An active security method for energy storage power stations

Abstract

This invention relates to the field of safety monitoring and proactive prevention and control technology for energy storage power stations, and discloses a proactive security method for energy storage power stations. This method collects data on five categories of risk factors: electrical, thermal, gas, mechanical, and insulation. After preprocessing and feature extraction, key feature indicators are obtained. Based on the risk evolution mechanism, multi-factor correlation analysis is performed to generate a risk situation index. Then, combined with the key feature triggering results, a four-level early warning level is determined according to the principle of choosing the highest level, and graded response and closed-loop optimization are implemented. This scheme can improve the early identification and proactive prevention and control capabilities of energy storage power stations.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring and active prevention and control technology for energy storage power stations, and in particular to an active security method for energy storage power stations. Background Technology

[0002] As an important regulating resource in the new power system, the safe and stable operation of energy storage power stations is directly related to the reliability of the power grid and the security of energy supply. With the large-scale and clustered application of electrochemical energy storage power stations, risks such as battery thermal runaway, insulation degradation, electrolyte leakage, accumulation of combustible gas, and equipment linkage failure exhibit multi-factor coupling and staged evolution characteristics. Traditional security methods that rely solely on single threshold alarms are no longer sufficient to meet the requirements of early identification and proactive prevention.

[0003] Chinese patent application CN116520165A, entitled "A method, system, device and storage medium for early warning of battery energy storage power station", discloses a technical approach for collecting multi-source operating data of battery energy storage power station, performing data analysis and outputting risk warnings. This approach can use operating data such as voltage, current, temperature and insulation resistance to identify safety risks of energy storage power station and trigger corresponding early warning or disposal measures according to the risk level.

[0004] However, the existing technologies mentioned above mainly focus on routine operational data and risk level assessment. In the early stages of accidents such as thermal runaway in energy storage power plants, there are sequential evolutionary and mutually corroborating relationships among risk factors such as electricity, heat, gas, force, and insulation. If only some parameters are judged independently, it can easily lead to incomplete risk perception dimensions, mismatch between early warning levels and accident evolution stages, and a lack of closed-loop optimization in post-early warning response strategies. Therefore, a technical solution is needed that can achieve full-element perception, multi-element correlation analysis, four-level accurate early warning, and proactive graded response on existing energy storage power plant infrastructure. Summary of the Invention

[0005] Technical issues

[0006] This invention addresses the problems in existing early warning technologies for energy storage power stations, such as incomplete risk perception elements, insufficient correlation of multi-source data, low matching degree between early warning levels and risk evolution stages, and lack of closed-loop optimization for proactive response. It proposes a proactive security method for energy storage power stations to improve their early risk identification and proactive prevention capabilities.

[0007] To address the aforementioned technical problems, in a first aspect, this invention provides an active security method for energy storage power stations, comprising: collecting data on five types of risk factors in the energy storage power station—electrical, thermal, gas, mechanical, and insulation—to form comprehensive sensing data; filtering and denoising the comprehensive sensing data, removing outliers, and extracting features to obtain key feature indicators for each type of risk factor; performing multi-factor correlation analysis on the key feature indicators based on risk evolution mechanisms to identify combined anomalies or sequential anomalies between at least two types of risk factors and generating a risk situation index; determining four warning levels based on the risk situation index and the triggering results of the key feature indicators, following the principle of choosing the highest level; executing corresponding active graded response strategies according to the four warning levels, and optimizing the warning judgment threshold and correlation rules based on the response effect evaluation results.

[0008] Furthermore, the five types of risk factor data include electrical characteristic parameters, thermal characteristic parameters, gas characteristic parameters, mechanical characteristic parameters, and insulation characteristic parameters. The electrical characteristic parameters include the rate of change of internal resistance and voltage inconsistency of battery cells or modules. The thermal characteristic parameters include the rate of temperature rise and temperature difference distribution. The gas characteristic parameters include the concentration of combustible gases and information on electrolyte decomposition products. The mechanical characteristic parameters include the internal pressure of the battery pack and the strain of the casing. The insulation characteristic parameters include the insulation resistance of the battery pack or power supply circuit.

[0009] Furthermore, the system performs unified time calibration and data integrity checks on various risk factor data; conducts horizontal difference analysis on similar parameters within the same battery pack or battery cluster; performs trend change analysis on continuous sampling data from the same measuring point or the same equipment unit; and outputs single-factor anomaly markers, rate of change markers, and trend markers for multi-factor correlation analysis.

[0010] Furthermore, the multi-factor correlation analysis includes: establishing a correlation rule base between risk factors; identifying a combined anomaly or a sequence anomaly when at least two of the following characteristics occur simultaneously or in a preset order within a preset correlation time window: increased internal resistance, accelerated temperature rise, detection of characteristic gases, increased internal pressure, strain change, or decreased insulation resistance; and generating the risk situation index based on the type, quantity, severity, and duration of the combined anomaly or sequence anomaly.

[0011] Furthermore, the four-level early warning system includes a level four warning, a level three warning, a level two warning, and a level one warning. The level four warning corresponds to the nascent stage of potential hazards, the level three warning corresponds to the risk development stage, the level two warning corresponds to the danger approaching stage, and the level one warning corresponds to the accident critical stage. When any key characteristic indicator reaches the trigger condition of a level higher than the level corresponding to the current risk situation index, the warning level will be upgraded to the higher level corresponding to that key characteristic indicator.

[0012] Furthermore, the proactive graded response strategy includes: outputting attention prompts and increasing inspection frequency during a Level 4 warning; outputting audible and visual warnings and notifying maintenance personnel to conduct on-site investigations during a Level 3 warning; notifying power plant duty personnel and the superior monitoring center and activating emergency plan preparations during a Level 2 warning; and coordinating proactive emergency response with the fire extinguishing system, fault circuit power outage system, or smoke exhaust ventilation system during a Level 1 warning.

[0013] Furthermore, the optimized early warning judgment threshold and association rules include: recording the early warning level, triggering characteristics, response measures, response timeliness, and risk resolution results for each early warning event; and adjusting the calculation weight of the risk situation index, the triggering conditions of key feature indicators, or the combined anomaly judgment conditions in the association rule base according to the response effect evaluation results.

[0014] The advantages of this invention compared to existing technologies are as follows:

[0015] This invention employs a comprehensive sensing approach encompassing five risk factors: electrical, thermal, gas, mechanical, and insulation. This approach reduces the blind spots caused by relying solely on conventional parameters such as temperature and voltage, enabling early signs of battery thermal runaway, insulation degradation, electrolyte decomposition, and abnormal internal pressure to be identified within the same data framework.

[0016] Through multi-factor correlation analysis, this invention transforms single-point parameter alarms into combined or sequential anomaly identification. It can utilize the mutual corroboration relationships between signals such as increased internal resistance, accelerated temperature rise, detection of characteristic gases, and decreased insulation resistance to improve the matching degree between early warning judgment and actual risk evolution process, and reduce the impact of false alarms or missed alarms of single parameters on security judgment.

[0017] By corresponding the risk evolution stages with four levels of early warning, and by adopting the principle of choosing the higher level over the lower level by combining the risk situation index with key characteristic indicators, this invention can output differentiated early warning results at different risk stages, providing clear technical basis for strengthening inspections, on-site investigations, emergency preparedness, and automatic linkage response.

[0018] Through response effect evaluation and continuous model optimization, this invention can iteratively adjust thresholds, weights, and association rules based on actual early warning and handling results, enabling the active security system of energy storage power stations to have continuous calibration capabilities, and is suitable for security upgrades of newly built or existing energy storage power stations. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating the full-element risk perception system architecture of the active security system for energy storage power stations according to the present invention.

[0020] Figure 2 This is a flowchart of the four-level early warning level determination process of the present invention;

[0021] Figure 3 This is a schematic diagram of the active security closed-loop mechanism of the present invention;

[0022] Figure 4 This is a schematic diagram of the multi-factor correlation analysis model of the present invention. Detailed Implementation

[0023] The following description, in conjunction with the accompanying drawings, further illustrates the method embodiments of the present invention. It should be understood that the following embodiments are used to illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention.

[0024] Example 1:

[0025] like Figure 1 As shown, this embodiment takes an electrochemical energy storage power station as the object and relies on the existing sensor network, data acquisition terminal, monitoring terminal, edge computing gateway, or station-end server of the energy storage power station to execute an active security method. When executing this method, the station-end server or monitoring terminal can be configured with a full-element perception module, a feature processing module, a fusion analysis module, an early warning judgment module, an active response module, and a closed-loop optimization module. These modules, as the execution units of this method, respectively undertake the functions of data acquisition, feature processing, fusion analysis, early warning judgment, graded response, and feedback optimization, and transmit full-element perception data, key feature indicators, early warning levels, and response effect data through a unified data interface. The station-end server, monitoring terminal, edge computing device, or scheduling and monitoring device may include a processor and a memory. When the computer program stored in the memory is executed by the processor, it completes the following method steps. This computer program may also be stored in a computer-readable storage medium.

[0026] Step S101: Identify the full-element sensing targets and establish a data collection plan. A survey is conducted on the battery types, system architecture, and operating conditions of the energy storage power station. Battery types include lithium iron phosphate batteries, ternary lithium batteries, or other electrochemical energy storage batteries. The system architecture includes modules, battery clusters, battery stacks, and battery compartments. Based on existing monitoring of temperature, smoke, voltage, and current, a data collection list is determined according to five risk elements: electrical, thermal, gas, mechanical, and insulation. This forms a full-element sensing plan covering early warning signs of risks such as battery thermal runaway, insulation degradation, electrolyte leakage, gas generation and swelling, and decreased insulation in the power supply circuit.

[0027] Step S102: Collect data on five types of risk factors and form comprehensive sensing data. Electrical characteristic parameters include the rate of change of internal resistance of individual battery cells or modules and voltage inconsistency. The rate of change of internal resistance can be calculated by the BMS system in conjunction with voltage and current data, with a monitoring accuracy configured to be no greater than 0.1 mΩ. Voltage inconsistency is characterized by the voltage range of individual cells within the same battery pack. Thermal characteristic parameters include the rate of temperature rise and temperature difference distribution. The rate of temperature rise is obtained from the temperature change per unit time at the same measuring point, and the temperature difference distribution is obtained from the temperature difference between different measuring points within the same battery pack. Gas characteristic parameters include the concentration of combustible gases such as hydrogen, carbon monoxide, and methane, as well as information on electrolyte decomposition products. The resolution of the combustible gas concentration acquisition module is configured to be at the 0.1 ppm level. Electrolyte decomposition products include carbonate vapors such as EMC and DMC. Mechanical characteristic parameters include the internal pressure of the battery pack and the strain of the casing. The internal pressure monitoring range is configured to be 0 to 200 kPa with an accuracy of ±0.5% FS, and the strain monitoring range is configured to be 0 to 5000 με. Insulation characteristic parameters include the insulation resistance of the battery pack or power supply circuit, with the monitoring range of the insulation resistance acquisition link covering 1MΩ to 100MΩ. The above five types of data are aggregated by a unified data acquisition terminal for synchronous acquisition, real-time transmission, and unified storage.

[0028] Step S103 involves preprocessing and feature extraction of the full-element sensing data. The feature processing module first performs unified time calibration and sampling time alignment on data from different acquisition modules, then performs filtering and noise reduction, outlier removal, and data integrity checks. For similar parameters within the same battery pack or battery cluster, lateral difference analysis is performed to identify anomalies such as individual cell voltage range, temperature difference distribution, internal resistance differences, or insulation resistance drift. For continuous sampling data from the same measuring point or the same equipment unit, trend change analysis is performed to identify trends such as accelerated temperature rise, continuous increase in internal pressure, continuous change in strain, or change in combustible gas concentration gradient. After the above processing, single-element anomaly markers, rate of change markers, and trend markers are output for multi-element correlation analysis, and key feature indicators of various risk elements are obtained.

[0029] Step S104 involves performing multi-factor correlation analysis and generating a risk situation index. For example... Figure 4 As shown, the station-side fusion analysis engine establishes a rule base and anomaly pattern base for the association between risk elements based on the risk evolution mechanism. When at least two of the following characteristics—increased internal resistance, accelerated temperature rise, detection of characteristic gases, increased internal pressure, strain change, and decreased insulation resistance—occur simultaneously within a preset association time window, or occur sequentially in a preset order, they are identified as combined anomalies or sequential anomalies. The fusion analysis engine then generates a risk situation index based on the type, quantity, severity, and duration of the combined or sequential anomalies. Let this risk situation index be... Its value satisfies ;in, The higher the value, the higher the risk of the current energy storage power station. During project implementation, the weighting of the risk situation index is calibrated based on battery type, operating conditions, historical operating data, and simulation verification results. At the same time, the fusion analysis engine conducts trend analysis based on historical and real-time data to make short-term predictions on the development trend of the risk situation index. The prediction time window is set to be no less than 30 minutes to make forward-looking judgments before the risk situation index reaches the warning threshold.

[0030] Step S105: Determine the Level 4 early warning level based on the risk situation index and the triggering results of key characteristic indicators. For example... Figure 2 As shown, the early warning judgment module first determines the basic early warning level based on the range of the risk situation index: Level 4 warning is the attention level, corresponding to the nascent stage of hidden dangers, with the risk situation index in the range of 0 to 25. Its characteristics are a slight abnormality in a single risk element, such as an excessive voltage inconsistency in a single unit or a slight acceleration in the local temperature rise rate; Level 3 warning is the warning level, corresponding to the risk development stage, with the risk situation index in the range of 26 to 50. Its characteristics are two or more obviously abnormal risk elements, such as the simultaneous occurrence of increased internal resistance and accelerated temperature rise, or the simultaneous occurrence of the detection of characteristic gases and a decrease in insulation resistance; Level 2 warning is the emergency level, corresponding to the danger approaching stage, with the risk situation index in the range of 51 to 75. Its characteristics are multiple severely abnormal and mutually corroborating risk elements, such as a significant increase in internal resistance, a sharp acceleration in the temperature rise rate, an excessive concentration of combustible gas, and an abnormal increase in internal pressure; Level 1 warning is the critical level, corresponding to the critical stage of an accident, with the risk situation index in the range of 76 to 100. Its characteristics are a continuous and rapid increase in the risk situation index, indicating that an accident is about to occur or has already occurred. When any key characteristic indicator reaches an independent trigger condition higher than the basic warning level, the warning determination module will raise the warning level to the higher level corresponding to that key characteristic indicator according to the principle of choosing the higher level.

[0031] Step S106: Implement the proactive tiered response strategy based on the four-level warning level. For example... Figure 3 As shown, during a Level 4 warning, the monitoring terminal outputs a notification and increases the frequency of inspections; during a Level 3 warning, the system outputs an audible and visual warning, notifying maintenance personnel to conduct on-site investigations, and can initiate remote diagnostics and enhanced monitoring; during a Level 2 warning, the system issues an emergency alarm, notifying the power plant duty personnel and the superior monitoring center, and initiating emergency response plan preparation and multi-party coordination; during a Level 1 warning, the system issues the highest-level emergency alarm and coordinates with the fire extinguishing system, fault circuit power outage system, or smoke exhaust ventilation system to perform proactive emergency response.

[0032] Step S107: Perform closed-loop optimization based on the response effect evaluation results. The closed-loop optimization module records the warning level, triggering characteristics, response measures, response timeliness, handling effect, and risk mitigation result for each warning event, and evaluates the accuracy, timeliness, and effectiveness of the warning based on the recorded results. If the evaluation results show that the warning lead time for a certain type of combined anomaly is insufficient, there are many false alarms, or the response measures are inadequate, the calculation weight of the risk situation index, the triggering conditions of key characteristic indicators, the combined anomaly judgment conditions in the association rule base, or the corresponding response strategy are adjusted. Through the above processing, the energy storage power station forms a closed-loop proactive security method of perception, analysis, warning, response, evaluation, and optimization.

[0033] Example 2:

[0034] This embodiment, based on Embodiment 1, further illustrates the deployment process of the active security method for energy storage power stations in newly built or existing energy storage power stations. This deployment process still falls under the category of method implementation, focusing on the configuration, debugging, and data transfer of each step at the engineering site.

[0035] Step S201 involves conducting a sensing requirements analysis and solution design. For the power station to be upgraded, obtain information on battery type, module / cluster / stall / cabin hierarchical structure, operating conditions, existing BMS functions, existing sensor layout, and monitoring communication conditions. Based on battery thermal runaway mechanisms, insulation aging mechanisms, and industry safety standards, determine the sampling points, sampling cycles, data interfaces, and acquisition module configuration plans for five categories of risk factors: electrical, thermal, gaseous, mechanical, and insulation. For data that existing equipment can provide, prioritize access via software interfaces; for data that existing equipment cannot provide but represents key risk precursors, supplementary access will be provided using a small number of acquisition modules.

[0036] Step S202: Configure and debug the sensing layer acquisition links. Based on the existing sensor network and data acquisition terminal, connect a 0.1ppm combustible gas concentration acquisition module and an electrolyte decomposition product detection module; connect an insulation resistance acquisition link covering 1MΩ to 100MΩ; connect a battery pack internal pressure sensor from 0 to 200kPa and a shell strain sensor from 0 to 5000με; and supplement the BMS system with internal resistance change rate calculation and voltage inconsistency analysis functions. After all acquisition links are connected, uniformly output time-stamped data records to the data acquisition terminal, enabling the five types of risk factors to enter the subsequent processing flow under the same time reference.

[0037] Step S203: Deploy the multi-source data fusion analysis engine. Deploy the fusion analysis engine software on monitoring terminals, station servers, or edge computing gateways. The fusion analysis engine includes data preprocessing, feature extraction, multi-factor correlation analysis, risk situation assessment, trend prediction, early warning determination, proactive response, and closed-loop optimization functions. The all-factor perception module sends five types of risk factor data to the feature processing module, which outputs key feature indicators and labeling results. The fusion analysis module calls the association rule base to identify combined anomalies or sequence anomalies and generates a risk situation index. The early warning determination module outputs four levels of early warning. The proactive response module sends corresponding control or notification commands to monitoring terminals, audible and visual alarms, duty notification systems, superior monitoring centers, fire extinguishing systems, fault circuit power outage systems, or smoke exhaust ventilation systems. The closed-loop optimization module feeds back the response results to the fusion analysis module and the early warning determination module.

[0038] Step S204: Set and calibrate the early warning judgment thresholds. Based on national standards, industry specifications, battery manufacturer technical parameters, and historical power plant operation data, set risk situation index ranges corresponding to Level 4, Level 3, Level 2, and Level 1 early warnings, and set independent trigger conditions for each key characteristic indicator. The setting is based not only on whether a single parameter exceeds its limit, but also on whether multiple abnormal factors occur simultaneously, whether they occur in a preset order, whether they persist, and whether they corroborate each other. When the basic early warning level and the independent trigger conditions are inconsistent, the higher level judgment rule is applied.

[0039] Step S205 involves system integration and simulation verification. After completing the hardware and software deployment, the completeness, real-time performance, and accuracy of the collected full-element sensing data are verified first, followed by verification of the correctness of the correlation analysis by the fusion analysis engine and the rationality of the situation assessment. During the integration process, combined abnormal scenarios of increased internal resistance, accelerated temperature rise, and detection of characteristic gases can be constructed, as well as sequential abnormal scenarios of decreased insulation resistance after the detection of characteristic gases, to verify the results of the four-level early warning level determination and the execution effect of the proactive response strategy. Based on the simulation verification results, the risk situation index weights, key characteristic indicator triggering conditions, and correlation rule base determination conditions are finely calibrated.

[0040] Step S206: After successful joint testing and verification, the method is put into operation. During operation, this method relies on the existing energy storage power station infrastructure and mainly upgrades software algorithms to improve active security capabilities. It does not require reconstruction of the existing energy storage system or large-scale addition of sensor hardware. For existing power stations, a smooth upgrade can be achieved by expanding the acquisition module, accessing existing BMS data, and deploying a station-side fusion analysis engine.

[0041] Example 3:

[0042] This embodiment, based on Embodiments 1 and 2, further illustrates the implementation process of the active security method for energy storage power stations during the trial operation, formal commissioning, and routine operation and maintenance phases. This implementation process ensures that the early warning judgment threshold, association rules, and response strategies continuously converge based on actual operational feedback.

[0043] Step S301: Initiate system trial operation and continuously collect operational data. The trial operation period is set to be no less than 30 days. During the trial operation, continuously collect data on five types of risk factors, key characteristic indicators, combined anomaly or sequence anomaly identification results, risk situation index, early warning level, and response records. For each early warning event that occurs during the trial operation, retain its occurrence time, involved battery pack or battery cluster, triggering characteristics, early warning level, output method, and handling process to form the data foundation for subsequent optimization.

[0044] Step S302: Evaluate the effectiveness of early warning and response. Based on the trial operation records, conduct statistical analysis on the accuracy, timeliness, response effectiveness, handling results, and risk mitigation status of early warnings. If an early warning event is handled before the risk escalates, it is considered a valid early warning sample; if an early warning event is inconsistent with the on-site verification results, it is considered a threshold or association rule calibration sample; if a response measure fails to mitigate the risk in a timely manner, it is considered a sample for adjusting the response strategy.

[0045] Step S303: Optimize the model and rules based on the evaluation results. The closed-loop optimization module iteratively adjusts the weights, key feature indicator triggering conditions, combined anomaly judgment conditions in the association rule base, sequence anomaly order relationships, and proactive hierarchical response strategies in the multi-source data fusion analysis model based on valid early warning samples, false alarm samples, missed risk samples, and insufficient response samples, until the system operates stably and the early warning results match the on-site risk evolution stage.

[0046] Step S304: Enter the routine operation and maintenance phase after formal commissioning. After formal commissioning, the system's operating status and data collection integrity will be checked daily via automatic inspection; the accuracy of various sensors and acquisition modules will be verified weekly; the early warning effect of the multi-source data fusion analysis model will be evaluated monthly and parameters will be adjusted; and a system operation analysis report will be generated quarterly to evaluate the overall effectiveness of full-element perception, multi-element fusion analysis, four-level early warning judgment, proactive hierarchical response, and closed-loop optimization.

[0047] Using the methods described in Examples 1 to 3, the energy storage power station can continuously perceive electrical, thermal, gas, mechanical, and insulation risk factors within the same data framework, identify multi-factor combination anomalies or sequence anomalies in the early stages of risk evolution, and output a four-level early warning level based on the risk situation index and key characteristic indicators. At the same time, through proactive graded response and response effect feedback, the early warning judgment threshold and correlation rules are continuously optimized, thereby achieving four-level precise proactive security under full-factor risk perception.

Claims

1. An active security method for an energy storage power station, characterized in that, include: Collect data on five risk factors in energy storage power stations: electrical, thermal, gas, mechanical, and insulation factors, to form comprehensive sensing data. The full-element perception data is filtered and denoised, outlier removal is performed, and features are extracted to obtain key feature indicators of various risk elements. Based on the risk evolution mechanism, multi-element correlation analysis is performed on the key feature indicators to identify combination anomalies or sequence anomalies between at least two types of risk elements and generate a risk situation index. Based on the risk situation index and the triggering results of the key feature indicators, a four-level early warning level is determined according to the principle of choosing the higher level. The corresponding proactive tiered response strategy is executed according to the four warning levels, and the warning judgment threshold and association rules are optimized based on the response effect evaluation results.

2. The method according to claim 1, characterized in that, The five types of risk factor data include electrical characteristic parameters, thermal characteristic parameters, gas characteristic parameters, mechanical characteristic parameters, and insulation characteristic parameters. The electrical characteristic parameters include the rate of change of internal resistance and voltage inconsistency of battery cells or modules. The thermal characteristic parameters include the rate of temperature rise and temperature difference distribution. The gas characteristic parameters include the concentration of combustible gases and information on electrolyte decomposition products. The mechanical characteristic parameters include the internal pressure of the battery pack and the strain of the casing. The insulation characteristic parameters include the insulation resistance of the battery pack or power supply circuit.

3. The method according to claim 1, characterized in that, The feature extraction includes: uniform time calibration and data integrity checks on various risk factor data; horizontal difference analysis of similar parameters within the same battery pack or battery cluster; trend change analysis of continuous sampling data from the same measuring point or the same equipment unit; and outputting single-factor anomaly markers, rate of change markers, and trend markers for multi-factor correlation analysis.

4. The method according to claim 1, characterized in that, The multi-factor correlation analysis includes: establishing a rule base for correlation between risk factors; identifying a combined anomaly or a sequence anomaly when at least two of the following characteristics occur simultaneously or in a preset order within a preset correlation time window: increased internal resistance, accelerated temperature rise, detection of characteristic gases, increased internal pressure, strain change, or decreased insulation resistance; and generating the risk situation index based on the type, quantity, severity, and duration of the combined or sequence anomaly.

5. The method according to claim 1, characterized in that, The four-level early warning system includes Level 4, Level 3, Level 2, and Level 1. Level 4 corresponds to the nascent stage of potential hazards, Level 3 corresponds to the risk development stage, Level 2 corresponds to the danger approaching stage, and Level 1 corresponds to the accident critical stage. When any key characteristic indicator reaches the trigger condition of a level higher than the current risk situation index, the early warning level will be upgraded to the higher level corresponding to that key characteristic indicator.

6. The method according to claim 5, characterized in that, The proactive graded response strategy includes: issuing attention prompts and increasing inspection frequency during Level 4 warnings; issuing audible and visual warnings and notifying maintenance personnel to conduct on-site investigations during Level 3 warnings; notifying power plant duty personnel and the superior monitoring center and activating emergency plan preparation during Level 2 warnings; and coordinating proactive emergency response with the fire extinguishing system, fault circuit power outage system, or smoke exhaust ventilation system during Level 1 warnings.

7. The method according to claim 1, characterized in that, The optimized early warning judgment threshold and association rules include: recording the early warning level, triggering characteristics, response measures, response timeliness, and risk resolution results for each early warning event; and adjusting the calculation weight of the risk situation index, the triggering conditions of key feature indicators, or the combined anomaly judgment conditions in the association rule base according to the response effect evaluation results.

Images