Security detection method and system of energy storage system and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a safety detection method and system of an energy storage system and a storage medium. The method comprises the steps that voltage, current, temperature and internal resistance data of the energy storage system are collected to form a basic monitoring data set, and a preprocessed data sequence is generated through data preprocessing; mapping the preprocessed data sequence to a multi-dimensional feature space to obtain a target feature set; establishing a security detection model through the target feature set; generating a risk assessment report based on the security detection model; and analyzing system performance according to the evaluation report and outputting a maintenance scheme. According to the invention, intelligent analysis and accurate fault diagnosis of multi-parameter data of the energy storage system are realized. The intelligent monitoring and early warning of the safety state of the energy storage system are realized by establishing a multi-level analysis framework with the functions of feature extraction, intelligent diagnosis, fault early warning and the like and comprehensively evaluating the system state in combination with a deep learning algorithm.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a safety detection method, system and storage medium for an energy storage system. Background Art

[0002] With the rapid development of the new energy industry, energy storage systems are increasingly widely used in fields such as power peak shaving, renewable energy grid connection, and distributed generation. As a key device for energy storage and conversion, the safety and reliability of energy storage systems directly affect the stable operation of the entire energy system. At present, the safety detection of energy storage systems mainly relies on regular inspections and offline tests, and the detection methods include electrical parameter measurement, thermal imaging scanning, and insulation performance inspection, etc. At the same time, some advanced online monitoring systems have also been gradually put into use, and these systems can collect the operation data of energy storage devices in real time and perform basic status monitoring and alarming.

[0003] However, the existing safety detection technologies for energy storage systems have deficiencies in many aspects. First, the traditional manual inspection method has problems such as long detection cycle, low efficiency, and easy missed detection, making it difficult to discover potential safety hazards in the system in a timely manner. Second, although the online monitoring system can collect data in real time, due to the lack of effective data analysis and processing methods, its ability to identify fault characteristics under complex working conditions is limited. Third, the existing detection methods often analyze single parameters in isolation and cannot effectively handle the complex correlation relationships between parameters in the energy storage system. Especially when facing characteristics such as multi-parameter coupling and non-linear changes, it is difficult to accurately judge the actual safety state of the system. In addition, the existing technology has weak capabilities in fault warning and trend prediction and cannot provide effective decision-making support for the preventive maintenance of the system. Summary of the Invention

[0004] This application provides a safety detection method, system and storage medium for an energy storage system, which are used to solve the technical problem of intelligent analysis of multi-parameter data of the energy storage system and accurate fault diagnosis. By establishing a multi-level analysis framework including functions such as feature extraction, intelligent diagnosis, and fault warning, and comprehensively evaluating the system state in combination with deep learning algorithms, intelligent monitoring and warning of the safety state of the energy storage system are realized.

[0005] In a first aspect, the present application provides a safety detection method for an energy storage system. The safety detection method for the energy storage system includes: monitoring and analyzing the operating parameters of the energy storage system, establishing a monitoring parameter system based on the multi-parameter, non-linear, and time-varying characteristics of the energy storage system, collecting voltage, current, temperature, and internal resistance data for the monitoring parameter system to obtain a basic monitoring data set; establishing a data preprocessing model according to the basic monitoring data set, performing data cleaning and standardization processing on the basic monitoring data set, extracting time series features and statistical features through feature engineering methods to generate a preprocessed data sequence; constructing a feature space model based on the preprocessed data sequence, mapping the preprocessed data sequence to a multi-dimensional feature space according to the parameter type, screening key feature indicators through a feature evaluation mechanism to form a target feature set; establishing an intelligent diagnosis model based on the target feature set, performing a correlation analysis on the electrical parameters, thermal parameters, and state parameters in the target feature set, optimizing the model parameters through a cross-validation method to obtain a safety detection model; setting a fault diagnosis rule according to the safety detection model, inputting the real-time operating data of the energy storage system into the safety detection model, comparing with a preset safety threshold to generate a fault risk assessment report; establishing a health assessment system based on the fault risk assessment report, performing a trend analysis on the system performance in combination with historical operating data, and outputting a maintenance decision plan.

[0006] In a second aspect, the present application provides a safety detection system for an energy storage system. The safety detection system for the energy storage system includes:

[0007] An acquisition module, configured to monitor and analyze the operating parameters of the energy storage system, establish a monitoring parameter system based on the multi-parameter, non-linear, and time-varying characteristics of the energy storage system, collect voltage, current, temperature, and internal resistance data for the monitoring parameter system to obtain a basic monitoring data set;

[0008] An extraction module, configured to establish a data preprocessing model according to the basic monitoring data set, perform data cleaning and standardization processing on the basic monitoring data set, extract time series features and statistical features through feature engineering methods to generate a preprocessed data sequence;

[0009] A mapping module, configured to construct a feature space model based on the preprocessed data sequence, map the preprocessed data sequence to a multi-dimensional feature space according to the parameter type, screen key feature indicators through a feature evaluation mechanism to form a target feature set;

[0010] A verification module, configured to establish an intelligent diagnosis model based on the target feature set, perform a correlation analysis on the electrical parameters, thermal parameters, and state parameters in the target feature set, optimize the model parameters through a cross-validation method to obtain a safety detection model;

[0011] An input module, configured to set a fault diagnosis rule according to the safety detection model, input real-time operation data of the energy storage system into the safety detection model, compare with a preset safety threshold, and generate a fault risk assessment report;

[0012] An analysis module, configured to establish a health assessment system according to the fault risk assessment report, perform trend analysis on the system performance in combination with historical operation data, and output a maintenance decision plan.

[0013] A third aspect of the present application provides a computer-readable storage medium, in which instructions are stored, and when they run on a computer, the computer is enabled to execute the above-mentioned safety detection method for the energy storage system.

[0014] In the technical solution provided by the present application, by establishing a data acquisition, processing, analysis and decision-making link, the comprehensive monitoring and intelligent diagnosis of the safety state of the energy storage system are realized. In the data acquisition link, a monitoring parameter system is established based on the multi-parameter, non-linear and time-varying characteristics of the energy storage system, and the accurate acquisition of key parameters such as voltage, current, temperature and internal resistance is realized, solving the problems of incomplete parameter acquisition and insufficient sampling accuracy in traditional methods. In the data preprocessing stage, a data preprocessing model is established to clean and standardize the basic monitoring data set, and time series features and statistical features are extracted in combination with feature engineering methods, significantly improving the data quality and usability. In the construction of the feature space, a parameter type mapping and feature evaluation mechanism are adopted to realize data dimensionality reduction and extraction of key features, providing high-quality feature input for subsequent analysis. The present invention adopts a deep learning network structure including a feature input layer, a parameter mapping layer, a feature extraction layer, a fault diagnosis layer and an output layer. This artificial intelligence model design optimized specifically for the characteristics of the energy storage system realizes the accurate diagnosis of the complex working conditions of the energy storage system. The design of each layer of the model fully considers the professional characteristics of the energy storage system. For example, the feature extraction layer adopts a parallel neural network structure to extract time series features, statistical features and correlation features respectively, and the fault diagnosis layer designs a specific loss function based on the professional knowledge of the energy storage system, significantly improving the accuracy and reliability of fault identification. By setting dynamic fault diagnosis rules, combining real-time operation data and preset safety thresholds, a real-time risk assessment mechanism is established. Finally, by establishing a health assessment system and performing trend analysis on the system performance in combination with historical operation data, a closed-loop management from fault diagnosis to maintenance decision-making is realized, providing reliable technical support for the safe operation and preventive maintenance of the energy storage system. The overall solution effectively solves technical problems such as data diversity, fault complexity and diagnosis real-time in the safety detection of the energy storage system through a multi-level data processing and multi-level analysis architecture, realizing the intelligence and automation of the safety detection of the energy storage system. Description of the Drawings

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0016] Figure 1 Schematic diagram of an embodiment of the safety detection method for the energy storage system in the embodiments of the present application;

[0017] Figure 2 Schematic diagram of the network structure of the safety detection model in the embodiments of the present application

[0018] Figure 3 Schematic diagram of an embodiment of the safety detection system for the energy storage system in the embodiments of the present application. Specific implementation manners

[0019] The embodiments of the present application provide a safety detection method, system and storage medium for an energy storage system. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned accompanying drawings of the present application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data used can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any deformation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] For ease of understanding, the following describes the specific process of the embodiments of the present application. Please refer to Figure 1 An embodiment of the safety detection method for the energy storage system in the embodiments of the present application includes:

[0021] Step S101: Monitor and analyze the operating parameters of the energy storage system, establish a monitoring parameter system based on the multi-parameter, non-linear and time-varying characteristics of the energy storage system, and collect voltage, current, temperature and internal resistance data for the monitoring parameter system to obtain a basic monitoring data set;

[0022] Step S102: Establish a data preprocessing model according to the basic monitoring data set, perform data cleaning and standardization processing on the basic monitoring data set, and extract time series features and statistical features through feature engineering methods to generate a preprocessed data sequence;

[0023] Step S103: Construct a feature space model based on the preprocessed data sequence, map the preprocessed data sequence to a multi-dimensional feature space according to the parameter type, and screen key feature indicators through a feature evaluation mechanism to form a target feature set;

[0024] Step S104: Establish an intelligent diagnosis model based on the target feature set, perform a correlation analysis on the electrical parameters, thermal parameters, and state parameters in the target feature set, and optimize the model parameters through a cross-validation method to obtain a safety detection model;

[0025] Step S105: Set fault diagnosis rules according to the safety detection model, input the real-time operation data of the energy storage system into the safety detection model, compare with the preset safety threshold, and generate a fault risk assessment report;

[0026] Step S106: Establish a health assessment system based on the fault risk assessment report, perform a trend analysis on the system performance in combination with historical operation data, and output a maintenance decision plan.

[0027] It can be understood that the execution subject of this application can be the safety detection system of the energy storage system, or it can also be a terminal or a server. Specifically, it is not limited here. In this embodiment of the application, the server is used as the execution subject for illustration.

[0028] Specifically, when collecting the operation parameters of the energy storage system through the sensor network, considering the multi-parameter characteristics of the system, multi-dimensional data including voltage, current, temperature, and internal resistance are collected. Considering the non-linear characteristics of the energy storage system, the coupling relationship between parameters needs to be considered when collecting data, such as the mutual influence between voltage and temperature. For time-varying characteristics, the sampling frequency needs to be dynamically adjusted according to the parameter change rate. The sampling frequencies of voltage and current are usually in milliseconds, and the sampling frequency of temperature is in seconds. During the data collection process, phase calibration is particularly important. By precisely controlling the sampling timing, the time alignment of different parameter data is ensured. The collected raw data is preliminarily processed by the data aggregation node to form a basic monitoring data set.

[0029] In the data preprocessing stage, the basic monitoring data set is cleaned to eliminate outliers caused by sensor failures or communication interferences. For the spikes in voltage data, they are identified and corrected by calculating the change rate of adjacent data points. The missing values in temperature data are supplemented by interpolation from adjacent measurement points. The standardization process unifies parameters with different dimensions to a comparable scale. Voltage data is usually normalized to the interval [0,1], and temperature data is converted into relative change amounts. In terms of feature engineering, when extracting time series features, the change rate, fluctuation amplitude, and periodic features of parameters are calculated. Statistical features include descriptive indicators such as mean, standard deviation, and skewness, generating a preprocessed data sequence containing multi-dimensional features. The feature space construction stage focuses on solving the problems of data dimensionality reduction and feature selection. When mapping the preprocessed data sequence to a multi-dimensional feature space, the physical associations between parameters are considered, such as voltage-temperature coupling features and current-internal resistance correlation features. Feature evaluation adopts a screening method based on information gain, calculating the contribution degree of each feature to the system state judgment. By setting a contribution degree threshold, a target feature set is screened and formed. The construction of the intelligent diagnosis model is carried out based on the target feature set. The model includes a network structure with a feature input layer, a parameter mapping layer, a feature extraction layer, a fault diagnosis layer, and an output layer. The feature input layer receives electrical parameters, thermal parameters, and state parameters, and the parameter mapping layer performs space transformation and scale adjustment. The feature extraction layer contains parallel processing units that respectively extract time series features, statistical features, and correlation features. The fault diagnosis layer performs pattern recognition based on the extracted features, and the output layer generates an evaluation result. The model parameters are optimized through cross-validation to improve the diagnostic accuracy.

[0030] The formulation of fault diagnosis rules is based on the safety detection model. The real-time operation data of the energy storage system is input into the model. After feature extraction and pattern recognition, it is compared with the preset safety thresholds. The setting of the thresholds considers the physical limits of parameters and historical statistical features, such as voltage overlimit thresholds and temperature alarm thresholds. The comparison results form a fault risk assessment report, which contains information such as fault types, risk levels, and development trends. The establishment of the health assessment system integrates the fault risk assessment results and historical operation data. By analyzing the fault types, development trends, and impact degrees, a fault-performance mapping relationship is established. Combining with the performance degradation law in historical data, life prediction and maintenance planning are carried out. The trend analysis of system performance is based on the change rate and degradation trajectory of performance indicators, and an output maintenance decision plan containing the priority of maintenance items, implementation timing, and resource requirements is generated.

[0031] For example, during the operation of the battery pack of a certain energy storage power station, the voltage sensor collects the voltage fluctuations of individual batteries. Data preprocessing reveals that the fluctuations have periodic characteristics and are correlated with temperature changes. Feature space analysis shows that the time lag between voltage fluctuations and temperature changes is approximately 10 minutes, indicating that temperature changes are a potential cause of voltage fluctuations. The diagnostic model identifies the abnormality of the temperature control system based on this feature, and the risk assessment report points out that this abnormality may lead to an accelerated decline in battery performance. The health assessment system based on historical data analysis shows that if similar faults are not handled in a timely manner, the battery capacity will be reduced by 15% within 2-3 months. The maintenance decision plan recommends overhauling the temperature control system within one week and strengthening the temperature monitoring frequency.

[0032] In the embodiment of the present application, by establishing a data acquisition, processing, analysis, and decision-making link, the comprehensive monitoring and intelligent diagnosis of the safety state of the energy storage system are realized. In the data acquisition link, a monitoring parameter system is established based on the multi-parameter, non-linear, and time-varying characteristics of the energy storage system, realizing the accurate acquisition of key parameters such as voltage, current, temperature, and internal resistance, and solving the problems of incomplete parameter acquisition and insufficient sampling accuracy in traditional methods. In the data preprocessing stage, a data preprocessing model is established to clean and standardize the basic monitoring data set, and time series features and statistical features are extracted by combining feature engineering methods, significantly improving the data quality and usability. In the construction of the feature space, a parameter type mapping and feature evaluation mechanism are adopted to realize data dimensionality reduction and the extraction of key features, providing high-quality feature input for subsequent analysis. The present invention adopts a deep learning network structure including a feature input layer, a parameter mapping layer, a feature extraction layer, a fault diagnosis layer, and an output layer. This artificial intelligence model design optimized specifically for the characteristics of the energy storage system realizes the accurate diagnosis of the complex working conditions of the energy storage system. The design of each layer of the model fully considers the professional characteristics of the energy storage system. For example, the feature extraction layer adopts a parallel neural network structure to extract time series features, statistical features, and correlation features respectively, and the fault diagnosis layer designs a specific loss function based on the professional knowledge of the energy storage system, significantly improving the accuracy and reliability of fault identification. By setting dynamic fault diagnosis rules and combining real-time operation data and preset safety thresholds, a real-time risk assessment mechanism is established. Finally, by establishing a health assessment system and conducting trend analysis on the system performance in combination with historical operation data, a closed-loop management from fault diagnosis to maintenance decision-making is realized, providing reliable technical support for the safe operation and preventive maintenance of the energy storage system. The overall solution effectively solves technical problems such as data diversity, fault complexity, and diagnosis real-time in the safety detection of the energy storage system through a multi-level data processing and multi-level analysis architecture, realizing the intelligence and automation of the safety detection of the energy storage system.

[0033] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0034] (1) Obtain the basic operation data of the energy storage system, conduct cross-analysis on the voltage data and current data in the basic operation data, and output electrical characteristic parameters;

[0035] (2) Conduct correlation analysis on the electrical characteristic parameters and temperature data, and correct the electrical characteristic parameters according to the change trend of the temperature data to form a temperature correction coefficient;

[0036] (3) Conduct data fusion on the temperature correction coefficient and internal resistance data, determine the degree of mutual influence between parameters through data correlation analysis, and generate a parameter correlation matrix;

[0037] (4) Classify the parameter correlation matrix according to the multi-parameter, non-linear and time-varying characteristics of the energy storage system, conduct time-series analysis on the classified data, and establish a monitoring parameter system;

[0038] (5) Set data acquisition rules according to the monitoring parameter system, and conduct hierarchical sampling on the voltage, current, temperature and internal resistance data according to the acquisition rules to generate a hierarchical data sequence;

[0039] (6) Conduct data verification and outlier processing on the hierarchical data sequence, and correct the abnormal data points through data compensation methods to obtain a basic monitoring data set.

[0040] Specifically, starting from the acquisition of basic operation data, by collecting the basic operation data in the energy storage system, focus on the cross-analysis of voltage data and current data. The cross-analysis of voltage data and current data forms a set of electrical characteristic parameters reflecting the operation state of the energy storage system through methods such as resistivity calculation, power factor analysis and harmonic content evaluation. These parameters include impedance characteristics, power characteristics and power quality indicators.

[0041] When the electrical characteristic parameters are obtained, the influence of temperature on the performance of the energy storage system needs to be considered. The correlation analysis between temperature and electrical characteristic parameters adopts the following calculation formula:

[0042]

[0043] Among them, M tc represents the temperature correction coefficient, P i represents the value of the i-th electrical characteristic parameter, ΔT i represents the temperature deviation at the corresponding parameter point, α i represents the temperature sensitivity coefficient, β i represents the temperature attenuation factor, γ represents the temperature change rate weight, ω represents the temperature dynamic response coefficient, represents the temperature change rate, and n represents the number of electrical characteristic parameters.

[0044] The fusion of the temperature correction coefficient and the internal resistance data adopts the weighted average method to calculate the mutual influence degree between parameters. The temperature correction coefficient is applied to the correction of the internal resistance data, and then the influence relationship between parameters is established through correlation analysis to generate a parameter correlation matrix. Each element in the parameter correlation matrix represents the influence strength between different parameters. The data classification of the parameter correlation matrix follows the three basic characteristics of the energy storage system: multi-parameter, non-linear, and time-varying. The multi-parameter characteristic is reflected in the grouping process of different types of parameters, the non-linear characteristic is reflected by the mapping of the non-linear relationship between parameters, and the time-varying characteristic is reflected by the calculation of the parameter change rate. The classified data is subjected to time series analysis, including trend analysis, periodic analysis, and mutation point detection, thereby establishing a monitoring parameter system.

[0045] Based on the data acquisition rules set by the monitoring parameter system, a hierarchical sampling strategy is adopted. High-frequency sampling is used for parameters such as voltage and current that change rapidly, and low-frequency sampling is used for parameters such as temperature and internal resistance that change relatively slowly. The selection of the sampling frequency matches the time-varying characteristics of the parameters, forming a multi-level sampling sequence. The hierarchical data sequence collected is subjected to data verification, including numerical range check, change rate test, and data continuity verification. For the detected abnormal data points, data compensation methods are used for correction. The data compensation methods include linear interpolation, spline interpolation, or data correction based on a physical model to obtain a reliable basic monitoring data set.

[0046] Taking the battery pack of a certain energy storage power station as an example, during the data processing process, the voltage and current data of the battery pack are collected. Through cross-analysis, it is found that the phase difference between the voltage and the current will change with the temperature. This phenomenon is input into the temperature correction model, and the influence coefficient of temperature on electrical parameters is calculated. Combining with the battery internal resistance data, a correlation relationship model between parameters is established. The data classification shows that there is an obvious hysteresis effect in the influence of temperature change on battery performance. Accordingly, a differentiated data sampling strategy is formulated to increase the sampling frequency during the period of significant temperature change. Voltage fluctuations caused by sudden temperature changes are found through outlier detection, and more accurate basic data for performance evaluation are obtained after data compensation.

[0047] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0048] (1) Construct a data quality evaluation system for the basic monitoring data set, perform data phase analysis on voltage, current, temperature, and internal resistance data, identify the phase shift in the data acquisition process, and form a phase correction sequence;

[0049] (2) Calibrate the phase correction sequence with the basic monitoring data set, eliminate the influence caused by data acquisition delay through the phase compensation method, and output the time series calibration data;

[0050] (3) Perform wavelet decomposition on the noise components in the timing calibration data, extract the data features in different frequency bands, separate the high-frequency noise from the effective signals, and generate the signal decomposition results;

[0051] (4) Perform data reconstruction on the signal decomposition results, weight and combine the effective signals in different frequency bands according to the energy contribution degree, and construct the signal recombination data;

[0052] (5) Segment the signal recombination data according to the working states of the energy storage system, extract the characteristic parameters in each working state, including charging characteristics, discharging characteristics, and static characteristics, and generate the state characteristic sequence;

[0053] (6) Perform data fusion processing on the state characteristic sequence, combine the time series characteristics with the state characteristics, and extract the dynamic relationships between the data through feature correlation analysis to generate the preprocessed data sequence.

[0054] Specifically, perform data quality assessment on the basic monitoring data set, and focus on the phase shift problem in the data acquisition process. There are phase differences in the voltage, current, temperature, and internal resistance data during the acquisition process due to factors such as the sensor response time and signal transmission delay. The data phase analysis uses the correlation analysis method to calculate the time delay relationship between different parameters, and determines the phase shift amount through the parameter time series comparison to generate a phase correction sequence containing the phase deviation information of each parameter.

[0055] The data calibration process of the phase correction sequence uses the phase compensation technology to align the data in time according to the physical correlation relationship between the parameters. For the rapidly changing voltage and current data, the compensation method is dynamically adjusted based on the parameter change rate; for the slowly changing temperature and internal resistance data, the linear interpolation method is used for time synchronization. Eliminate the influence of data acquisition delay through phase compensation to form the timing calibration data. The timing calibration data often contains noise components of various frequencies and needs to be processed by the wavelet decomposition method. Wavelet decomposition decomposes the original signal into sub-signals in different frequency bands. The high-frequency components mainly contain measurement noise and interference signals, while the low-frequency components contain the basic change trends of the parameters. By setting the frequency threshold, distinguish the high-frequency noise from the effective signals to generate a multi-level signal decomposition result.

[0056] During the data reconstruction process, an energy contribution analysis method is adopted for signals in different frequency bands. Calculate the energy proportion of signals in each frequency band. The high-energy region usually corresponds to the main change characteristics of the parameters, while the low-energy region reflects the secondary change information. Based on the energy contribution, set the weight coefficient, and perform weighted combination on the effective signals in different frequency bands to construct the signal reconstruction data that reflects the true change characteristics of the parameters. The working state analysis of the signal reconstruction data is based on the operating characteristics of the energy storage system. According to the change rules of voltage and current, the data is divided into three working states: charging, discharging, and standing still. The charging characteristics include parameters such as the charging voltage curve and the charging current change rate; the discharging characteristics include indicators such as the depth of discharge and the discharging rate; the standing still characteristics reflect the self-discharge characteristics of the system and the temperature change rule. Extract the characteristic parameters for each working state to form a state characteristic sequence.

[0057] The fusion processing of the state characteristic sequence and the time series characteristics adopts a multi-dimensional feature analysis method. The time series characteristics reflect the change trend and periodic characteristics of the parameters, while the state characteristics describe the performance indicators under different working states. Through feature correlation analysis, establish the dynamic correlation relationship between parameters, including the sequence, influence intensity, and duration of parameter changes, and generate a preprocessing data sequence containing complete feature information.

[0058] Taking the lithium battery pack of a certain energy storage power station as an example, it is found that there is an obvious acquisition delay in the voltage and temperature data during the data processing. Determine through phase analysis that the temperature data lags behind the voltage data by about 2 seconds, and generate a phase correction value accordingly. After phase compensation, the corresponding relationship between temperature change and voltage change is more accurate. Perform wavelet decomposition on the calibrated data to separate the high-frequency interference caused by grid fluctuations. In data reconstruction, the energy contribution during the charging process mainly comes from the medium-frequency band signals, corresponding to the main charging characteristics of the battery. The working state analysis shows that during the fast charging stage, there is a significant positive correlation between the temperature rise rate and the charging current. Through feature fusion analysis, a dynamic correlation model among the charging current, temperature change, and battery performance is established.

[0059] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0060] (1) Construct a dynamic feature evaluation system for the preprocessing data sequence, extract the mutation points in the time series data based on the working condition switching characteristics of the energy storage system, and generate a working condition conversion sequence;

[0061] (2) Divide the data segments for the working condition conversion sequence, extract the working condition characteristics of the voltage data, current data, temperature data, and internal resistance data in each data segment, and construct a working condition feature space;

[0062] (3) Analyze the energy distribution of the data in the operating condition feature space, calculate the energy proportion of each parameter under different operating conditions, and output the energy feature matrix;

[0063] (4) Perform feature mapping based on the energy feature matrix, project the parameters in the high-energy region onto the multi-dimensional feature space, and form a key feature mapping diagram;

[0064] (5) Conduct data coupling degree analysis on the key feature mapping diagram, generate a parameter coupling matrix by calculating the phase relationship and amplitude relationship between parameters;

[0065] (6) Conduct correlation analysis on the coupling degree values in the parameter coupling matrix and the operating condition conversion sequence, extract the characteristic parameters with significant operating condition recognition ability, and form the target feature set.

[0066] Specifically, when performing dynamic feature evaluation on the preprocessed data sequence, focus on the operating condition switching characteristics of the energy storage system, mainly including key nodes such as charge-discharge switching, load change, and working mode conversion. By analyzing the parameter change rate in the time-series data, when the parameter change rate exceeds the set threshold, it is marked as an operating condition switching point, thereby generating an operating condition conversion sequence reflecting the conversion of the system operation state. Based on the generated operating condition conversion sequence, segment the data. For each operating condition segment, extract the characteristics of voltage data, current data, temperature data, and internal resistance data respectively. The operating condition feature extraction includes statistical features (mean, standard deviation), time-domain features (rise time, stabilization time), and frequency-domain features (main frequency component, harmonic content), and these features together constitute the operating condition feature space.

[0067] When analyzing the energy distribution of the data in the operating condition feature space, the following calculation method is adopted:

[0068]

[0069] Among them, E dist represents the energy distribution eigenvalue, λ k represents the weight coefficient of the k-th operating condition, W i represents the weight of the i-th parameter, D ki represents the normalized value of the i-th parameter under the k-th operating condition, R j represents the j-th reference value, H j represents the corresponding adjustment factor, ξ k represents the operating condition attenuation coefficient, m represents the number of operating conditions, n represents the number of parameters, and p represents the number of reference values.

[0070] The data coupling degree analysis of the key feature mapping diagram adopts the following formula:

[0071]

[0072] Among them, C coup represents the coupling degree coefficient, represents the cross-correlation coefficient of parameters x and y, V x and V y represent the amplitudes of the parameters, S x and S y represent the standard deviations of the parameters, θ x and θ y represent the phase angles of the parameters, ψ xy represents the phase difference sensitivity, and q represents the total number of parameters.

[0073] After the coupling degree analysis is completed, the parameter coupling matrix is correlated with the working condition conversion sequence. By calculating the change characteristics of the parameter coupling degree under different working conditions, the characteristic parameters sensitive to the change of working conditions are identified to form a target characteristic set.

[0074] Taking the battery pack of a certain energy storage power station as an example, during the charging and discharging process, the charging switching point and the discharging switching point are identified through dynamic characteristic evaluation. Near the switching point, the change rates of voltage and current increase significantly, and the temperature change shows a certain lag. Feature extraction is performed on the data segments before and after the switching, and it is found that there is an obvious correlation between the voltage rise characteristic and the temperature change during the charging process, while the internal resistance change is closely related to the current magnitude during the discharging process. Through energy distribution analysis, the importance of each parameter under different working conditions is determined. For example, during the fast charging working condition, the energy proportions of current and temperature are relatively high. Feature mapping and coupling degree analysis show that the voltage-temperature coupling degree increases significantly at the end of charging, while the current-internal resistance coupling degree is stronger at the beginning of discharging.

[0075] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0076] (1) Group the target characteristic set according to the functional attributes, calculate the data correlation intensity among electrical parameters, thermal parameters, and state parameters, and generate a parameter correlation degree matrix;

[0077] (2) Introduce time-delay analysis for the parameter correlation degree matrix, calculate the response delay characteristics between different parameters, and establish a parameter time-delay mapping table;

[0078] (3) Combine the time-delay data in the parameter time-delay mapping table with the correlation degree data, construct a parameter influence link, and quantify the causal relationship in the link to output a parameter causal relationship diagram;

[0079] (4) Perform hierarchical analysis on the parameter causal relationship diagram, identify the direct influence path and indirect influence path between parameters, and construct a hierarchical influence network;

[0080] (5) Establish parameter interaction rules through a hierarchical influence network, match the parameter change patterns in historical fault data with the hierarchical influence network, and generate a fault feature library;

[0081] (6) Conduct data verification based on the fault feature library, verify and correct the parameter interaction rules through a cross-validation method to obtain a security detection model, where the security detection model includes a feature input layer, a parameter mapping layer, a feature extraction layer, a fault diagnosis layer, and an output layer.

[0082] Specifically, divide the data into three categories according to functional attributes: electrical parameters (including voltage and current), thermal parameters (including temperature), and state parameters (including internal resistance), and calculate the data correlation strength between parameters. The calculation of the data correlation strength uses the cross-correlation analysis method to evaluate the change relationship between different parameters and generate a parameter correlation matrix.

[0083] The time-delay analysis adopts the following calculation method:

[0084]

[0085] where, T lag represents the parameter time-delay eigenvalue, κ ij represents the correlation coefficient between parameters i and j, t i and t j represent the response times of parameters i and j respectively, η ij represents the response decay factor, ζ represents the time-delay sensitivity coefficient, ρ ij represents the ratio of the parameter change rates, and n represents the number of parameters.

[0086] The quantification of the causal relationship of the parameter influence link adopts the following formula:

[0087]

[0088] where, C rel represents the causal relationship strength, μ k represents the weight of the kth link node, represents the change amount of the cause parameter, represents the change amount of the result parameter, μ k represents the link sensitivity, σ k represents the time decay coefficient, Δt k represents the node time interval, and m represents the number of link nodes.

[0089] Based on the above mathematical model, the time-delay data and correlation data are combined to construct a parameter influence link that reflects the influence relationship between parameters. By quantifying the strength of the causal relationship between different parameters, a parameter causal relationship diagram is formed to clearly show the influence transmission path between parameters. During the hierarchical analysis process, the direct influence path and indirect influence path between parameters are identified. The direct influence path represents the direct action relationship between parameters, such as the temperature change directly caused by the current change; the indirect influence path reflects the influence between parameters through intermediate parameters, such as the voltage change affecting the temperature through the current change. These influence paths together constitute a hierarchical influence network.

[0090] During the matching process of the hierarchical influence network and historical fault data, the parameter change rules in the historical fault data are extracted, corresponding to the influence paths in the network, establishing fault-parameter association rules, and forming a fault feature library containing multiple fault modes.

[0091] The cross-validation method is used to verify and correct the parameter interaction rules, and a security detection model with a five-layer structure is constructed: the feature input layer receives the original parameter data; the parameter mapping layer performs data standardization and mapping transformation; the feature extraction layer contains multiple parallel processing units, which respectively extract time series features, statistical features and correlation features; the fault diagnosis layer identifies the fault mode based on the features; the output layer generates the diagnostic result. Taking the power battery pack of a certain energy storage power station as an example, when a battery thermal runaway fault occurs, it is observed that the battery temperature rises abnormally. Through time-delay analysis, it is found that the voltage fluctuates, the current increases, and the internal resistance decreases before the temperature anomaly. The parameter influence link shows that the increase in current leads to a decrease in internal resistance, which in turn accelerates the temperature rise, forming a positive feedback effect. This positive feedback loop identified in the hierarchical influence network is a typical feature of the thermal runaway fault. Based on this feature, the security detection model can identify potential thermal runaway risks early when the current-internal resistance-temperature link is abnormal.

[0092] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0093] (1) Extract key monitoring point combinations from the feature input layer of the security detection model, input voltage, current, temperature and internal resistance data into the feature input layer, and calculate the change rate and phase relationship of the data through the parameter mapping layer to generate a dynamic early warning matrix;

[0094] (2) Input the dynamic early warning matrix into the parameter mapping layer for spatial transformation, identify the inflection point position of the parameter change trend through the data distribution characteristics, and divide the boundaries for different risk levels to form a fault classification standard;

[0095] (3) Input the fault classification standard into the feature extraction layer, and respectively extract time series features, statistical features and correlation features through the parallel feature processing unit, construct the response link between parameters, and output the fault propagation path diagram;

[0096] (4) Perform feature matching at the fault diagnosis layer for the fault propagation path diagram, calculate the similarity between the real-time operation data of the energy storage system and the historical fault modes, and generate the fault source identification result;

[0097] (5) Conduct risk quantification analysis on the fault source identification result at the fault diagnosis layer, and calculate based on the difference between the fault development trend and the system safety boundary to output the risk level quantification index;

[0098] (6) Conduct multi-dimensional comprehensive analysis on the risk level quantification index at the output layer, and combine the fault type discrimination result, development trend prediction and impact degree assessment to generate the fault risk assessment report.

[0099] Specifically, as Figure 2 shown, it is a schematic diagram of the network structure of the safety detection model in the embodiment of the present application. This structure includes five main levels: the feature input layer receives the original data of voltage, current, temperature and internal resistance; the parameter mapping layer performs spatial transformation and data standardization processing; the feature extraction layer includes three parallel processing units of time series features, statistical features and correlation features, which are respectively responsible for feature extraction in different dimensions; the fault diagnosis layer executes feature matching and risk quantification analysis; the output layer generates the risk assessment report.

[0100] The feature input layer receives the collected voltage, current, temperature and internal resistance data and performs preliminary processing through the parameter mapping layer. In this process, calculate the change rate of each parameter, including the voltage change rate dV / dt, the current change rate dI / dt, the temperature change rate dT / dt and the internal resistance change rate dR / dt. At the same time, analyze the phase relationship between the parameters to determine the sequence and time difference of parameter changes, and form a dynamic warning matrix reflecting the dynamic change characteristics of the parameters. In the parameter mapping layer, perform spatial transformation processing on the dynamic warning matrix. This process identifies the inflection point positions in the parameter change trend by analyzing the distribution characteristics of the data, including statistical quantities such as mean, variance, and kurtosis. The inflection point positions usually represent significant changes in the system state. Based on these inflection points, divide the risk level boundaries. The division process considers the amplitude, rate and duration of parameter changes to form a fault classification standard including multiple risk levels.

[0101] After the fault classification standard is input into the feature extraction layer, multi-dimensional analysis is carried out through parallel feature processing units. The time-series feature processing unit is responsible for extracting the time-domain features of parameters, such as change trends, periodicity, and mutation features; the statistical feature processing unit calculates the statistical features of parameters, including distribution features and correlation indicators; the correlation feature processing unit analyzes the mutual influence relationship between parameters. These features jointly construct the response link between parameters, describe the propagation path of faults in the system, and output the fault propagation path diagram. In the fault diagnosis layer, feature matching analysis is performed on the fault propagation path diagram. The real-time operation data of the energy storage system is compared with the typical fault patterns recorded in the historical fault database, and the similarity index is calculated. The similarity calculation considers the time-series features, amplitude features, and correlation features of parameter changes, and determines the most matching fault type through comprehensive scoring, generating an identification result including the fault source location and fault type.

[0102] The fault diagnosis layer is also responsible for quantitatively analyzing the risk of the fault source identification result. By calculating the difference between the fault development trend and the preset safety boundary of the system, the severity and development speed of the fault are evaluated. This process comprehensively considers the degree of parameter overrun, change rate, and cumulative impact, and outputs a quantitative index reflecting the fault risk level. In the output layer, multi-dimensional comprehensive analysis is carried out on the risk level quantitative index. Combining the fault type discrimination result, the typical features and influence range of the fault are analyzed; based on the development trend prediction, the evolution direction and speed of the fault are evaluated; through the impact degree evaluation, the impact of the fault on the overall performance of the system is determined. These analysis results are integrated to form a detailed fault risk assessment report.

[0103] Taking the lithium iron phosphate battery pack of a certain energy storage power station as an example, the working process of the safety detection model is as follows: When the feature input layer receives the monitoring data of the battery pack, it is found that the temperature of a single battery rises abnormally. The parameter mapping layer analysis shows that the temperature change rate exceeds the normal range and forms an abnormal phase relationship with the voltage and current parameters. The feature extraction layer discovers a significant correlation between the temperature anomaly and the reduction of internal resistance and the increase of current through parallel processing. In the fault diagnosis layer, this parameter change feature highly matches the typical features of battery thermal runaway. The risk quantitative analysis shows that the temperature rise rate has approached the safety boundary, and there is a risk of thermal diffusion. The evaluation report points out that this is a typical early sign of thermal runaway and gives specific disposal suggestions.

[0104] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0105] (1) Perform spatio-temporal decoupling analysis on the fault type, development trend, and impact degree data in the fault risk assessment report, associate the development characteristics of different types of faults with the performance degradation law, and generate a fault-performance mapping matrix;

[0106] (2) Analyze the health status based on the fault - performance mapping matrix. By calculating the degradation rate of each parameter and the performance decay curve, construct a sequence of performance degradation indices.

[0107] (3) Conduct a correlation analysis between the sequence of performance degradation indices and historical operation data, extract the performance change characteristics of the energy storage system under different working conditions, and output a working condition - performance correlation map.

[0108] (4) Conduct a life characteristic analysis based on the working condition - performance correlation map. By calculating the corresponding relationship between the degradation trajectory of key performance parameters and the remaining life, form a life prediction curve.

[0109] (5) Establish an evaluation criterion for maintenance timing for the life prediction curve. Combining the system operation constraint conditions and maintenance cost factors, generate a maintenance priority ranking.

[0110] (6) Conduct a comprehensive analysis of the maintenance priority ranking and the performance degradation trend, overall plan the urgency of maintenance items and resource allocation, and output a maintenance decision plan.

[0111] Specifically, the spatio - temporal decoupling analysis analyzes the data of fault types, development trends, and impact degrees in the time dimension and space dimension respectively. The time - dimension analysis focuses on the sequential characteristics of fault development, and the space - dimension analysis focuses on the propagation characteristics of faults within the system. By establishing a mapping relationship between the development characteristics of different types of faults and the corresponding performance degradation laws, a mapping matrix reflecting the fault - performance correspondence is generated.

[0112] When analyzing the health status of the fault - performance mapping matrix, the following calculation method is used:

[0113]

[0114] where D idx represents the performance degradation index, χ i represents the weight coefficient of the i - th performance parameter, F i represents the fault eigenvalue, ω i represents the degradation sensitivity, υ i represents the time decay factor, P i represents the current performance value, P 0 represents the initial performance value, t represents the operation time, and n represents the number of performance parameters. Through this formula, a sequence of performance degradation indices reflecting the system health status is calculated. The correlation analysis between the sequence of performance degradation indices and historical operation data focuses on the performance change characteristics under different working conditions. By extracting the change rules of performance parameters under each working condition, establishing the corresponding relationship between working condition conditions and performance degradation, a working condition - performance correlation map is formed. This map shows the influence degree and characteristics of different working conditions on the system performance.

[0115] The life characteristic analysis is carried out based on the operating condition - performance correlation map. By analyzing the deterioration trajectories of key performance parameters, the health state of the system is evaluated and the remaining life is calculated. The life prediction takes into account the performance degradation rate, operating condition influencing factors and historical data statistical rules, and draws a prediction curve reflecting the remaining life of the system. The maintenance timing evaluation criterion is established based on the life prediction curve. The evaluation criterion comprehensively considers the system operation constraint conditions (such as operation load requirements, environmental restrictions, etc.) and maintenance cost factors (including maintenance costs, downtime losses, etc.). Through comprehensive evaluation of multiple factors, different maintenance items are prioritized to form a maintenance execution order.

[0116] The maintenance priority ranking is comprehensively analyzed with the performance deterioration trend. By evaluating the urgency of maintenance items and the required resources, the maintenance plan is reasonably arranged, and finally a maintenance decision plan including specific implementation plans is output.

[0117] Taking the battery pack maintenance of a certain energy storage power station as an example: Through spatio - temporal decoupling analysis, it is found that the capacity attenuation of some single cells in the battery pack accelerates, and this attenuation is more significant under high - temperature operating conditions. The health state analysis shows that the performance deterioration index of the affected batteries continues to rise, and the attenuation rate exceeds the normal level. Correlation analysis finds that high - temperature operating conditions and charge - discharge depth jointly affect battery performance degradation. The life characteristic analysis predicts that if the current usage mode is maintained, some batteries will reach the replacement threshold in the short term. Considering the replacement cost and performance requirements comprehensively, the maintenance decision suggests replacing the aging batteries in batches and adjusting the charge - discharge strategy to delay the attenuation rate of other batteries.

[0118] In a specific embodiment, the process of performing the spatio - temporal decoupling analysis step on the data of fault types, development trends and influence degrees in the fault risk assessment report may specifically include the following steps:

[0119] (1) Analyze the parameter data in the fault risk assessment report in the time dimension, associate the fault types with the time - series characteristics, extract the fault occurrence time, development cycle and evolution rate, and form a fault time - series characteristic table;

[0120] (2) Decompose the fault time - series characteristic table in the space dimension, quantitatively describe the propagation paths and influence scopes of the fault development trends among different system components, and generate a fault space distribution map;

[0121] (3) Conduct cross - analysis of the fault space distribution map and the influence degree data. By calculating the influence intensity and attenuation law of the fault at different spatial positions, output a fault influence weight matrix;

[0122] (4) Conduct performance correlation analysis on the fault influence weight matrix, quantitatively calculate the influence degree of the fault on each performance index of the energy storage system, and construct a performance degradation factor sequence;

[0123] (5) Match the performance degradation factor sequence with historical failure cases, and output a failure-performance correspondence table by identifying the performance degradation law under similar failure modes.

[0124] (6) Perform data fusion on the failure-performance correspondence table, comprehensively evaluate the performance change characteristics caused by different types of failures, and generate a failure-performance mapping matrix.

[0125] Specifically, by extracting the time-series data from the failure risk assessment report, a timeline analysis is established for each type of failure. The time-series analysis focuses on the occurrence time of the failure (determined by the first occurrence time of parameter anomalies), the development cycle (the time interval from the appearance of failure symptoms to the complete manifestation of the failure), and the evolution rate (the change speed of the parameter deviation from the normal value). These time-series characteristics are sorted and recorded in the failure time-series feature table. The spatial dimension decomposition process focuses on the propagation characteristics of the failure among different components of the energy storage system. Map the data in the failure time-series feature table to space, and track the process of the failure spreading from the occurrence location to the surrounding area. By analyzing the physical connection relationship and energy transfer path among different components, determine the propagation direction and diffusion speed of the failure. The quantitative description includes spatial characteristics such as propagation distance, diffusion area, and influence depth, and generates a failure spatial distribution map reflecting the spatial distribution characteristics of the failure.

[0126] The cross-analysis of the failure spatial distribution map and the impact degree data is carried out from multiple dimensions. Calculate the impact intensity of the failure at each position in space, considering the attenuation law of the failure energy during the propagation process. The calculation of the impact intensity is based on factors such as parameter deviation value, component distance, and propagation impedance, while the attenuation law reflects the weakening trend of the failure impact with the increase of distance. These analysis results are sorted into a failure impact weight matrix to quantitatively describe the impact degree of the failure on components at different positions. In the performance correlation analysis stage, a correspondence relationship is established between the failure impact weight matrix and the performance indicators of the energy storage system. Analyze the impact degree of the failure on core performance indicators such as capacity, efficiency, and life respectively, and conduct quantitative calculations. By establishing the mapping relationship between the failure impact and performance degradation, construct a performance degradation factor sequence reflecting the degradation degree of each performance indicator. This sequence records the degradation rate and trend characteristics of different performance indicators.

[0127] The matching process between the performance degradation factor sequence and historical failure cases adopts a pattern recognition method. By comparing the performance degradation characteristics caused by the current failure with the degradation patterns recorded in historical cases, similar failure types and development laws are identified. The matching process considers multiple feature dimensions such as degradation rate, degradation pattern, and impact degree, and outputs a failure-performance correspondence table containing detailed correspondence relationships. The data fusion process systematically integrates the information in the failure-performance correspondence table. For different types of failures, the performance change characteristics caused by them are analyzed, including direct and indirect impacts. By comprehensively evaluating the impact degree of failures on various performance indicators, a failure-performance mapping relationship is established, and a failure-performance mapping matrix is generated.

[0128] Taking the battery pack failure analysis of a certain energy storage power station as an example, when a temperature anomaly in the battery pack is detected, it is found through time series analysis that the temperature increase first appears in a certain single battery and then spreads to the surrounding area. Spatial analysis shows that the failure starts from this single battery and spreads to adjacent batteries along the heat conduction path, and the influence range shows a circular diffusion characteristic. Cross-analysis shows that the batteries closer to the failure source have a faster temperature rise, while the influence on the distant batteries weakens with the increase in distance. Performance analysis finds that the temperature anomaly leads to a decrease in battery capacity, and the capacity degradation rate is positively correlated with the duration and peak value of the temperature anomaly. Comparing with historical cases, it is found that this temperature anomaly-capacity degradation mode highly coincides with the typical thermal runaway failure. The failure-performance mapping matrix clearly shows how the temperature anomaly affects various performance indicators of the entire battery pack through heat diffusion.

[0129] The safety detection method of the energy storage system in the embodiment of the present application has been described above. Next, the safety detection system of the energy storage system in the embodiment of the present application will be described. Please refer to Figure 3 , an embodiment of the safety detection system of the energy storage system in the embodiment of the present application includes:

[0130] An acquisition module, configured to monitor and analyze the operating parameters of the energy storage system, establish a monitoring parameter system based on the multi-parameter, non-linear, and time-varying characteristics of the energy storage system, collect voltage, current, temperature, and internal resistance data for the monitoring parameter system, and obtain a basic monitoring data set;

[0131] An extraction module, configured to establish a data preprocessing model according to the basic monitoring data set, perform data cleaning and standardization processing on the basic monitoring data set, extract time series features and statistical features through feature engineering methods, and generate a preprocessed data sequence;

[0132] A mapping module, configured to construct a feature space model according to the preprocessed data sequence, map the preprocessed data sequence to a multi-dimensional feature space according to the parameter type, and screen key feature indicators through a feature evaluation mechanism to form a target feature set;

[0133] A verification module, which is used to establish an intelligent diagnosis model based on the target feature set, perform correlation analysis on electrical parameters, thermal parameters and state parameters in the target feature set, optimize model parameters through a cross-validation method, and obtain a safety detection model;

[0134] An input module, which is used to set fault diagnosis rules according to the safety detection model, input real-time operation data of the energy storage system into the safety detection model, compare with a preset safety threshold, and generate a fault risk assessment report;

[0135] An analysis module, which is used to establish a health assessment system based on the fault risk assessment report, perform trend analysis on system performance in combination with historical operation data, and output a maintenance decision plan.

[0136] Through the collaborative cooperation of the above-mentioned various components, by establishing a data acquisition, processing, analysis and decision-making link, the comprehensive monitoring and intelligent diagnosis of the safety state of the energy storage system are realized. In the data acquisition link, a monitoring parameter system is established based on the multi-parameter, non-linear and time-varying characteristics of the energy storage system, and the accurate acquisition of key parameters such as voltage, current, temperature and internal resistance is realized, solving the problems of incomplete parameter acquisition and insufficient sampling accuracy in traditional methods. In the data preprocessing stage, a data preprocessing model is established to clean and standardize the basic monitoring data set, and time series features and statistical features are extracted in combination with feature engineering methods, significantly improving the data quality and usability. In terms of feature space construction, a parameter type mapping and feature evaluation mechanism are adopted to realize data dimensionality reduction and extraction of key features, providing high-quality feature input for subsequent analysis. The present invention adopts a deep learning network structure including a feature input layer, a parameter mapping layer, a feature extraction layer, a fault diagnosis layer and an output layer. This artificial intelligence model design optimized specifically for the characteristics of the energy storage system realizes the accurate diagnosis of the complex working conditions of the energy storage system. The design of each layer of the model fully considers the professional characteristics of the energy storage system. For example, the feature extraction layer adopts a parallel neural network structure to extract time series features, statistical features and correlation features respectively, and the fault diagnosis layer designs a specific loss function based on the professional knowledge of the energy storage system, significantly improving the accuracy and reliability of fault identification. By setting dynamic fault diagnosis rules, combining real-time operation data and preset safety thresholds, a real-time risk assessment mechanism is established. Finally, by establishing a health assessment system and performing trend analysis on system performance in combination with historical operation data, a closed-loop management from fault diagnosis to maintenance decision-making is realized, providing reliable technical support for the safe operation and preventive maintenance of the energy storage system. The overall solution effectively solves technical problems such as data diversity, fault complexity and diagnosis real-time in the safety detection of the energy storage system through a multi-level data processing and multi-level analysis architecture, realizing the intelligence and automation of the safety detection of the energy storage system.

[0137] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions run on a computer, the computer is caused to execute the steps of the safety detection method of the energy storage system.

[0138] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0139] If the integrated unit is implemented in the form of 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 application, 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. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0140] As described above, the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A safety detection method for an energy storage system, characterized in that: The safety detection method of the energy storage system comprises: Monitor and analyze the operating parameters of the energy storage system, establish a monitoring parameter system based on the multi-parameter, nonlinear and time-varying nature of the energy storage system, collect voltage, current, temperature and internal resistance data for the monitoring parameter system, and obtain a basic monitoring data set; Establishing a data preprocessing model based on the basic monitoring data set, performing data cleaning and standardization on the basic monitoring data set, extracting time series features and statistical features through feature engineering methods, and generating a preprocessed data sequence; Constructing a feature space model based on the preprocessed data sequence, mapping the preprocessed data sequence to a multidimensional feature space according to parameter types, screening key feature indicators through a feature evaluation mechanism, and forming a target feature set; An intelligent diagnosis model is established based on the target feature set, a correlation analysis is performed on the electrical parameters, thermal parameters and state parameters in the target feature set, and the model parameters are optimized by a cross-validation method to obtain a safety detection model; Setting fault diagnosis rules according to the safety detection model, inputting real-time operation data of the energy storage system into the safety detection model, comparing with the preset safety threshold, and generating a fault risk assessment report; A health assessment system is established based on the fault risk assessment report, and a trend analysis of system performance is performed in combination with historical operating data to output a maintenance decision plan.

2. The safety detection method for energy storage system according to claim 1, characterized in that: The operating parameters of the energy storage system are monitored and analyzed, and a monitoring parameter system is established based on the multi-parameter, nonlinear and time-varying nature of the energy storage system. Voltage, current, temperature and internal resistance data are collected for the monitoring parameter system to obtain a basic monitoring data set, including: Obtaining basic operation data of the energy storage system, cross-analyzing voltage data and current data in the basic operation data, and outputting electrical characteristic parameters; Correlation analysis is performed on the electrical characteristic parameters and the temperature data, and the electrical characteristic parameters are corrected according to the change trend of the temperature data to form a temperature correction coefficient; Performing data fusion on the temperature correction coefficient and internal resistance data, determining the mutual influence degree between parameters through data association analysis, and generating a parameter association matrix; Classify the parameter correlation matrix according to the multi-parameter, nonlinear and time-varying nature of the energy storage system, perform time series analysis on the classified data, and establish a monitoring parameter system; According to the monitoring parameter system, data collection rules are set, and voltage, current, temperature and internal resistance data are graded and sampled according to the collection rules to generate a graded data sequence; The hierarchical data sequence is subjected to data verification and outlier processing, and the abnormal data points are corrected by a data compensation method to obtain a basic monitoring data set.

3. The safety detection method for energy storage system according to claim 1, characterized in that: The data preprocessing model is established according to the basic monitoring data set, data cleaning and standardization are performed on the basic monitoring data set, time series features and statistical features are extracted by feature engineering methods, and a preprocessed data sequence is generated, including: A data quality assessment system is constructed for the basic monitoring data set, and data phase analysis is performed on the voltage, current, temperature and internal resistance data to identify the phase offset during the data acquisition process and form a phase correction sequence; Calibrate the phase correction sequence with the basic monitoring data set, eliminate the influence of data acquisition delay by a phase compensation method, and output timing calibration data; Performing wavelet decomposition on the noise components in the timing calibration data, extracting data features of different frequency bands, separating high-frequency noise from effective signals, and generating signal decomposition results; Reconstructing data based on the signal decomposition result, weighting and combining effective signals in different frequency bands according to their energy contribution, and constructing signal reconstruction data; Segmenting the signal reorganization data according to the working state of the energy storage system, extracting characteristic parameters under each working state, including charging characteristics, discharging characteristics and static characteristics, and generating a state characteristic sequence; The state feature sequence is subjected to data fusion processing, the time series features are combined with the state features, the dynamic relationship between the data is extracted through feature association analysis, and a preprocessed data sequence is generated.

4. The method for safety detection of energy storage system according to claim 1, characterized in that: The feature space model is constructed based on the preprocessed data sequence, the preprocessed data sequence is mapped to a multidimensional feature space according to the parameter type, and key feature indicators are screened through a feature evaluation mechanism to form a target feature set, including: Constructing a dynamic feature evaluation system for the preprocessed data sequence, extracting mutation points in the time series data based on the operating condition switching characteristics of the energy storage system, and generating an operating condition conversion sequence; Divide the data into segments according to the working condition conversion sequence, extract working condition features from the voltage data, current data, temperature data and internal resistance data in each segment of data, and construct a working condition feature space; Performing energy distribution analysis on the data in the working condition feature space, calculating the energy proportion of each parameter under different working conditions, and outputting an energy feature matrix; Perform feature mapping based on the energy feature matrix, project the parameters of the high energy area into a multi-dimensional feature space, and form a key feature mapping diagram; Performing data coupling analysis on the key feature map, and generating a parameter coupling matrix by calculating the phase relationship and amplitude relationship between parameters; The coupling degree values ​​in the parameter coupling matrix are correlated with the operating condition conversion sequence to extract characteristic parameters with significant operating condition recognition capabilities to form a target feature set.

5. The safety detection method for energy storage system according to claim 1, characterized in that: The intelligent diagnosis model is established based on the target feature set, the electrical parameters, thermal parameters and state parameters in the target feature set are analyzed for correlation, the model parameters are optimized by a cross-validation method, and a safety detection model is obtained, including: The target feature set is grouped according to functional attributes, the data correlation strength between electrical parameters, thermal parameters and state parameters is calculated, and a parameter correlation matrix is ​​generated; Introducing time-delay analysis for the parameter correlation matrix, calculating the response delay characteristics between different parameters, and establishing a parameter time-delay mapping table; Combining the time delay data in the parameter time delay mapping table with the correlation data, constructing a parameter influence link, quantifying the causal relationship in the link, and outputting a parameter causal relationship graph; Performing hierarchical analysis on the parameter causal relationship diagram, identifying direct and indirect impact paths between parameters, and constructing a hierarchical impact network; Establishing parameter interaction rules through the hierarchical influence network, matching the parameter change rules in the historical fault data with the hierarchical influence network, and generating a fault feature library; Data verification is performed based on the fault feature library, and the parameter interaction rules are verified and corrected through a cross-validation method to obtain a safety detection model, wherein the safety detection model includes a feature input layer, a parameter mapping layer, a feature extraction layer, a fault diagnosis layer and an output layer.

6. The method for safety detection of energy storage system according to claim 5, characterized in that: The method of setting a fault diagnosis rule according to the safety detection model, inputting the real-time operation data of the energy storage system into the safety detection model, comparing the data with the preset safety threshold, and generating a fault risk assessment report includes: Extracting key monitoring point combinations from the feature input layer in the safety detection model, inputting voltage, current, temperature and internal resistance data into the feature input layer, calculating the rate of change and phase relationship of the data through the parameter mapping layer, and generating a dynamic warning matrix; The dynamic warning matrix is ​​input into the parameter mapping layer for spatial transformation, the inflection point position of the parameter change trend is identified through data distribution characteristics, the boundaries of different risk levels are divided, and a fault classification standard is formed; The fault classification standard is input into the feature extraction layer, and the time series features, statistical features and correlation features are respectively extracted by the parallel feature processing unit, a response link between parameters is constructed, and a fault propagation path diagram is output; Perform feature matching on the fault diagnosis layer for the fault propagation path diagram, calculate similarity between the real-time operation data of the energy storage system and the historical fault mode, and generate a fault source identification result; Perform risk quantification analysis on the fault source identification result at the fault diagnosis layer, calculate the difference between the fault development trend and the system safety boundary, and output a risk level quantitative index; The risk level quantitative indicators are subjected to a multi-dimensional comprehensive analysis at the output layer, and a fault risk assessment report is generated by combining the fault type identification results, development trend prediction and impact assessment.

7. The method for safety detection of energy storage system according to claim 1, characterized in that: The health assessment system is established based on the fault risk assessment report, and the system performance trend is analyzed in combination with the historical operation data to output a maintenance decision plan, including: Performing spatiotemporal decoupling analysis on the fault types, development trends and impact degree data in the fault risk assessment report, correlating the development characteristics of different types of faults with performance degradation laws, and generating a fault-performance mapping matrix; Performing health status analysis on the fault-performance mapping matrix, and constructing a performance degradation index sequence by calculating the degradation rate and performance decay curve of each parameter; Correlation analysis is performed on the performance degradation index sequence and historical operation data, the performance change characteristics of the energy storage system under different working conditions are extracted, and a working condition-performance correlation map is output; Based on the working condition-performance correlation map, life characteristic analysis is performed, and a life prediction curve is formed by calculating the corresponding relationship between the degradation trajectory of key performance parameters and the remaining life; Establishing maintenance timing evaluation criteria for the life prediction curve, and generating maintenance priority ranking by combining system operation constraints and maintenance cost factors; The maintenance priority ranking and performance degradation trend are comprehensively analyzed, the urgency and resource allocation of maintenance projects are comprehensively planned, and a maintenance decision plan is output.

8. The method for safety detection of energy storage system according to claim 7, characterized in that: The time-space decoupling analysis of the fault type, development trend and impact degree data in the fault risk assessment report is performed to associate the development characteristics of different types of faults with the performance degradation law to generate a fault-performance mapping matrix, including: Performing time dimension analysis on the parameter data in the fault risk assessment report, associating the fault type with the timing characteristics, extracting the fault occurrence time, development cycle and evolution rate, and forming a fault timing characteristics table; Decomposing the fault time series feature table in spatial dimension, quantitatively describing the propagation path and impact range of the fault development trend between different system components, and generating a fault spatial distribution map; Cross-analyze the fault spatial distribution map and the impact degree data, and output the fault impact weight matrix by calculating the impact intensity and attenuation law of the fault at different spatial positions; Performing performance correlation analysis on the fault impact weight matrix, quantifying the impact of the fault on various performance indicators of the energy storage system, and constructing a performance degradation factor sequence; Perform feature matching on the performance degradation factor sequence and historical fault cases, and output a fault-performance correspondence table by identifying the performance degradation rules under similar fault modes; Data fusion is performed on the fault-performance correspondence table, and the performance change characteristics caused by different types of faults are comprehensively evaluated to generate a fault-performance mapping matrix.

9. A safety detection system for an energy storage system, used to implement the safety detection method for an energy storage system according to any one of claims 1 to 8, characterized in that: The safety detection system of the energy storage system includes: The acquisition module is used to monitor and analyze the operating parameters of the energy storage system, establish a monitoring parameter system based on the multi-parameter, nonlinear and time-varying nature of the energy storage system, and collect voltage, current, temperature and internal resistance data for the monitoring parameter system to obtain a basic monitoring data set; An extraction module is used to establish a data preprocessing model according to the basic monitoring data set, perform data cleaning and standardization on the basic monitoring data set, extract time series features and statistical features through feature engineering methods, and generate a preprocessed data sequence; A mapping module is used to construct a feature space model based on the preprocessed data sequence, map the preprocessed data sequence to a multidimensional feature space according to parameter type, screen key feature indicators through a feature evaluation mechanism, and form a target feature set; A verification module, used to establish an intelligent diagnosis model based on the target feature set, perform correlation analysis on the electrical parameters, thermal parameters and state parameters in the target feature set, optimize the model parameters by a cross-validation method, and obtain a safety detection model; An input module, used to set fault diagnosis rules according to the safety detection model, input real-time operation data of the energy storage system into the safety detection model, compare with a preset safety threshold, and generate a fault risk assessment report; The analysis module is used to establish a health assessment system based on the fault risk assessment report, conduct trend analysis on system performance in combination with historical operation data, and output a maintenance decision plan.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the safety detection method for the energy storage system according to any one of claims 1 to 8 is implemented.

Images