes in the current series are calculated as basic statistical features. The peak value of the cross-correlation function between the gas concentration series and the pressure series is calculated as the correlation feature; Based on the battery electrothermal model, temperature and current data are coupled and analyzed to generate electrothermal coupling features, and gas-pressure correlation features are generated based on the gas diffusion model and gas concentration and pressure data, which serve as model-driven features. The basic statistical features, the correlation features, and the model-driven features are weighted and combined to generate the fused feature vector. The risk warning output unit is used to input the fused feature vector into the pre-trained dynamic risk assessment model and output the battery's comprehensive risk score and corresponding warning level. The linkage control unit is used to execute a matching linkage control strategy according to the warning level and following the principle of hierarchical progression.

[0085] The functional explanation of each unit in this embodiment is the same as that of a multi-level early warning linkage control method for an energy storage system, and the technical effects are the same, so it will not be repeated here.

[0086] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0087] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0088] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0089] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention 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 or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the specification of the present invention.

Claims

1. A multi-stage early warning linkage control method for an energy storage system, characterized in that, Comprise: Step S1, obtaining the multi-modal real-time data stream of the battery monomer / battery box level in the energy storage system, including voltage, current, temperature, specific gas concentration and internal pressure of the box; Step S2, time alignment and feature level fusion are carried out on the multi-modal real-time data stream to generate a fusion feature vector, including: The multi-modal real-time data stream is intercepted and time aligned by using a sliding time window. In each aligned window, the following features are extracted synchronously: Calculate the rising gradient of the temperature sequence, the fluctuation variance of the voltage sequence and the mutation times of the current sequence as the basic statistical features; Calculate the cross-correlation function peak value of the gas concentration sequence and the pressure sequence as the correlation feature; Based on the battery electro-thermal model, the coupling analysis of temperature and current data is carried out to generate electro-thermal coupling features, and based on the gas diffusion model, the correlation analysis of gas concentration and pressure data is carried out to generate gas-pressure correlation features as model-driven features; The basic statistical features, the correlation features and the model-driven features are combined by weighting to generate the fusion feature vector; Step S3, input the fusion feature vector into the pre-trained dynamic risk assessment model to output the comprehensive risk score of the battery and the corresponding warning level; Step S4, according to the warning level, follow the principle of hierarchical progression to execute the matching linkage control strategy.

2. The multi-stage early warning linkage control method of an energy storage system according to claim 1, characterized in that, The weighting combination of the basic statistical features, the correlation features and the model-driven features comprises: Obtain the current health status and operating condition of the battery in the energy storage system; According to the current health status of the battery, select the adaptive feature calculation subset from the preconfigured feature extraction strategy library; According to the current operating condition of the battery, dynamically adjust the fusion weight of each feature in the selected feature calculation subset; Wherein, the preconfigured feature extraction strategy library at least includes: the first type of strategy focusing on battery performance and life degradation evaluation, and the second type of strategy focusing on battery safety and mutation fault response.

3. The multi-stage early warning linkage control method of an energy storage system according to claim 2, characterized in that, According to the current health status of the battery, select the adaptive feature calculation subset from the preconfigured feature extraction strategy library, comprising: According to the current health status of the battery, divide the health status of the battery into multiple intervals, and preset an optimized feature subset for each health status interval, wherein the optimized feature subset is obtained by analyzing the data before the fault occurs in the corresponding interval in the historical fault case library.

4. The multi-stage early warning linkage control method of an energy storage system according to claim 2, characterized in that, According to the current operating condition of the battery, dynamically adjust the fusion weight of each feature in the selected feature calculation subset, comprising: According to the current operating condition of the battery, determine the risk dominant type most related to the current condition; Based on the risk dominant type, dynamically adjust the weight proportion of the corresponding feature in the fusion feature vector; Wherein, the operating condition at least includes ambient temperature and charge-discharge rate, and the risk dominant type includes thermal runaway risk dominant, electrical over-stress risk dominant or gas leakage risk dominant.

5. The multi-stage early warning linkage control method of an energy storage system according to claim 1, characterized in that, The correlation analysis of gas concentration and cabin pressure data based on the gas diffusion model to generate gas-pressure correlation features comprises: Calculate the cross-correlation function of the gas concentration sequence and the pressure sequence in the sliding window to obtain the corresponding peak value; When the peak value exceeds a first preset threshold, triggering correlation analysis based on a gas diffusion model, calculating the model fitness between the gas concentration change and the pressure change as the gas-pressure correlation feature.

6. The multi-stage early warning linkage control method of an energy storage system according to claim 1, characterized in that, The coupling analysis of the temperature and current data based on the battery electro-thermal model to generate the electro-thermal coupling feature includes: Based on the current sequence and battery parameters in the sliding window, the theoretical temperature rise curve of the battery is calculated by using an electrochemical-thermal coupling model; Comparing the theoretical temperature rise curve with the measured temperature sequence in the same window, calculating the residual sum of squares, and generating the electro-thermal coupling feature representing the balance state of heat generation and dissipation.

7. The multi-stage early warning linkage control method of an energy storage system according to claim 1, characterized in that, The dynamic risk assessment model is a time series prediction model based on a long short-term memory network and an attention mechanism, which is trained by historical normal data and multi-level fault data, and is used to output a comprehensive risk score of the battery system according to the time evolution trend of the fusion feature vector, and determine the corresponding warning level based on the comprehensive risk score.

8. The multi-stage early warning linkage control method of an energy storage system according to claim 7, characterized in that, The fusion feature vector is input into a pre-trained dynamic risk assessment model to output a comprehensive risk score of the battery system and a corresponding warning level, which includes: The time series dependency pattern of the fusion feature vector sequence is extracted by a long short-term memory network to output a hidden state sequence; The hidden state sequence is analyzed by an attention mechanism to calculate the importance weight of different time step features for current risk assessment and identify key abnormal segments; The final output layer of the model generates a continuous comprehensive risk score and a multi-dimensional risk vector based on the weighted context vector, and the multi-dimensional risk vector represents the sub-scores of thermal runaway risk, electrical fault risk and gas leakage risk, respectively.

9. A multi-stage early warning linkage control system for an energy storage system, characterized in that, The method of any one of claims 1-8, comprising: a data acquisition unit for acquiring multi-modal real-time data streams of battery cells / battery box level in the energy storage system, including voltage, current, temperature, specific gas concentration and internal pressure of the box; a feature fusion unit for time series alignment and feature level fusion of the multi-modal real-time data streams to generate a fusion feature vector, including: using a sliding time window to intercept and time series align the multi-modal real-time data streams, and in each aligned window, the following features are extracted synchronously: calculating the rising gradient of the temperature sequence, the fluctuation variance of the voltage sequence and the mutation number of the current sequence as basic statistical features; calculating the cross-correlation function peak value of the gas concentration sequence and the pressure sequence as a correlation feature; based on the battery electro-thermal model, coupling analysis of temperature and current data to generate electro-thermal coupling features, and based on the gas diffusion model, correlation analysis of gas concentration and pressure data to generate gas-pressure correlation features as model-driven features; weighting and combining the basic statistical features, the correlation features and the model-driven features to generate the fusion feature vector; a risk warning output unit for inputting the fusion feature vector into a pre-trained dynamic risk assessment model to output a comprehensive risk score of the battery and a corresponding warning level; The linkage control unit is used for executing the matched linkage control strategy according to the early warning level and following the principle of hierarchical progression.

Images