Power grid natural disaster early warning method and system based on artificial intelligence

Abstract

The embodiment of the application provides a power grid natural disaster early warning method and system based on artificial intelligence, and belongs to the technical field of artificial intelligence. Firstly, a multi-source power grid and disaster information evolution association set containing power grid equipment operation dynamic information and corresponding regional natural disaster occurrence dynamic information and through regional identification real-time association and information evolution trend labeling is constructed, then time-series risk transmission analysis is performed, time-series association results of power grid equipment risk and disaster transmission are obtained, then a pre-trained power grid disaster intelligent hierarchical prediction model is called to perform multi-stage feedback prediction processing, power grid natural disaster hierarchical prediction results are generated, early warning strategy dynamic iterative optimization is executed according to the results, early warning strategy dynamic iterative adjustment parameters are generated, finally, power grid natural disaster accurate early warning instructions are generated based on the early warning strategy dynamic iterative adjustment parameters and sent to the power grid monitoring center terminal, and the power grid natural disaster can be early warned comprehensively, accurately and dynamically.

Description

Artificial Intelligence-Based Power Grid Natural Disaster Early Warning Method and System Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based method and system for early warning of natural disasters in power grids. Background Technology

[0002] In the field of power grid operation and management, natural disasters pose a serious threat to the safe and stable operation of power grid equipment. Traditional methods for early warning of natural disasters in power grids mainly rely on a single information source, such as disaster warnings issued solely by meteorological departments or monitoring only the operating parameters of power grid equipment. However, these single-source approaches have many limitations.

[0003] On the one hand, single meteorological information cannot accurately reflect the actual impact of a disaster on specific power grid equipment. Factors such as the terrain of different regions, the layout of power grid equipment, and the characteristics of the equipment itself can all lead to different impacts of the same natural disaster on different equipment. For example, in mountainous areas, strong winds may cause serious damage to transmission lines located on mountaintops, while having a smaller impact on lines located in valleys; different types of transformers also have different flood resistance capabilities when facing heavy rain.

[0004] On the other hand, focusing solely on the operating parameters of power grid equipment while ignoring the dynamic development of natural disasters makes it difficult to accurately predict the risks that equipment may face in advance. The occurrence and development of natural disasters is a dynamic process, with its intensity, scope, and duration of impact constantly changing. For example, the intensity and path of a typhoon change significantly before and after landfall. If this dynamic information cannot be obtained in a timely manner, it is impossible to accurately assess its impact on power grid equipment, and thus impossible to take effective preventive measures in advance.

[0005] Furthermore, most existing early warning methods lack the ability to capture the timing characteristics of risk transmission. Power grid equipment has complex electrical connections and interrelationships. When one piece of equipment malfunctions due to a disaster, the risk may be transmitted to other equipment through electrical connections, triggering a cascading failure. Traditional early warning methods cannot accurately analyze the timing characteristics of this risk transmission, making it difficult to provide effective early warning and intervention before a failure occurs.

[0006] Therefore, existing power grid natural disaster early warning methods are significantly inadequate in terms of accuracy and timeliness, and cannot meet the needs of safe and stable operation of modern power grids. Summary of the Invention

[0007] In view of this, the purpose of this application is to provide a power grid natural disaster early warning method and system based on artificial intelligence.

[0008] According to a first aspect of this application, an artificial intelligence-based method for early warning of natural disasters in power grids is provided, the method comprising:

[0009] A multi-source power grid and disaster information evolution association set is constructed. The multi-source power grid and disaster information evolution association set includes dynamic information on the operation of power grid equipment and dynamic information on the occurrence of natural disasters in the corresponding region. The dynamic information on the operation of power grid equipment and the dynamic information on the occurrence of natural disasters are linked in real time through regional identifiers, and the information evolution trend is marked.

[0010] A time-series risk transmission analysis is performed on the multi-source power grid and disaster information evolution association set to capture the time-series characteristics of risk transmission of different equipment types under the influence of disasters, and to obtain the time-series association results of power grid equipment risk and disaster transmission. The time-series association results include equipment risk transmission time-series curves and disaster impact diffusion time-series curves.

[0011] The pre-trained intelligent hierarchical prediction model for power grid disasters is invoked to perform multi-stage feedback prediction processing on the time-series correlation results. The prediction deviation is corrected through the internal feedback mechanism of the model to generate hierarchical prediction results for power grid natural disasters. The hierarchical prediction results for power grid natural disasters include dynamic prediction information of equipment risk level and dynamic prediction information of regional disaster impact.

[0012] Based on the power grid natural disaster stratified prediction results, the early warning strategy is dynamically iteratively optimized. The early warning parameters are adjusted based on the temporal changes of the prediction information, and the early warning strategy dynamic iterative adjustment parameters are generated. The early warning strategy dynamic iterative adjustment parameters include the equipment early warning intensity iterative adjustment content and the regional early warning range iterative adjustment content.

[0013] Based on the aforementioned early warning strategy, parameters are dynamically iteratively adjusted to generate precise early warning instructions for power grid natural disasters. These instructions include dynamic early warning intensity indicators, dynamic early warning coverage area indicators, and dynamic suggestions for emergency control of equipment. The precise early warning instructions for power grid natural disasters are then sent to the power grid monitoring center terminal.

[0014] According to a second aspect of this application, an artificial intelligence-based power grid natural disaster early warning system is provided. The artificial intelligence-based power grid natural disaster early warning system includes a processor and a readable storage medium. The readable storage medium stores a program that, when executed by the processor, implements the aforementioned artificial intelligence-based power grid natural disaster early warning method.

[0015] Based on any of the above aspects, by constructing a multi-source power grid and disaster information evolution association set, the dynamic information of power grid equipment operation and the dynamic information of natural disaster occurrence in the corresponding region are linked in real time through regional identifiers and the information evolution trend is marked. Then, a time-series risk transmission analysis is performed on the multi-source information to capture the time-series characteristics of risk transmission of different equipment types under the influence of disasters. The time-series association results, including equipment risk transmission time-series curves and disaster impact diffusion time-series curves, can deeply analyze the impact process and risk transmission mechanism of disasters on power grid equipment, making the early warning analysis more accurately reflect the actual situation. Next, a pre-trained intelligent hierarchical prediction model for power grid disasters is called to perform multi-stage feedback prediction processing on the time-series association results. The feedback mechanism corrects prediction biases, generating tiered prediction results for power grid natural disasters that include dynamic predictions of equipment risk levels and regional disaster impacts. This improves the accuracy and reliability of predictions, enabling early prediction of potential risk levels for power grid equipment and the degree of disaster impact on different regions. Then, based on the tiered prediction results, the early warning strategy is dynamically iteratively optimized. Early warning parameters are adjusted based on the temporal changes in prediction information, generating dynamic iterative adjustment parameters for the early warning strategy that include iterative adjustments to equipment early warning intensity and regional early warning range. This allows for timely adjustments to the early warning strategy based on dynamic changes in disasters and equipment status, making early warnings more relevant to actual conditions and improving their targeting and effectiveness. Finally, based on the dynamically iterative adjustment parameters of the early warning strategy, a precise early warning command for power grid natural disasters is generated, including dynamic early warning intensity identifiers, dynamic early warning coverage area identifiers, and dynamic suggestions for equipment emergency management. This command is then sent to the power grid monitoring center terminal, achieving precise early warning and timely management of power grid natural disasters. This effectively reduces the impact of natural disasters on power grid equipment, ensures the safe and stable operation of the power grid, and improves the power grid's ability and level of response to natural disasters. Attached Figure Description

[0016] Figure 1 shows a flowchart of the artificial intelligence-based power grid natural disaster early warning method provided in an embodiment of this application. Detailed Implementation

[0017] Figure 1 shows a flowchart of the power grid natural disaster early warning method based on artificial intelligence provided in an embodiment of this application. The detailed steps are described below.

[0018] Step S110: Construct a multi-source power grid and disaster information evolution association set. The multi-source power grid and disaster information evolution association set includes dynamic information on the operation of power grid equipment and dynamic information on the occurrence of natural disasters in the corresponding region. The dynamic information on the operation of power grid equipment and the dynamic information on the occurrence of natural disasters are linked in real time through regional identifiers, and the information evolution trend is marked.

[0019] In this embodiment, the impact of typhoon disasters and their resulting secondary disasters such as torrential rain and storm surges on the power grid system in a coastal area during typhoon season is used as the unified application scenario throughout the text. This coastal power grid covers multiple administrative regions and includes various types of equipment such as transmission lines, substations, distribution transformers, and dispatch centers. Historically, it has been affected by typhoon disasters multiple times, resulting in widespread power outages. The following steps are all described based on this scenario.

[0020] Step S111: Collect dynamic information on the operation of power grid equipment. The dynamic information on the operation of power grid equipment includes real-time operating status information of equipment, equipment load change information, equipment connection link on / off information and equipment environmental perception information. Each piece of information carries a collection time identifier.

[0021] In this coastal scenario, dynamic information on the operation of power grid equipment is collected through sensor networks and intelligent monitoring terminals deployed on various devices. For transmission lines, real-time operating status information includes parameters such as conductor temperature, sag, and insulator leakage current, collected via fiber optic grating sensors, infrared thermometers, and leakage current monitoring devices installed on the lines. Equipment load change information is obtained through smart meters at substations at both ends of the line, recording the active and reactive power transmitted at different times. The connectivity information of equipment links is determined by the auxiliary contact status signals of line circuit breakers and disconnectors; when the contacts are closed, the link is connected, and when they are open, the link is disconnected. Environmental perception information is collected by micro-weather stations installed on the towers, including data such as wind speed, wind direction, ambient temperature, humidity, and rainfall. Each piece of collected information is marked with a timestamp module accurate to the second; for example, the conductor temperature information of a transmission line at a certain moment will carry the time stamp of that moment.

[0022] Step S112: Collect dynamic information on natural disasters occurring in the corresponding area. The dynamic information on natural disasters includes information on changes in disaster type, information on the expansion of the disaster-affected area, information on fluctuations in disaster intensity, and information on the duration of the disaster. Each piece of information carries a collection time identifier.

[0023] For this coastal region, the collection of dynamic information on natural disasters primarily relies on the meteorological department's monitoring system and geological disaster monitoring stations. Information on changes in disaster type is determined through a comprehensive analysis of meteorological satellite cloud images, radar echo data, and information from ground observation stations, such as the process of a tropical depression gradually developing into a strong typhoon. Information on the expansion of the disaster's affected area is determined through typhoon path forecasts and wind and rain impact range forecasts issued by the meteorological department, and the affected area is divided into different levels of warning zones using a Geographic Information System (GIS). Information on fluctuations in disaster intensity includes real-time changes in parameters such as typhoon central pressure, maximum wind speed, and total rainfall, monitored through meteorological radar and an automatic weather station network. Information on the duration of the disaster begins counting from when the typhoon enters the warning range of the coastal region until the typhoon center moves out of the area and the impact weakens below the safe threshold. Each piece of natural disaster information also carries a collection time identifier to ensure correspondence with power grid equipment operation information in the time dimension.

[0024] Step S113: Extract the equipment installation area identifier from the dynamic information of the operation of the power grid equipment, wherein the equipment installation area identifier is a unique identifier of the geographical location of the power grid equipment; and extract the disaster impact area identifier from the dynamic information of the occurrence of natural disasters, wherein the disaster impact area identifier is a unique identifier of the geographical location currently affected by the natural disaster.

[0025] The equipment installation area identification for power grid equipment adopts a GIS-based coding system. For example, the tower number of a transmission line includes the administrative region code, line number, and tower serial number, forming a unique identifier such as "District-Line-Number". Substations are identified by a combination of their township name and station name, such as "Substation of Township X". These identifiers are pre-stored in the power grid equipment asset database and are automatically retrieved when collecting equipment operation information. Disaster impact area identification is determined based on the warning area codes defined by the meteorological department. For example, coastal areas are divided into multiple typhoon impact zones, each corresponding to a unique area code, such as "A01" and "A02". This, combined with township administrative boundary identification, enables precise location of the disaster impact range.

[0026] Step S114: Perform real-time matching processing between the equipment installation area identifier and the disaster-affected area identifier to generate an area identifier matching result. The area identifier matching result includes successfully matched identifier pairs and unmatched identifier pairs.

[0027] Regional identifier matching is achieved by establishing a mapping table. The administrative region code in the equipment installation area identifier is compared with the early warning zone code and township boundary identifier in the disaster impact area identifier. If the administrative region code of the equipment is located within the early warning zone or township boundary of the disaster impact area identifier, the match is considered successful, forming an identifier pair; otherwise, the match fails. For example, the equipment installation area identifier of a transmission line tower is "District-Line-Number," where "District" is the administrative region code. If this administrative region code is included in the "A01" early warning zone of the disaster impact area identifier, then the tower and the "A01" early warning zone form a successfully matched identifier pair. The matching process is executed every 5 minutes by a real-time data processing server to ensure the timeliness of the matching results.

[0028] Step S115: For the successfully matched identifiers, perform association and binding processing on the corresponding dynamic information of power grid equipment operation and dynamic information of natural disaster occurrence, add the same association and binding identifier, and sort them according to the collection time identifier to form an information time sequence.

[0029] For successfully matched identifier pairs, such as a substation and its corresponding typhoon-affected zone, their operational dynamics and disaster occurrence dynamics are associated by adding a binding identifier. This binding identifier is generated using a UUID (Universally Unique Identifier) ​​to ensure that each identifier pair corresponds to a unique UUID. For example, a UUID "uuid-001" is generated for the matching pair between a substation and the "A01" warning zone, and this UUID is added to all operational information of the substation and all disaster information of the "A01" zone. Subsequently, according to the order of the data collection time identifiers, all information associated with the same UUID is sorted to form a time-series information sequence, such as starting 24 hours before the typhoon's arrival, with hourly substation load information and corresponding typhoon intensity information arranged sequentially.

[0030] Step S116: For the identifiers that failed to match, the corresponding dynamic information of the power grid equipment operation and the dynamic information of natural disasters are temporarily stored. A timed rematching mechanism is set up to perform the regional identifier matching process again at preset intervals until the match is successful or the preset storage time is exceeded.

[0031] For failed matching pairs, such as a distribution transformer located in an inland area and an identifier for a current typhoon-affected area, the corresponding equipment operation information and disaster information are temporarily stored in a distributed cache database. During temporary storage, each piece of information is marked with the storage time and the reason for the matching failure, such as "the equipment area is not within the disaster impact range." A timed re-matching mechanism is implemented through a scheduled task set in the server, with a preset time interval of 15 minutes. That is, every 15 minutes, the temporarily stored information is retrieved from the cache database, and the area identifier matching process in step S114 is re-executed. If a match is still not successful within the preset storage period (e.g., 72 hours), the temporarily stored information is archived to the historical database and no longer participates in real-time association processing.

[0032] Step S117: Analyze the changing trends of dynamic information on power grid equipment operation based on information time series analysis, and mark the evolution direction of equipment operation status. At the same time, analyze the changing trends of dynamic information on natural disaster occurrence, and mark the evolution direction of disaster impact.

[0033] Taking a successfully matched transmission line information time series as an example, the changing trends of equipment operation dynamic information are analyzed. By curve fitting of parameters such as conductor temperature and load power at multiple consecutive time points, if the conductor temperature gradually increases over time and the load power shows a fluctuating upward trend, the evolution direction of the equipment operation status is marked as "increased risk"; if the parameters tend to stabilize or decrease, it is marked as "stable risk" or "decreased risk". For the dynamic information of natural disasters, similarly, by analyzing the time series curves of intensity parameters such as typhoon wind speed and rainfall, if the wind speed continues to increase and the rainfall continues to increase, the evolution direction of the disaster impact is marked as "intensity enhancement"; otherwise, it is marked as "intensity weakening". The evolution direction labeling results are added as attributes to each record of the information time series.

[0034] Step S118: Integrate the associated and bound information, the marked evolution direction, and the information temporarily stored for matching, classify and sort them according to the regional identifier, generate a multi-source power grid and disaster information evolution association set, and add a timestamp identifier to each piece of information in the multi-source power grid and disaster information evolution association set.

[0035] The integration process first aggregates the associated and bound information time sequence, labeled evolution direction information, and temporarily stored information awaiting matching into a unified data processing platform. Then, it categorizes the information according to the hierarchical relationship between equipment installation area identifiers and disaster impact area identifiers, for example, first dividing it into broad categories by municipal administrative regions, and then further subdividing it by districts, counties, and townships. Under each regional category, all relevant information is sorted in chronological order of collection timestamps, forming a complete set of multi-source power grid and disaster information evolution associations. Each piece of information in the set, in addition to its original collection timestamp, is given a globally unique timestamp, accurate to the millisecond level, ensuring time consistency during cross-regional data integration.

[0036] Step S120: Perform time-series risk transmission analysis on the multi-source power grid and disaster information evolution association set to capture the time-series characteristics of risk transmission of different equipment types under the influence of disasters, and obtain the time-series association results of power grid equipment risk and disaster transmission. The time-series association results include equipment risk transmission time-series curves and disaster impact diffusion time-series curves.

[0037] In this coastal typhoon disaster scenario, time-series risk transmission analysis aims to reveal the time-series characteristics of risk from occurrence to spread for different types of power grid equipment under the influence of typhoons and their secondary disasters. Through in-depth mining of various equipment operation data and disaster data in the multi-source power grid and disaster information evolution correlation set, a dynamic correspondence between equipment risk and disaster transmission is established.

[0038] Step S121: Classify the dynamic information of power grid equipment operation in the multi-source power grid and disaster information evolution association set according to equipment type to obtain the dynamic subset of power transmission equipment operation, the dynamic subset of power substation equipment operation, the dynamic subset of power distribution equipment operation, and the dynamic subset of dispatching equipment operation. Each subset is sorted by timestamp.

[0039] From the multi-source power grid and disaster information evolution association set, the dynamic information of power grid equipment operation is classified according to the equipment type field. The dynamic subset of transmission equipment operation includes the operation information of all transmission equipment such as transmission lines, towers, and insulators, such as conductor temperature and sag data of a 220kV transmission line; the dynamic subset of substation equipment operation covers the information of transformers, circuit breakers, disconnectors, and other substation equipment, such as oil temperature and winding temperature data of main transformers; the dynamic subset of distribution equipment operation includes the information of distribution transformers, distribution cabinets, overhead distribution lines, and other distribution equipment, such as load rate data of a distribution transformer in a certain area; the dynamic subset of dispatching equipment operation includes the operation status information of dispatching equipment such as servers, communication equipment, and SCADA systems in the dispatching center. After each subset is extracted, it is sorted in ascending order according to the timestamp identifier in the information to form a sequence of equipment operation data with time as the axis.

[0040] Step S122: Classify the dynamic information of natural disaster occurrence in the multi-source power grid and disaster information evolution association set according to disaster type to obtain dynamic subsets of rainstorm disaster occurrence, dynamic subsets of strong wind disaster occurrence, dynamic subsets of ice and snow disaster occurrence, and dynamic subsets of lightning disaster occurrence. Each subset is sorted by timestamp.

[0041] For the dynamic information on natural disaster occurrences in the multi-source power grid and disaster information evolution association set, classification is performed based on the disaster type field. In the typhoon disaster scenario in this coastal area, the dynamic subset of rainstorm disaster occurrence includes information such as rainfall amount, rainfall intensity, and duration caused by the typhoon; the dynamic subset of strong wind disaster occurrence covers data such as wind speed, wind direction, and time of maximum wind occurrence during the typhoon; since this region is located in the subtropics, the probability of snow and ice disasters is low, and the dynamic subset of snow and ice disaster occurrence is temporarily empty in this scenario, but the system still retains this classification to deal with special situations; the dynamic subset of lightning disaster occurrence includes information such as the number of lightning activities accompanying the typhoon, the number of thunderstorm days, and the location of lightning strikes. Each disaster subset is also sorted in ascending order by timestamp to ensure the temporal sequence of disaster data.

[0042] Step S123: For the dynamic subset of power transmission equipment operation and the dynamic subset of rainstorm disaster occurrence, extract the power transmission equipment operation status parameters and rainstorm disaster intensity parameters at different timestamps, analyze the time-series correspondence between rainstorm intensity changes and power transmission equipment insulation performance changes, and generate the first risk transmission time-series data.

[0043] Insulation performance parameters such as insulator leakage current and insulation resistance at different time points are extracted from the dynamic subset of power transmission equipment operation. Rainfall intensity parameters such as rainfall amount and intensity at corresponding time points are extracted from the dynamic subset of rainstorm disaster occurrence. The temporal correspondence between the rainstorm intensity parameters and insulation performance parameters is analyzed by calculating the rate of change of these parameters at different time points. For example, when rainfall gradually increases from a lower value to a higher value, does the insulator leakage current show an upward trend, and does the insulation resistance decrease accordingly? These correspondences are recorded with time on the horizontal axis and the rate of change of parameters on the vertical axis, generating the first risk transmission time series data. This first risk transmission time series data reflects the time process of how changes in rainstorm disaster intensity are transmitted to changes in the insulation performance of power transmission equipment.

[0044] Step S124: For the dynamic subset of substation equipment operation and the dynamic subset of lightning disaster occurrence, extract the substation equipment operation status parameters and lightning disaster intensity parameters at different timestamps, analyze the time sequence correspondence between lightning intensity changes and substation equipment circuit fault frequency changes, and generate second risk transmission time sequence data.

[0045] Circuit fault parameters, such as the number of circuit breaker trips and the number of relay protection actions at each time point, are extracted from the dynamic subset of power equipment operation. Lightning intensity parameters, such as the number of lightning activities and the magnitude of lightning current, are extracted from the dynamic subset of lightning disaster occurrences at corresponding time points. The analysis examines whether the number of circuit breaker trips and the number of relay protection actions increase accordingly when the number of lightning activities or the magnitude of lightning current increases at different time points, i.e., the change in circuit fault frequency. The correspondence between changes in lightning intensity parameters and changes in circuit fault frequency is organized into a time series to generate second risk propagation time series data. This second risk propagation time series data reflects the temporal impact characteristics of lightning disaster intensity on power equipment circuit faults.

[0046] Step S125: For the dynamic subset of power distribution equipment operation and the dynamic subset of snow and ice disaster occurrence, extract the power distribution equipment operation status parameters and snow and ice disaster intensity parameters at different timestamps, analyze the time-series correspondence between the changes in snow and ice thickness and the changes in the over-limit frequency of power distribution equipment load, and generate third risk transmission time-series data.

[0047] In this coastal typhoon disaster scenario, due to the low probability of snow and ice disasters, data analysis of the dynamic subsets of power distribution equipment operation and snow and ice disaster occurrence is only retained as a routine function of the system. If special circumstances lead to snow and ice disasters, load parameters such as ice thickness of power distribution lines and tower pressure are extracted from the dynamic subset of power distribution equipment operation, and intensity parameters such as ice and snow thickness and density are extracted from the dynamic subset of snow and ice disaster occurrence. The changes in the frequency of power distribution equipment load exceeding limits (such as the number of times the line ice thickness exceeds the design value) as ice and snow thickness increases are analyzed, and third-party risk transmission time-series data are generated according to the time series.

[0048] Step S126: For the dynamic subset of the operation of the dispatching equipment and the dynamic subset of the occurrence of the wind disaster, extract the operating status parameters of the dispatching equipment and the intensity parameters of the wind disaster at different timestamps, analyze the time-series correspondence between wind speed changes and the frequency changes of poor heat dissipation of the dispatching equipment, and generate the fourth risk transmission time-series data.

[0049] The system extracts heat dissipation-related parameters such as CPU temperature and ambient temperature of the dispatch center server from the dynamic subset of dispatch equipment operation, and wind speed parameters corresponding to the timestamps from the dynamic subset of strong wind disaster occurrences. Dispatch equipment is typically installed indoors, but strong winds may cause malfunctions in the server room ventilation system or reduce the heat dissipation efficiency of outdoor cooling towers, thus affecting equipment heat dissipation. The system analyzes whether the frequency of dispatch equipment CPU temperature exceeding the normal range (frequency of poor heat dissipation) increases when wind speed exceeds a certain threshold. The correlation between wind speed changes and changes in the frequency of poor heat dissipation is recorded in time series to generate fourth-risk transmission time-series data.

[0050] Step S127: Based on the first risk transmission time series data, the second risk transmission time series data, the third risk transmission time series data, and the fourth risk transmission time series data, plot the equipment risk transmission time series curve corresponding to each equipment type. The horizontal axis of the equipment risk transmission time series curve is the timestamp, and the vertical axis is the frequency of equipment risk occurrence.

[0051] Using timestamps as the horizontal axis and the frequency of equipment risk occurrence as the vertical axis, risk propagation time-series curves are plotted for different equipment types. For transmission equipment, based on the first risk propagation time-series data, the frequency of risk events caused by insulation performance degradation (such as the frequency of insulator flashover risk) at each timestamp is statistically analyzed, and the risk propagation time-series curve for transmission equipment is plotted. For substation equipment, based on the second risk propagation time-series data, the frequency of circuit faults over time is statistically analyzed, and the risk propagation time-series curve for substation equipment is plotted. For distribution equipment and dispatching equipment, based on the third and fourth risk propagation time-series data, the frequencies of overload risk and poor heat dissipation risk are statistically analyzed, and their respective risk propagation time-series curves are plotted. The curves are plotted using data visualization tools, presenting the correspondence between risk occurrence frequency and time as a continuous curve.

[0052] Step S128: Extract the disaster impact area expansion parameters under different timestamps, and draw the disaster impact diffusion time series curve corresponding to each disaster type. The horizontal axis of the disaster impact diffusion time series curve is the timestamp, and the vertical axis is the area of ​​the disaster impact area.

[0053] The boundary coordinates of disaster-affected areas at different timestamps are extracted from the dynamic information of natural disasters. The area of ​​the disaster-affected area corresponding to each timestamp is calculated using a Geographic Information System (GIS) as a parameter for expanding the disaster-affected area. For rainstorm disasters, the boundary coordinates of the rainfall-affected area at different timestamps are extracted, and the area is calculated. For gale disasters, the radius of the typhoon's wind circle or the boundary coordinates of its affected area are extracted, and the area is calculated. A disaster impact diffusion time-series curve is plotted with the timestamp as the horizontal axis and the disaster-affected area as the vertical axis. For example, the curve shows the process of the area affected by typhoons and gale-force winds gradually expanding from a small initial area to a maximum value and then gradually shrinking over time.

[0054] Step S129: Analyze the correlation between the equipment risk transmission time series curve and the corresponding disaster impact diffusion time series curve, calculate the similarity of the change trend of the equipment risk transmission time series curve and the corresponding disaster impact diffusion time series curve at the same time stamp, and mark the time interval where the similarity meets the preset threshold.

[0055] The Dynamic Time Warping (DTW) algorithm is used to calculate the similarity of the trends between the equipment risk transmission time-series curve and the corresponding disaster impact diffusion time-series curve at the same time stamp. For example, the curve shapes of the transmission equipment risk transmission time-series curve (vertical axis: insulation risk frequency) and the rainstorm disaster impact diffusion time-series curve (vertical axis: rainstorm impact area) at the same time stamp are compared, and the similarity value between the two is calculated. The preset similarity threshold is determined based on historical data statistical analysis, for example, set to 0.7. When the calculated similarity value is greater than or equal to 0.7, the time interval containing that time stamp is marked as a high-correlation interval, indicating that the equipment risk transmission and disaster impact diffusion have a strong temporal correlation within this interval.

[0056] Step S130: Integrate all equipment risk transmission time series curves, disaster impact diffusion time series curves, and labeled similarity time intervals to generate time series correlation results between power grid equipment risk and disaster transmission.

[0057] The completed risk transmission time-series curves for transmission, transformation, distribution, and dispatching equipment, as well as the disaster impact diffusion time-series curves for rainstorms, strong winds, and lightning (without snow disaster curves), and the time intervals where the similarity between each curve meets a preset threshold, are all integrated into a single data structure. Corresponding equipment type, disaster type, and time range identifiers are added to each curve and time interval, forming a time-series correlation result between power grid equipment risk and disaster transmission. This result is stored in both visual reports and structured data formats for easy use in subsequent model predictions.

[0058] Step S130: Call the pre-trained intelligent hierarchical prediction model for power grid disasters to perform multi-stage feedback prediction processing on the time-series correlation results, correct the prediction deviation through the internal feedback mechanism of the model, and generate hierarchical prediction results for power grid natural disasters. The hierarchical prediction results for power grid natural disasters include dynamic prediction information on equipment risk levels and dynamic prediction information on regional disaster impacts.

[0059] In this coastal typhoon disaster scenario, the pre-trained intelligent hierarchical prediction model for power grid disasters is a deep learning model trained on historical typhoon disaster data, power grid equipment failure data, and meteorological data. This deep learning model receives the temporal correlation results between power grid equipment risk and disaster transmission as input. Through multi-stage feedback prediction processing, it first predicts the risk level at the equipment level, then predicts the disaster impact at the regional level. Simultaneously, it utilizes the model's internal feedback mechanism to correct deviations generated during the prediction process, ultimately outputting a hierarchical prediction result for power grid natural disasters that includes dynamic prediction information at both the equipment and regional levels.

[0060] Step S131: Perform feature parsing processing on the time series correlation results, extract the equipment risk transmission time series features, disaster impact diffusion time series features and curve similarity features from the time series correlation results, and generate multi-dimensional prediction input features, which are arranged according to the time dimension.

[0061] Feature analysis is performed on the time-series correlation results between power grid equipment risks and disaster propagation. Time-series features such as slope, peak value, trough value, and periodicity of the equipment risk propagation curve are extracted as equipment risk propagation time-series features. Similarly, time-series features such as slope, peak value, and rate of change of affected area are extracted from the disaster impact diffusion time-series curve as disaster impact diffusion time-series features. Similarity values, interval length, and interval start time are extracted from the labeled similarity time intervals as curve similarity features. These features are arranged according to the time dimension; for example, each timestamp corresponds to a set of equipment risk propagation time-series features, disaster impact diffusion time-series features, and curve similarity features, forming a multi-dimensional prediction input feature sequence. Each element of this multi-dimensional prediction input feature sequence contains multiple feature dimensions, arranged in chronological order.

[0062] Step S132: Perform the first stage prediction processing based on the multi-dimensional prediction input features, predict the equipment risk level, and perform weighted calculation on the equipment risk transmission time series features and curve similarity features in the multi-dimensional prediction input features to generate a comprehensive time series feature value of equipment risk. The weighting coefficient is dynamically adjusted based on historical prediction errors.

[0063] The first stage of predictive processing focuses on equipment risk level prediction. It selects equipment risk transmission time-series features and curve similarity features from multi-dimensional prediction input features as input, and merges them into a comprehensive equipment risk time-series feature value through weighted calculation. The initial values ​​of the weighting coefficients are determined based on the contribution of the two types of features to equipment risk level prediction in historical data; for example, the initial weighting coefficient for equipment risk transmission time-series features is set to 0.6, and the initial weighting coefficient for curve similarity features is set to 0.4. During the prediction process, the prediction results are periodically compared with the actual equipment risk level, and the historical prediction error is calculated. If the error exceeds a preset range, the weighting coefficients are dynamically adjusted according to the magnitude and direction of the error. For example, when the prediction deviation is caused by excessively high weights for curve similarity features, its weighting coefficient is reduced, while the weighting coefficient for equipment risk transmission time-series features is increased.

[0064] Step S1321: Call the pre-stored feature weight configuration table in the intelligent hierarchical prediction model for power grid disasters. The feature weight configuration table includes the weight coefficients of the equipment risk transmission time series features and the weight coefficients of the curve similarity features. The initial weight coefficients are dynamically adjusted based on historical prediction errors.

[0065] The intelligent hierarchical prediction model for power grid disasters has a feature weight configuration table built during the training phase. This table stores the weight coefficients of the time-series features of equipment risk transmission and the curve similarity features. The initial weight coefficients are determined through error analysis of a large number of historical prediction cases. For example, in historical data, when the prediction error of the time-series features of equipment risk transmission is small, a higher initial weight coefficient is assigned; conversely, a lower initial weight coefficient is assigned. The feature weight configuration table is continuously updated during model training and practical application to adapt to the prediction needs of different scenarios.

[0066] Step S1322: Extract the feature quantization value corresponding to the device risk transmission time series feature in the multi-dimensional prediction input features. The feature quantization value is calculated by the slope and amplitude of the time series curve.

[0067] The equipment risk transmission time-series feature in the multi-dimensional prediction input features exists in the form of a curve. By calculating the slope (reflecting the rate of feature change) and amplitude (reflecting the range of feature fluctuation) of the curve in different time intervals, it is converted into a quantified value. For example, if the slope of the equipment risk transmission time-series curve is positive and the value is large within a certain period, it indicates that the equipment risk is rising rapidly during that period, and its quantified value will be correspondingly large; a large curve amplitude indicates that the equipment risk fluctuates drastically, which will also be reflected in the quantified value. A specialized feature quantization algorithm completes the conversion from curve to quantized value, ensuring that the quantized value accurately reflects the characteristics of the time-series curve.

[0068] Step S1323: Extract the feature quantization value corresponding to the curve similarity feature in the multi-dimensional prediction input features. The feature quantization value is calculated by the correlation coefficient of the two curves.

[0069] The quantitative value of the curve similarity feature is obtained by calculating the correlation coefficient between the equipment risk transmission time series curve and the disaster impact diffusion time series curve. The correlation coefficient is calculated using the Pearson correlation coefficient method, which measures the strength and direction of the linear correlation between the two curves. When the two curves have highly consistent trends, the correlation coefficient is close to 1; when the trends are completely opposite, the correlation coefficient is close to -1; and when there is no obvious linear relationship, the correlation coefficient is close to 0. The calculated correlation coefficient is used as the quantitative value of the curve similarity feature, which reflects the degree of linear association between equipment risk and disaster impact.

[0070] Step S1324: Standardize the quantified values ​​of the time-series characteristics of equipment risk transmission and the quantified values ​​of curve similarity characteristics, respectively, and convert them into dimensionless standard scores to make different characteristic values ​​comparable.

[0071] Because the quantified values ​​of equipment risk transmission time-series features and curve similarity features may have different dimensions and orders of magnitude, direct weighted calculation can affect the accuracy of the results. Therefore, standardization is necessary. The standardization process uses the Z-score standardization method, subtracting the mean of each feature in historical data from its quantified value, and then dividing by the standard deviation of that feature to obtain a dimensionless standard score. For example, for the quantified values ​​of equipment risk transmission time-series features, the difference between each value and the mean of all historical quantified values ​​of equipment risk transmission time-series features is calculated, and then divided by the historical standard deviation to obtain the standard score for that feature. The quantified values ​​of curve similarity features are processed in the same way, ensuring that the two types of feature values ​​are on the same order of magnitude and comparable.

[0072] Step S1325: Multiply the standardized equipment risk transmission time series feature quantization value with the equipment risk transmission time series feature weight coefficient to obtain the equipment risk transmission time series feature weighted value.

[0073] Obtain the weight coefficient of the equipment risk transmission time series feature from the feature weight configuration table, and multiply this coefficient by the standardized quantized value of the equipment risk transmission time series feature. For example, if the standardized quantized value of the equipment risk transmission time series feature is a certain value, and the weight coefficient of the equipment risk transmission time series feature is another value, the result of multiplying the two is the weighted value of the equipment risk transmission time series feature, which reflects the degree of contribution of the equipment risk transmission time series feature to the comprehensive feature.

[0074] Step S1326: Multiply the standardized curve similarity feature quantization value with the curve similarity feature weight coefficient to obtain the curve similarity feature weighted value.

[0075] Similarly, the curve similarity feature weight coefficients are obtained from the feature weight configuration table and multiplied by the standardized curve similarity feature quantification value to obtain the curve similarity feature weighted value. This curve similarity feature weighted value reflects the contribution of the curve similarity feature to the comprehensive feature; the larger the weight coefficient, the greater the influence of the curve similarity feature in the comprehensive evaluation.

[0076] Step S1327: Sum the weighted value of the equipment risk transmission time series feature and the weighted value of the curve similarity feature to obtain the initial equipment risk time series comprehensive feature value.

[0077] The initial comprehensive time-series feature value of equipment risk is obtained by adding the weighted values ​​of the equipment risk transmission time-series feature and the curve similarity feature. For example, if the weighted value of the equipment risk transmission time-series feature is one value and the weighted value of the curve similarity feature is another value, the sum of the two is the initial comprehensive time-series feature value of equipment risk, which integrates information from both equipment risk transmission and curve similarity.

[0078] Step S1328: Call the historical prediction error database in the intelligent hierarchical prediction model for power grid disasters and extract the prediction error values ​​corresponding to historical scenarios similar to the current prediction scenario.

[0079] The intelligent hierarchical prediction model for power grid disasters includes a historical prediction error database, which stores the model's prediction error values ​​under different historical prediction scenarios. Using a scenario similarity matching algorithm, historical scenarios similar to the current typhoon disaster scenario in coastal areas are searched from the database. For example, scenarios involving typhoons of similar intensity and path affecting similar power grid areas in the past are identified. The error values ​​between the equipment risk levels predicted by the model and the actual equipment risk levels under these similar historical scenarios are extracted and used as a reference for adjusting the current prediction deviation.

[0080] Step S1329: If the prediction error value of the historical scene exceeds the preset error threshold, the weight coefficients in the feature weight configuration table are adjusted according to the magnitude of the error value. The larger the error value, the greater the adjustment range. Priority is given to adjusting the feature weight coefficients that have been verified by historical data to have an impact on the prediction result exceeding the preset proportion.

[0081] The preset error threshold is set according to the model's prediction accuracy requirements, for example, as a certain percentage. When the extracted historical scene prediction error value exceeds this threshold, it indicates that the current feature weight configuration may be unreasonable and needs adjustment. The adjustment magnitude is directly proportional to the error value; the larger the error value, the greater the adjustment magnitude of the weight coefficient. During the adjustment process, the influence ratio of each feature on the prediction result is determined through historical data statistical analysis, and the weight coefficients of features whose influence ratio exceeds the preset ratio (e.g., 30%) are adjusted first. For example, if the influence ratio of the curve similarity feature on the prediction result exceeds 30%, and the historical prediction error is mainly caused by the unreasonable weight of this feature, then the weight coefficient of the curve similarity feature should be adjusted primarily.

[0082] Step S1330: Recalculate the standardized weighted values ​​of equipment risk transmission time series features and curve similarity features using the adjusted weighting coefficients, thereby obtaining the adjusted comprehensive feature value of equipment risk time series, and record the weighting coefficients used in this weighting calculation and the basis for adjustment.

[0083] The adjusted weighting coefficients for the time-series features of equipment risk transmission and the curve similarity features are re-multiplied by the standardized quantified values ​​of the time-series features of equipment risk transmission and the curve similarity features to obtain the adjusted weighted values ​​for the two features. These two weighted values ​​are then added together to obtain the adjusted comprehensive time-series feature value of equipment risk. Simultaneously, the weighting coefficient values ​​used in this adjustment, the changes before and after the adjustment, and the basis for the adjustment (such as historical prediction error values ​​and the proportion of feature influence) are recorded in detail in the model log for subsequent model optimization and auditing.

[0084] Step S133: Match the comprehensive feature value of equipment risk time series with the preset risk level mapping table to determine the preliminary equipment risk level prediction value. The risk level mapping table is divided into different mapping rules according to time intervals.

[0085] The pre-defined risk level mapping table divides the comprehensive time-series characteristic values ​​of equipment risk into multiple intervals, each interval corresponding to a risk level, such as low risk, medium risk, high risk, and extremely high risk. The risk level mapping table uses different mapping rules based on time intervals. For example, within 24 hours before a typhoon, a certain interval of the comprehensive time-series characteristic values ​​of equipment risk might correspond to a high-risk level; while after the typhoon, the same characteristic value interval might correspond to a medium-risk level. The adjusted comprehensive time-series characteristic values ​​of equipment risk are matched with the mapping rules corresponding to the current time interval to determine a preliminary predicted equipment risk level. For example, if the characteristic value falls within a high-risk interval, it is predicted to be a high-risk level.

[0086] Step S134: Compare the preliminary equipment risk level prediction with the preset historical equipment risk database. If the predicted value deviates from the actual risk level under the same historical time series scenario by more than the preset value, adjust the weighting coefficient, recalculate the equipment risk time series comprehensive characteristic value and risk level prediction value until the deviation meets the requirements, and obtain the dynamic prediction information of equipment risk level.

[0087] A pre-set historical equipment risk database stores a large number of actual equipment risk level values ​​under similar historical time-series scenarios. The preliminary predicted equipment risk level is compared with the actual risk levels under similar scenarios (such as the same type of equipment, the same type of disaster, and similar time-series characteristics) in the database, and the deviation is calculated. If the deviation exceeds a preset value (e.g., one risk level), the process returns to step S1329 to readjust the feature weight coefficients. The comprehensive time-series feature value and predicted risk level are recalculated according to the new weight coefficients, and compared again with the historical database until the deviation meets the requirements. The final determined predicted equipment risk levels are arranged in a time series, and each predicted value is labeled with a timestamp and deviation value, forming dynamic equipment risk level prediction information.

[0088] Step S135: Perform the second stage of prediction processing based on the multi-dimensional prediction input features to predict the regional disaster impact. First, extract the disaster impact diffusion time series features and curve similarity features from the multi-dimensional prediction input features to generate a disaster diffusion time series comprehensive feature vector.

[0089] The second stage of predictive processing focuses on regional disaster impact prediction. It extracts disaster impact diffusion time-series features (such as the rate of change of disaster-affected area and diffusion speed) and curve similarity features (such as the correlation coefficient between equipment risk and disaster diffusion) from the multi-dimensional prediction input features. These features are then arranged in a specific dimensional order to form a comprehensive disaster diffusion time-series feature vector. For example, the first element of the vector is the rate of change of disaster-affected area at a certain timestamp, the second element is the diffusion speed at that timestamp, the third element is the correlation coefficient, etc., and the length of the vector equals the total number of features.

[0090] Step S136: Simulate the spread process of the disaster within a preset time period based on the comprehensive feature vector of the disaster spread time series, and generate preliminary regional disaster impact prediction values. The simulation process incorporates the trend changes in the time series features.

[0091] Step S1361: Extract disaster type information from the disaster diffusion time series comprehensive feature vector, and call the corresponding disaster diffusion simulation algorithm according to the disaster type information. For rainstorm disasters, call the liquid diffusion simulation algorithm; for strong wind disasters, call the gas flow diffusion simulation algorithm; for snow and ice disasters, call the solid accumulation diffusion simulation algorithm; and for lightning disasters, call the pulse diffusion simulation algorithm.

[0092] The disaster diffusion time-series comprehensive feature vector includes a disaster type identifier field, which determines the type of disaster to be simulated. In this coastal typhoon disaster scenario, the main disasters involved are rainstorms and strong winds. For rainstorm disasters, a liquid diffusion simulation algorithm is used, which simulates the flow and infiltration of rainwater on the ground, considering the influence of factors such as terrain slope and soil permeability on disaster diffusion. For strong wind disasters, a gas flow diffusion simulation algorithm is used, which is based on fluid dynamics principles to simulate the movement and diffusion of typhoon airflow, considering factors such as pressure gradient and surface roughness.

[0093] Step S1362: Extract the disaster diffusion time series features from the disaster diffusion time series comprehensive feature vector, and determine the initial velocity, acceleration and trend change direction of disaster diffusion. The trend change direction is determined by the slope change of the time series curve.

[0094] Features related to the disaster spread rate are extracted from the comprehensive feature vector of the disaster spread time series, such as the increase in the disaster-affected area per unit time as the initial velocity; acceleration, i.e., the rate of change of velocity, is calculated based on the velocity changes at different timestamps. The direction of trend change is determined by analyzing the slope changes of the disaster spread time series curve. If the slope of the curve gradually increases, it indicates that the disaster spread trend is strengthening; if the slope gradually decreases, it indicates that the spread trend is weakening; if the slope remains basically unchanged, it indicates that the spread trend remains stable.

[0095] Step S1363: Extract curve similarity features from the disaster diffusion time series comprehensive feature vector, determine the synergistic relationship between equipment risk and disaster diffusion, and use this synergistic relationship as a constraint condition for the simulation process.

[0096] Curve similarity characteristics reflect the synergistic relationship between equipment risk and disaster spread. For example, when curve similarity is high, equipment risk increases synchronously with disaster spread. This synergistic relationship can be translated into constraints for the simulation process. For instance, when simulating disaster spread, if the affected area expands to a certain extent, the equipment risk level must reach the corresponding level; otherwise, the simulation results need to be corrected.

[0097] Step S1364: Set the time step for disaster spread simulation. The time step is determined based on the data acquisition interval in the time series characteristics, so that the simulation results are consistent with the time dimension of the time series data.

[0098] The data acquisition interval in time series features is typically 5 or 10 minutes. To ensure that the disaster spread simulation results are consistent with the time series data in the time dimension, the simulation time step is set to be the same as the data acquisition interval. For example, if the data acquisition interval is 10 minutes, then each time step in the simulation is 10 minutes, and the disaster spread status is calculated once after each time step.

[0099] Step S1365: Based on the disaster diffusion simulation algorithm, initial velocity, acceleration, trend change direction and constraints, calculate the boundary coordinates of the disaster-affected area at each time step.

[0100] Using initial velocity, acceleration, and the direction of trend change as input parameters, and combined with constraints on coordinated change relationships, the boundary coordinates of the disaster-affected area are calculated at each time step. For example, for strong wind disasters, based on the gas flow diffusion algorithm, the latitude and longitude coordinates of the typhoon's impact range at each time step are calculated according to the initial wind speed, wind acceleration, and the direction of trend change; at the same time, it is ensured that the risk level of equipment within this boundary coordinate range is coordinated with the degree of disaster diffusion.

[0101] Step S1366: Based on the boundary coordinates at each time step, calculate the area of ​​the disaster-affected region at the corresponding time step, and generate a sequence of changes in the area of ​​regional disaster-affected region within a future preset time period.

[0102] Using Geographic Information System (GIS) tools, the area of ​​the disaster-affected region is calculated based on the boundary coordinates of the region at each time step. The area values ​​at different time steps are arranged in chronological order to generate a sequence of changes in the disaster-affected area, which reflects the dynamic changes in the disaster-affected range within a future preset time period (such as the next 24 hours).

[0103] Step S1367: Based on the regional disaster impact area change sequence and combined with the impact intensity parameters corresponding to the disaster type, calculate the degree of impact of the disaster on the equipment in the region at each time step.

[0104] Different disaster types correspond to different impact intensity parameters. For example, the impact intensity parameters for rainstorm disasters include rainfall amount and rainfall intensity, while those for strong wind disasters include maximum wind speed and average wind speed. By multiplying the area value in the regional disaster impact area change sequence by the impact intensity parameter at the corresponding time step, and then multiplying by an empirical coefficient, we obtain the degree of impact of the disaster on equipment within the region at each time step. This degree of impact comprehensively reflects the impact of the disaster's scope and intensity on the equipment.

[0105] Step S1368: Integrate the regional disaster impact area and impact degree values ​​at each time step to generate preliminary regional disaster impact prediction values, which are arranged according to time step.

[0106] The calculated regional disaster impact area and impact severity values ​​at each time step are combined into a data pair and arranged in order of time step to form a preliminary regional disaster impact prediction value. For example, the prediction value at the first future time step (10 minutes later) includes the disaster impact area and impact severity values ​​for that time step, and the prediction value at the second future time step also includes the corresponding area and severity values, and so on.

[0107] Step S1369: Based on the trend changes in the disaster diffusion time series characteristics, perform trend smoothing on the preliminary regional disaster impact prediction values ​​to ensure that the changing trend of the regional disaster impact prediction values ​​is consistent with the trend in the time series characteristics.

[0108] The preliminary regional disaster impact forecasts are trend-smoothed using the moving average method or exponential smoothing method to eliminate random fluctuations in the forecasts and ensure that the trend of the forecasts aligns with the trend of the disaster diffusion time series characteristics. For example, if the disaster diffusion time series characteristics show an increasing trend in disaster impact, the smoothed regional disaster impact forecasts should also show a gradual upward trend to avoid abnormal downward fluctuations.

[0109] Step S1370: Record the algorithm, parameters and constraints used in the disaster diffusion simulation process.

[0110] Record the specific algorithm name (such as liquid diffusion simulation algorithm), the initial velocity, acceleration, time step and other parameter values ​​used, as well as the constraints of the cooperative change relationship in the simulation log during the disaster diffusion simulation process, so as to facilitate the subsequent traceability and analysis of the prediction results.

[0111] Step S137: Compare the preliminary regional disaster impact prediction value with the preset historical disaster impact database. If the deviation between the prediction value and the actual disaster impact range under the same historical time series scenario exceeds the preset value, adjust the simulation parameters and re-execute the simulation process until the deviation does not exceed the preset value, and obtain the dynamic prediction information of regional disaster impact.

[0112] A pre-defined historical disaster impact database stores data on the actual disaster impact range of similar historical time-series scenarios. The preliminary regional disaster impact prediction values ​​are compared with the actual data of similar scenarios in the database, and the deviation value (e.g., percentage deviation of affected area) is calculated. If the deviation value exceeds a preset value, the process returns to step S1362 to adjust the initial velocity, acceleration, and other simulation parameters of disaster spread, re-executes the simulation process, generates new regional disaster impact prediction values, and compares them again until the deviation does not exceed the preset value. The final regional disaster impact prediction values ​​are arranged in a time series to form dynamic regional disaster impact prediction information.

[0113] Step S138: The dynamic prediction information of equipment risk level and the dynamic prediction information of regional disaster impact are linked and integrated, prediction time identifier and prediction deviation identifier are added, and the power grid natural disaster hierarchical prediction result is generated. Based on the temporal correspondence between equipment risk level and regional disaster impact in the power grid natural disaster hierarchical prediction result, the coordinated change range of the two is marked.

[0114] The dynamic prediction information of equipment risk levels and the dynamic prediction information of regional disaster impacts are linked by timestamps, so that the predicted values ​​of equipment risk levels and regional disaster impacts at the same point in time correspond to each other. A prediction time identifier (e.g., predicting a specific future time point) and a prediction deviation identifier (e.g., deviation value, deviation level) are added to each prediction information. By analyzing the correspondence between the two over time, the coordinated change intervals in which equipment risk levels and regional disaster impacts rise or fall synchronously are marked. For example, if the area affected by a regional disaster expands while the equipment risk level also increases within a certain time period, this time period is the coordinated change interval. The integrated information, identifiers, and coordinated change intervals together constitute the hierarchical prediction results of power grid natural disasters.

[0115] Step S140: Perform dynamic iterative optimization of the early warning strategy based on the stratified prediction results of natural disasters in the power grid, adjust the early warning parameters based on the temporal changes of the prediction information, and generate dynamic iterative adjustment parameters for the early warning strategy. The dynamic iterative adjustment parameters for the early warning strategy include iterative adjustment content for equipment early warning intensity and iterative adjustment content for regional early warning range.

[0116] In this coastal typhoon disaster scenario, the dynamic iterative optimization of the early warning strategy aims to adjust the early warning parameters in real time based on the dynamic changes in equipment risk levels and regional disaster impacts in the power grid's tiered natural disaster prediction results. This ensures that the early warning strategy can adapt to the development of the disaster. Through continuous iterative optimization, it ensures that the early warning intensity and scope accurately reflect the actual threat of the disaster.

[0117] Step S141: Extract the dynamic prediction information of equipment risk level from the power grid natural disaster stratified prediction results, identify the risk level of each type of equipment at different prediction time steps, and the risk level includes the equipment core function risk level, equipment auxiliary function risk level and equipment cascading impact risk level.

[0118] Dynamic prediction information on equipment risk levels is extracted from the stratified prediction results of natural disasters in the power grid, and classified according to equipment type (e.g., transmission equipment, substation equipment, distribution equipment, dispatching equipment). For each type of equipment, its risk level is extracted at different prediction time steps (e.g., the next 1 hour, 2 hours, ..., 24 hours), including the risk level of the equipment's core function (e.g., the conductivity risk of transmission lines), the risk level of the equipment's auxiliary function (e.g., the communication risk of transmission lines), and the risk level of the equipment's cascading effects (e.g., the impact risk of a line fault on related lines). By parsing the risk level field and time step field in the prediction information, the specific numerical values ​​of the three risk levels for each type of equipment at each time step are determined.

[0119] Step S142: Call the preset warning intensity iterative adjustment rule library. The warning intensity iterative adjustment rule library contains warning intensity adjustment schemes corresponding to the temporal changes of different risk levels. Each warning intensity adjustment scheme includes a warning signal frequency iterative adjustment method, a warning information push frequency iterative adjustment method, and a warning response priority iterative adjustment method.

[0120] The pre-defined rule base for iterative adjustment of early warning intensity is a knowledge base built upon power grid safety regulations and historical emergency response experience. Each rule in the rule base corresponds to a risk level time-series change pattern, such as continuously increasing risk level, risk level first increasing and then decreasing, or risk level fluctuating. Each time-series change pattern is associated with an early warning intensity adjustment scheme. For example, for the continuously increasing risk level pattern, the iterative adjustment method for early warning signal frequency might be "increasing signal output once per time step," the iterative adjustment method for early warning information push frequency might be "increasing push frequency once per time step," and the iterative adjustment method for early warning response priority might be "increasing priority level by one per time step." The rule base is stored in structured data format for easy retrieval and matching by computer programs.

[0121] Step S143: For each type of device, according to the predicted time step order, match the risk level of each time step with the rules in the early warning intensity iterative adjustment rule base to determine the corresponding early warning intensity adjustment parameters for that time step, including the adjustment range of early warning signal frequency, the adjustment range of early warning information push frequency, and the adjustment range of early warning response priority.

[0122] Step S1431: For each type of equipment, extract the risk level under different prediction time steps from its equipment risk level dynamic prediction information, arrange them in order of prediction time steps, and form a time series sequence of equipment risk levels.

[0123] Taking a certain transmission line as an example, the risk levels of the core functions, auxiliary functions, and cascading effects of the equipment are extracted from the dynamic prediction information of the equipment risk level in the next 24 prediction time steps. These are arranged in order from the 1st to the 24th time step to form the time series sequence of the equipment risk level of the transmission line. Each element in the sequence contains three risk level values.

[0124] Step S1432: Encode the risk level corresponding to each predicted time step in the equipment risk level time series to generate a risk level time series code. The risk level time series code includes a time step code segment, a core function risk level code segment, an auxiliary function risk level code segment, and a chain effect risk level code segment.

[0125] A preset coding rule is used to encode the risk level of each predicted time step in the equipment risk level time series. The time step coding segment uses numbers to represent the predicted time step number (e.g., 01 represents the first time step, 02 represents the second time step, etc.); the core function risk level coding segment, the auxiliary function risk level coding segment, and the cascading impact risk level coding segment use letters or numbers to represent different risk levels (e.g., L represents low risk, M represents medium risk, H represents high risk, VH represents extremely high risk). For example, if the core function risk level is high risk, the auxiliary function risk level is medium risk, and the cascading impact risk level is low risk in the first time step, then its risk level time series coding might be "01-HML".

[0126] Step S1433: Call the rule encoding table in the rule base for iterative adjustment of early warning intensity. The rule encoding table contains the rule code corresponding to each rule. The structure of the rule code is consistent with the structure of the risk level time sequence code, and different adjustment parameters are associated according to the time step.

[0127] The rule coding table in the early warning intensity iterative adjustment rule base converts each rule into a rule code with the same time-series coding structure as the risk level. For example, a rule with the structure "Time step 01, core function risk level H, auxiliary function risk level M, cascading impact risk level L" has the rule code "01-HML" and is associated with the corresponding early warning intensity adjustment parameters. The rule codes in the rule coding table are organized by time step, with different time steps corresponding to different adjustment parameters.

[0128] Step S1434: Match the risk level time sequence code of the device with the rule codes in the rule code table segment by segment. First, match the time step code segment and filter out the rule subset corresponding to the current time step.

[0129] The time step code segment in the risk level time sequence code of the device is compared with the time step code segment of the rule code in the rule code table. Rule codes with the same time step code segment are selected to form a rule subset corresponding to the current time step. For example, if the risk level time sequence code of the first time step is "01-HML", then all rule codes in the rule code table with the time step code segment "01" are selected as the rule subset.

[0130] Step S1435: Within the rule subset, sequentially match the core function risk level code segment, the auxiliary function risk level code segment, and the cascading impact risk level code segment to determine the rule entry with the most complete matches or the most matched segments.

[0131] Within the rule subset corresponding to the time step, the core function risk level code segment of the equipment risk level time sequence code is matched with the core function risk level code segment of the rule code. Then, the auxiliary function and cascading impact risk level code segments are matched sequentially. The number of matching segments between each rule code and the equipment risk level time sequence code is counted, and the rule entry with the most complete matches (all four code segments are the same) or the most matching segments is selected as the matching result.

[0132] Step S1436: If there is a completely matching rule entry, directly extract the warning intensity adjustment parameters corresponding to the rule entry, including the warning signal frequency adjustment range, the warning information push frequency adjustment range, and the warning response priority adjustment range.

[0133] When a rule entry that perfectly matches the timing code of the device risk level is found within the rule subset, the associated warning intensity adjustment parameters are directly read from that rule entry. For example, the warning signal frequency adjustment range is "+1 times / time step", the warning information push frequency adjustment range is "+1 times / time step", and the warning response priority adjustment range is "+1 level".

[0134] Step S1437: If there is no completely matching rule entry, select the rule entry with the most matching segments as the reference rule, combine it with the warning adjustment data of the device under the same historical time step, modify the adjustment parameters of the reference rule, and generate warning intensity adjustment parameters that are adapted to the current time step.

[0135] If no rule entry matches exactly, the rule entry with the most matching segments is selected as the reference rule. The device's historical warning adjustment data for the same historical time step (e.g., the first time step) is extracted from the device's historical warning database. The differences between the actual warning adjustment parameters and the reference rule adjustment parameters for similar risk level combinations in historical data are analyzed, and the adjustment parameters of the reference rule are corrected based on these differences. For example, if the reference rule's warning signal frequency adjustment range is "+1 times / time step," while the actual adjustment range for similar scenarios in the same historical time step is "+2 times / time step," then the adjustment range of the reference rule is corrected to "+2 times / time step."

[0136] Step S1438: Record the matching rule entries, adjustment parameters, and correction basis under the current time step.

[0137] The code of the rule entry matched at the current time step, the final determined warning intensity adjustment parameters, and the basis for correction (such as historical warning adjustment data, number of matching segments, etc.) are recorded in detail in the device warning log for subsequent auditing and rule base optimization.

[0138] Step S1439: Repeat the above matching, extraction or correction process according to the prediction time step order to obtain the warning intensity adjustment parameters of the device under all prediction time steps.

[0139] Starting from the first prediction time step, the matching, extraction, or correction process of steps S1434 to S1438 is executed sequentially for each time step until the processing of all prediction time steps is completed, so as to obtain the adjustment range of the warning signal frequency, the adjustment range of the warning information push frequency, and the adjustment range of the warning response priority of the device at each time step.

[0140] Step S14310: Arrange all warning intensity adjustment parameters under the prediction time step in chronological order, label the risk level code and rule matching result corresponding to each parameter, and form the warning intensity adjustment parameter sequence of the device.

[0141] Arrange the warning intensity adjustment parameters of the device in order of time step across all prediction time steps, and label each parameter with the corresponding risk level code (e.g., "01-HML") and rule matching results (e.g., fully matched rule codes or reference rule codes and correction status), forming a complete sequence of warning intensity adjustment parameters for the device.

[0142] Step S14311: Analyze the parameter change trend in the early warning intensity adjustment parameter sequence, mark the increasing or decreasing range of the parameters, and make the parameter changes consistent with the risk level change trend.

[0143] The changing trend of parameters is analyzed by calculating the difference between parameters at adjacent time steps in the warning intensity adjustment parameter sequence. If the parameter value gradually increases with the time step, it is marked as an increasing interval; if it gradually decreases, it is marked as a decreasing interval. It is ensured that the increasing or decreasing intervals of the parameters correspond to the changing trend of the risk level in the equipment risk level time series. For example, an interval of continuously rising risk level corresponds to an increasing interval of the warning intensity adjustment parameter.

[0144] Step S144: Based on the warning intensity adjustment parameters under different time steps, generate an equipment warning intensity iteration sequence. The equipment warning intensity iteration sequence is arranged according to the time step and the adjustment basis corresponding to each time step is marked.

[0145] The equipment's warning intensity adjustment parameters (adjustment range of warning signal frequency, adjustment range of warning information push frequency, and adjustment range of warning response priority) at different prediction time steps are arranged in order of time step to form an iterative sequence of equipment warning intensity. The adjustment basis is marked at each time step position of the sequence, such as the matched rule code, historical data correction, etc., so that each adjustment parameter in the sequence has a clear source.

[0146] Step S145: Integrate the early warning intensity iteration sequences of all devices according to device type, add device type identifier and time step identifier, and generate device early warning intensity iteration adjustment content.

[0147] The iterative sequences for early warning intensity of different types of equipment, such as transmission equipment, substation equipment, distribution equipment, and dispatching equipment, are categorized separately. Each sequence is labeled with an equipment type identifier (e.g., "Transmission-Line 001", "Substation-Transformer 002") and a time step identifier (e.g., "Prediction Time Step 01", "Prediction Time Step 02"). The iterative sequences for early warning intensity of the same type of equipment are then aggregated to form equipment early warning intensity adjustment content categorized by equipment type, facilitating subsequent association with regional early warning range adjustments.

[0148] Step S146: Extract the regional disaster impact dynamic prediction information from the power grid natural disaster stratified prediction results, and identify the disaster impact range of each region under different prediction time steps. The disaster impact range includes the core impact area, the secondary impact area and the potential impact area.

[0149] Dynamic forecasting information on regional disaster impacts is extracted from the stratified forecasting results of natural disasters in the power grid, and different regions are divided according to administrative regions or power grid zones. For each region, the disaster impact range at different forecast time steps is identified. By analyzing the predicted disaster impact area, boundary coordinates, and impact severity values, the core impact area (the area with the highest disaster impact severity value), secondary impact area (the area with the next highest disaster impact severity value), and potential impact area (the area that may be affected by the disaster but with a lower degree of impact) are divided. For example, at the 5th forecast time step, the core impact area of ​​a certain region is a 5-square-kilometer area in the center of the region, the secondary impact area is a 10-square-kilometer area extending outward from the center, and the potential impact area is a further 15-square-kilometer area extending outward from the center.

[0150] Step S147: Call the preset warning range iterative adjustment rule library. The warning range iterative adjustment rule library contains warning range adjustment schemes corresponding to different impact range time sequence changes. Each warning range adjustment scheme includes a range expansion iteration method, a range contraction iteration method, and a range hierarchical iteration method.

[0151] The pre-defined rule base for iterative adjustment of early warning ranges formulates corresponding adjustment schemes based on the temporal change patterns of regional disaster impact ranges (such as continuous expansion, expansion followed by contraction, and tiered diffusion). The range expansion iteration method specifies how the early warning range should expand when the disaster impact range expands (e.g., proportionally expanding the boundary); the range contraction iteration method specifies how the early warning range should contract when the disaster impact range shrinks; and the tiered range iteration method specifies how the early warning ranges for core, secondary, and potential impact areas should be adjusted tiered (e.g., the core area has the highest early warning level, and the potential area has the lowest). The rule base is also stored in structured data format, containing descriptions of the temporal change patterns of impact ranges and corresponding adjustment schemes.

[0152] Step S148: For each region, according to the predicted time step order, match the disaster impact range under each time step with the rules in the early warning range iterative adjustment rule base to determine the early warning range adjustment parameters corresponding to that time step. The early warning range adjustment parameters include the adjustment range of the expansion boundary, the adjustment range of the contraction boundary, and the adjustment method of the graded identifier.

[0153] Step S1481: For each region, extract the disaster impact range under different prediction time steps from its regional disaster impact dynamic prediction information, arrange them in order of prediction time steps, and form a time series sequence of regional impact range.

[0154] Taking a coastal administrative region as an example, the boundary coordinates or area data of the core impact area, secondary impact area and potential impact area under the next 24 prediction time steps are extracted from the regional disaster impact dynamic prediction information and arranged in the order of time steps to form the time series sequence of the regional impact range of the region.

[0155] Step S1482: Extract the boundary coordinates of the disaster impact range corresponding to each predicted time step in the time series of the regional impact range to obtain the boundary coordinate set of the core impact area, the boundary coordinate set of the secondary impact area, and the boundary coordinate set of the potential impact area. Each coordinate set is labeled according to the time step.

[0156] Using Geographic Information System (GIS) tools, the boundary coordinates of the core impact area, secondary impact area, and potential impact area are extracted from the disaster impact range data at each prediction time step in the regional impact range time series, forming three boundary coordinate sets. Each coordinate set contains multiple coordinate points (longitude and latitude) and is labeled with the corresponding prediction time step number. For example, the boundary coordinate set of the core impact area at the third prediction time step contains all coordinate points on the boundary of the core area at that time step.

[0157] Step S1483: Call the range description table in the rule base for iterative adjustment of the warning range. The range description table contains the warning range boundary coordinate template corresponding to each rule. The warning range boundary coordinate template is classified according to the time step. Each warning range boundary coordinate template contains the standard boundary coordinate structure of the core area, secondary area and potential area.

[0158] The range description table in the rule base for iterative adjustment of the early warning range converts the early warning range adjustment scheme of each rule into a boundary coordinate template. These templates are categorized according to the prediction time step. Each time step's template contains the standard boundary coordinate structure (such as a sequence of polygon vertex coordinates) for the core region, secondary region, and potential region. For example, a rule in the third time step corresponds to an early warning range boundary coordinate template, where the core region template is a sequence of polygon boundary coordinates, and the secondary and potential region templates are similar.

[0159] Step S1484: Calculate the similarity between the core influence area boundary coordinate set at the current time step and the core area boundary coordinate template at the same time step in the range description table. Use the sum of coordinate point distance deviations as the similarity index to select the rule subset corresponding to the core area template whose similarity index meets the preset minimum threshold.

[0160] Calculate the distance between each coordinate point in the core influence area boundary coordinate set at the current time step and the corresponding coordinate point in the core area boundary coordinate template at the same time step in the range description table. Sum all distances to obtain the total distance deviation. The smaller the total distance deviation, the higher the similarity. A preset minimum threshold is set as the upper limit of the total distance deviation. When the calculated total deviation is less than this threshold, the rule corresponding to the core area template is considered to match the current area influence range, and it is included in the rule subset.

[0161] Step S1485: Within the rule subset, calculate the similarity between the secondary impact area boundary coordinate set and the secondary area template, and the potential impact area boundary coordinate set and the potential area template at the current time step, to further filter the rules and determine the rule entries that meet the preset minimum threshold for the similarity index of the disaster impact range at the current time step.

[0162] Within the rule subset selected by template matching in the core region, the same method of summing coordinate point distance deviations is used to sequentially calculate the similarity between the boundary coordinate set of the secondary influence region and the secondary region template within the rule subset at the current time step, and the similarity between the boundary coordinate set of the potential influence region and the potential region template. Only when the similarity indices of the core, secondary, and potential regions all meet their respective preset minimum thresholds is the rule entry determined as the final matching rule.

[0163] Step S1486: If the selected rule item is a range expansion scheme, extract the expansion boundary adjustment range from the scheme, including the core area expansion range, the secondary area expansion range and the potential area expansion range, and extract the hierarchical identifier adjustment method.

[0164] When the warning range adjustment scheme corresponding to the matched rule entry is a range expansion scheme, the expansion range of the core area (such as expanding the boundary outward by a certain distance), the expansion range of the secondary area, the expansion range of the potential area, and the adjustment method of the hierarchical label (such as upgrading the warning label of the core area from yellow to orange) are read from the scheme.

[0165] Step S1487: If the selected rule item is a range shrinkage scheme, then extract the shrinkage boundary adjustment range from the scheme, including the core area shrinkage range, the secondary area shrinkage range and the potential area shrinkage range, and extract the hierarchical identification adjustment method.

[0166] When the matched rule entry is a range shrinkage scheme, extract the core area shrinkage range (such as shrinking the boundary inward by a certain distance), secondary area shrinkage range, potential area shrinkage range, and the graded label adjustment method (such as downgrading the potential area warning label from blue to green) from the scheme.

[0167] Step S1488: Record the matching rule entries, adjustment parameters, and similarity calculation results at the current time step.

[0168] The rule entry number matched at the current time step, the extracted warning range adjustment parameters (expansion / contraction range, hierarchical identification adjustment method), and the similarity calculation results (total distance deviation) of the core, secondary, and potential areas are recorded in the regional warning log.

[0169] Step S1489: Repeat the above boundary extraction, similarity calculation, rule filtering and parameter extraction processes in the order of prediction time steps to obtain the warning range adjustment parameters for the region under all prediction time steps.

[0170] From the first prediction time step to the last prediction time step, steps S1484 to S1488 are executed sequentially for each time step to obtain the warning range adjustment parameters for the region at each time step.

[0171] Step S14810: Arrange all warning range adjustment parameters under the prediction time step in chronological order, label the disaster impact range boundary coordinate set and rule matching results corresponding to each parameter, and form the warning range adjustment parameter sequence for the region.

[0172] Arrange the warning range adjustment parameters for the region in chronological order across all prediction time steps. Next to each parameter, label the corresponding disaster impact range boundary coordinate set identifier (e.g., "core boundary set 05") and the rule matching result (e.g., the matched rule entry number), thus forming the warning range adjustment parameter sequence for the region.

[0173] Step S14811: Analyze the parameter change trend in the parameter sequence of the warning range adjustment, mark the increasing or decreasing range of the parameters, and make the parameter changes consistent with the change trend of the disaster impact range.

[0174] Analyze the changing trends of the adjustment magnitudes of the expansion or contraction boundaries in the parameter sequence for adjusting the warning range. If the expansion magnitude gradually increases with the time step, it indicates that the warning range is expanding rapidly, and this is marked as an increasing interval; if the contraction magnitude gradually increases, it indicates that the warning range is contracting rapidly, and this is marked as a decreasing interval. Ensure that the parameter changing trends are consistent with the actual changing trends of the disaster impact range.

[0175] Step S14812: Integrate the area identifier, warning range adjustment parameter sequence, and trend label.

[0176] Add a regional identifier (such as "Coastal Region A") to the warning range adjustment parameter sequence for this region, and attach trend labeling results to complete the integration of the warning range adjustment parameter sequence for this region.

[0177] Step S149: Based on the warning range adjustment parameters under different time steps, generate a regional warning range iteration sequence. The regional warning range iteration sequence is arranged according to the time step and the adjustment basis corresponding to each time step is marked.

[0178] The adjustment parameters for the warning range of each region at different prediction time steps are arranged in order of time step to form an iterative sequence of regional warning ranges. The basis for adjustment (such as matching rule entries, similarity calculation results, etc.) is marked at each time step position in the sequence, so that each adjustment parameter in the sequence has a clear source.

[0179] Step S150: Integrate the iterative sequences of the warning range for all regions according to the region identifier, add the region identifier and time step identifier, and generate the iterative adjustment content of the regional warning range.

[0180] The iterative sequences of regional early warning ranges for each administrative region or power grid zone are categorized and summarized according to regional identifiers. A regional identifier and a time step identifier are added to each sequence to form the iterative adjustment content of the regional early warning range. This iterative adjustment content of the regional early warning range corresponds to the iterative adjustment content of the equipment early warning intensity, and together they form the basis for the dynamic adjustment of the early warning strategy.

[0181] Step S151: Establish the correlation between the iterative adjustment content of equipment warning intensity and the iterative adjustment content of regional warning range, so that under the same time step, the adjustment of equipment warning intensity and the adjustment of warning range in the same region remain coordinated and consistent.

[0182] By using region identifiers and time step identifiers, the device warning intensity adjustment parameters for the same region and time step in the device warning intensity iterative adjustment content are associated with the corresponding regional warning range adjustment parameters in the regional warning range iterative adjustment content. For example, if the regional warning range is adjusted to an expanded state at the 5th prediction time step, the warning intensity adjustment of all devices in that region should also be coordinated to appropriately increase the warning intensity, avoiding the contradictory situation of the regional warning range expanding while the device warning intensity decreases.

[0183] Step S152: Integrate the associated device warning intensity iterative adjustment content and regional warning range iterative adjustment content, add adjustment parameter generation time identifier and rule matching identifier, and generate warning strategy dynamic iterative adjustment parameters.

[0184] The associated device warning intensity iterative adjustment content and regional warning range iterative adjustment content are merged into a single data structure. The specific time (accurate to the second) for generating the adjustment parameters and rule matching identifiers (such as the total number of rules matched in this adjustment, the matching success rate, etc.) are added to ultimately generate dynamic iterative adjustment parameters for the warning strategy. These dynamic iterative adjustment parameters for the warning strategy contain detailed information on warning adjustments at both the device and regional levels for each future time step.

[0185] Step S153: Based on the early warning strategy, dynamically iterate and adjust the adjustment trend of parameters at different time steps, and mark the direction of parameter iterative optimization.

[0186] Analyze the changing trends of the equipment warning intensity and regional warning range adjustment parameters in the dynamic iterative adjustment parameters of the early warning strategy at different time steps. If most parameters show an increasing trend in the next few time steps, it indicates that the early warning strategy needs to be iteratively optimized in the direction of strengthening the early warning; if most parameters show a decreasing trend, it needs to be iteratively optimized in the direction of weakening the early warning.

[0187] Step S150: Based on the early warning strategy, dynamically iteratively adjust the parameters to generate a precise early warning instruction for power grid natural disasters. The precise early warning instruction for power grid natural disasters includes a dynamic early warning intensity identifier, a dynamic early warning coverage area identifier, and dynamic suggestions for emergency control of equipment. The precise early warning instruction for power grid natural disasters is then sent to the power grid monitoring center terminal.

[0188] In this typhoon disaster scenario in the coastal area, parameters are dynamically and iteratively adjusted based on the early warning strategy. The abstract adjustment parameters are transformed into specific early warning instructions, which specify when, where, and at what intensity the early warning is issued, as well as what emergency control measures are taken for which equipment. Finally, the instructions are sent to the power grid monitoring center terminal to guide the actual power grid emergency response work.

[0189] For example, step S151: Extract the device early warning intensity iterative adjustment content from the dynamic iterative adjustment parameters of the early warning strategy, and obtain the frequency adjustment range, push frequency adjustment range, and priority adjustment range for each device under different prediction time steps.

[0190] The iterative adjustment content of equipment early warning intensity is extracted from the dynamic iterative adjustment parameters of the early warning strategy. This is then retrieved by equipment type and prediction time step to obtain the adjustment range of early warning signal frequency (e.g., "+2 times / hour"), early warning information push frequency (e.g., "+1 times / 30 minutes"), and early warning response priority (e.g., "+1 level") for each type of equipment at each prediction time step. For example, the frequency adjustment range for transmission line 001 at the third prediction time step is "+2 times / hour", the push frequency adjustment range is "+1 times / 30 minutes", and the priority adjustment range is "+1 level".

[0191] Step S152: Determine the output frequency identifier of the early warning signal corresponding to each prediction time step according to the frequency adjustment amplitude under different prediction time steps, determine the push frequency identifier of the early warning information corresponding to each prediction time step according to the push frequency adjustment amplitude, and determine the early warning response priority identifier corresponding to each prediction time step according to the priority adjustment amplitude.

[0192] The system establishes a pre-defined correspondence between the output frequency identifier and the frequency adjustment range of the pre-set warning signal (e.g., "F1" means once per hour, "F2" means twice per hour, and so on). The corresponding frequency identifier is selected based on the frequency adjustment range of the device under different prediction time steps. Similarly, the system pre-defined warning information push frequency identifier (e.g., "P1" means once every 30 minutes, "P2" means once every 15 minutes, etc.) and warning response priority identifier (e.g., "PR1" means priority level 1, "PR2" means priority level 2, etc.) are determined based on the push frequency adjustment range and priority adjustment range, respectively. For example, a frequency adjustment range of "+2 times / hour" corresponds to the frequency identifier "F3", a push frequency adjustment range of "+1 times / 30 minutes" corresponds to the push identifier "P2", and a priority adjustment range of "+1 level" corresponds to the priority identifier "PR2".

[0193] Step S153: Integrate the frequency identifier, push frequency identifier, and priority identifier under the same prediction time step to generate a dynamic warning intensity identifier corresponding to the prediction time step, and arrange them in order of prediction time step to form a dynamic warning intensity identifier sequence.

[0194] The frequency identifier of the warning signal output, the frequency identifier of the warning information push, and the priority identifier of the warning response for a device at the same prediction time step are combined into a single string to serve as the dynamic warning intensity identifier for that prediction time step. For example, if the frequency identifier for the third prediction time step is "F3", the push frequency identifier is "P2", and the priority identifier is "PR2", then the dynamic warning intensity identifier is "F3-P2-PR2". The dynamic warning intensity identifiers for this device at all prediction time steps are then arranged in order of time step to form a dynamic warning intensity identifier sequence.

[0195] Step S154: Extract the regional early warning range iterative adjustment content from the dynamic iterative adjustment parameters of the early warning strategy, and obtain the adjustment range of the expansion boundary, the adjustment range of the contraction boundary, and the adjustment method of the hierarchical label for each region under different prediction time steps.

[0196] The iterative adjustment content of the regional early warning range is extracted from the dynamic iterative adjustment parameters of the early warning strategy. This is retrieved by region identifier and prediction time step to obtain the adjustment range of the expansion boundary (e.g., "expanding eastward by 2 kilometers"), the adjustment range of the contraction boundary (e.g., "contracting westward by 1 kilometer"), and the adjustment method of the graded identifier (e.g., "core area identifier upgraded to red") for each region at different prediction time steps. For example, the expansion boundary adjustment range of coastal region A at the 5th prediction time step is "expanding eastward by 2 kilometers," and the graded identifier adjustment method is "core area identifier upgraded to red."

[0197] Step S155: Based on the adjustment range of the expanded boundary or the adjustment range of the contracted boundary under different prediction time steps, calculate the boundary coordinates of the warning range corresponding to each prediction time step, and convert the boundary coordinates into standardized geographic area identifiers.

[0198] Using Geographic Information System (GIS) tools, the boundary coordinates of the current warning area are adjusted and calculated based on the expansion or contraction of the boundary at different prediction time steps, resulting in new boundary coordinates. These new boundary coordinates are then matched with a pre-defined geographic region coding system and converted into standardized geographic region identifiers (such as "350203001" based on administrative division codes or "G10234567" based on grid codes) to ensure that the warning area can be accurately identified by the power grid monitoring system.

[0199] Step S156: Combine the hierarchical identifier adjustment method corresponding to each prediction time step to generate a dynamic early warning coverage area identifier corresponding to that prediction time step, and arrange them in order of prediction time step to form a dynamic early warning coverage area identifier sequence.

[0200] By combining the standardized geographic region identifier with the hierarchical identifier adjustment method at each prediction time step, a dynamic early warning coverage area identifier is generated. For example, if the geographic region identifier is "350203001" and the hierarchical identifier adjustment method is "core area identifier upgraded to red, secondary area identifier to yellow, and potential area identifier to blue", then the dynamic early warning coverage area identifier is "350203001-red-yellow-blue". The dynamic early warning coverage area identifiers for this region at all prediction time steps are arranged in time step order to form a dynamic early warning coverage area identifier sequence.

[0201] Step S157: Call the preset equipment emergency management suggestion library. The equipment emergency management suggestion library contains management suggestions corresponding to different combinations of dynamic early warning intensity indicators and dynamic early warning coverage area indicators. Each management suggestion includes suggestions for dynamic adjustment of equipment operating parameters, suggestions for dynamic adjustment of equipment inspection focus, and suggestions for emergency handling of equipment failures.

[0202] The pre-built equipment emergency management suggestion library is a structured knowledge base constructed based on power grid equipment operation and maintenance procedures, historical fault handling cases, and expert experience. Each record in this library is indexed by a combination of dynamic warning intensity identifier and dynamic warning coverage area identifier, linking it to the corresponding equipment emergency management suggestion. For example, when the dynamic warning intensity identifier is "F3-P2-PR2" and the dynamic warning coverage area identifier is "350203001-Red-Yellow-Blue", the indexed management suggestions might include: dynamic adjustment suggestions for equipment operating parameters (such as reducing the transmission power of transmission lines by a certain percentage), dynamic adjustment suggestions for equipment inspection priorities (such as increasing the inspection frequency of tower foundations in the core impact area), and dynamic suggestions for emergency handling of equipment faults (such as preparing backup power vehicles to deal with possible substation power outages). Each suggestion entry in the library details the operation object, operation content, operation timing, and expected goals to ensure its feasibility.

[0203] Step S158: According to the prediction time step order, match the dynamic early warning intensity identifier and dynamic early warning coverage area identifier under each prediction time step with the combination in the equipment emergency management suggestion library to determine the equipment emergency management dynamic suggestion corresponding to the prediction time step.

[0204] Starting from the first prediction time step, the dynamic warning intensity identifier and dynamic warning coverage area identifier for that time step are extracted sequentially. These two identifiers are combined into search keywords and precisely matched in the equipment emergency management suggestion library. If a completely matching combination exists, the corresponding management suggestion is directly extracted as the equipment emergency management dynamic suggestion for that time step. If no completely matching combination exists, a combination with the same dynamic warning intensity identifier and consistent core area level in the dynamic warning coverage area identifier is selected as an approximate match. The approximate matching management suggestion is then adaptively modified based on the equipment type and regional characteristics in the current scenario to generate an adapted equipment emergency management dynamic suggestion. For example, if there is no completely matching item in the suggestion library for a certain time step identifier combination, but there is a combination with the same warning intensity and core area level, the inspection focus suggestion for that combination can be adjusted to the inspection focus for a specific equipment type in the current area.

[0205] Step S159: Integrate the dynamic early warning intensity identifier, dynamic early warning coverage area identifier, and dynamic equipment emergency control suggestions under each prediction time step according to the preset instruction format, add the instruction generation time, instruction unique code, and instruction generation basis, and generate a power grid natural disaster accurate early warning instruction sequence arranged according to the prediction time step.

[0206] The preset instruction format defines the structure and fields of the instruction, including an instruction header, instruction body, and instruction tail. The instruction header contains the instruction generation time (accurate to milliseconds) and a unique instruction code (using UUID format); the instruction body contains the dynamic warning intensity identifier, the dynamic warning coverage area identifier, and detailed information on the dynamic recommendations for emergency control of equipment; the instruction tail contains the basis for instruction generation (such as the associated warning strategy dynamic iterative adjustment parameter number and the matching rule entry number). Following this format, the above information for each prediction time step is integrated into an independent precise early warning instruction for power grid natural disasters. All instructions are arranged in order of prediction time steps, forming an instruction sequence. For example, the instruction for the first time step includes the intensity identifier, area identifier, control recommendations, generation time "2024-07-20T10:00:00.000", unique code "uuid-0001", and generation basis "adjustment parameter ID:AP20240720001, rule ID:R001".

[0207] Step S1510: The sequence of accurate early warning instructions for natural disasters in the power grid is sent to the power grid monitoring center terminal through an encrypted communication link, and transmitted sequentially according to the predicted time step.

[0208] The transmission of precise early warning command sequences for natural disasters in the power grid employs an encrypted communication link based on the SSL / TLS protocol to ensure the confidentiality and integrity of the commands during transmission. Before transmission, the command sequence is compressed to reduce data volume, and a Message Authentication Code (MAC) is added for the receiving end to verify data integrity. The sending end sends each command sequentially according to the predicted time step. After each command is sent, it waits for confirmation from the receiving end before sending the next command. If a transmission error occurs, it automatically retransmits until all commands are successfully sent. After receiving the commands, the power grid monitoring center terminal parses the command content and displays dynamic early warning information on the monitoring system interface. Simultaneously, it pushes dynamic suggestions for emergency equipment management to the terminal devices of relevant maintenance teams.

[0209] Step S1511: Receive confirmation information for each prediction time step instruction from the power grid monitoring center terminal, and record the instruction sending status according to the prediction time step.

[0210] After sending each instruction, the sending end starts a timeout timer, waiting for a reception confirmation from the power grid monitoring center terminal. The confirmation message includes a unique instruction code and the reception status (e.g., "successfully received," "format error," "verification failure," etc.). Based on the confirmation message, the sending end records the transmission status of each instruction according to the predicted time step. Instructions that are "successfully received" are marked as "delivered," instructions with "format error" or "verification failure" are marked as "retransmission required," and instructions that have not received confirmation within the timeout period are marked as "timeout." A command transmission status report is generated periodically, containing statistics such as the number of commands sent, the number of successes, and the reasons for failures at each time step. This information is used for communication link quality assessment and subsequent command transmission strategy optimization.

Claims

1. An artificial intelligence-based method for early warning of natural disasters in power grids, characterized in that, The method includes: constructing a multi-source power grid and disaster information evolution association set, which contains dynamic information on the operation of power grid equipment and dynamic information on the occurrence of natural disasters in the corresponding region. The dynamic information on the operation of power grid equipment and the dynamic information on the occurrence of natural disasters are linked in real time through regional identifiers, and the information evolution trend is marked; performing time-series risk transmission analysis on the multi-source power grid and disaster information evolution association set to capture the time-series characteristics of risk transmission of different equipment types under the influence of disasters, and obtaining the time-series association results of power grid equipment risk and disaster transmission. The time-series association results include equipment risk transmission time-series curves and disaster impact diffusion time-series curves; calling a pre-trained intelligent hierarchical prediction model for power grid disasters to perform multi-stage feedback prediction processing on the time-series association results, through the model's internal feedback mechanism. The system corrects prediction biases and generates tiered prediction results for natural disasters affecting the power grid. These tiered prediction results include dynamic prediction information on equipment risk levels and dynamic prediction information on regional disaster impacts. Based on these tiered prediction results, the system performs dynamic iterative optimization of early warning strategies, adjusting early warning parameters according to the temporal changes in prediction information to generate dynamic iterative adjustment parameters for the early warning strategies. These dynamic iterative adjustment parameters include iterative adjustments to equipment early warning intensity and regional early warning range. Based on these dynamic iterative adjustment parameters, the system generates precise early warning commands for natural disasters affecting the power grid. These precise early warning commands include dynamic early warning intensity identifiers, dynamic early warning coverage area identifiers, and dynamic suggestions for equipment emergency management. These precise early warning commands are then sent to the power grid monitoring center terminal.

2. The power grid natural disaster early warning method based on artificial intelligence according to claim 1, characterized in that, The construction of a multi-source power grid and disaster information evolution association set includes dynamic information on the operation of power grid equipment and dynamic information on the occurrence of natural disasters in the corresponding region. The dynamic information on the operation of power grid equipment and the dynamic information on the occurrence of natural disasters are linked in real time through regional identifiers, and the information evolution trend is marked. This includes: collecting dynamic information on the operation of power grid equipment, which includes real-time operating status information, load change information, connectivity information, and environmental perception information of the equipment, with each piece of information carrying a collection time identifier; collecting dynamic information on the occurrence of natural disasters in the corresponding region, which includes information on changes in disaster type, expansion of disaster-affected area, fluctuations in disaster intensity, and duration of disaster, with each piece of information carrying a collection time identifier; extracting the equipment installation area identifier from the dynamic information on the operation of power grid equipment, which is a unique identifier of the geographical location of the power grid equipment; and extracting the disaster-affected area identifier from the dynamic information on the occurrence of natural disasters, which is a unique identifier of the geographical location currently affected by the natural disaster. The process involves: 1) Real-time matching of the equipment installation area identifier with the disaster-affected area identifier to generate an area identifier matching result, which includes successfully matched and unmatched identifier pairs; 2) Association and binding of the corresponding power grid equipment operation dynamic information and natural disaster occurrence dynamic information for successfully matched identifiers, adding identical association and binding identifiers, and sorting them by collection time identifiers to form an information time sequence; 3) Temporary storage of the corresponding power grid equipment operation dynamic information and natural disaster occurrence dynamic information for unmatched identifiers, setting a timed re-matching mechanism to perform area identifier matching again at preset intervals until a match is successful or the preset storage time is exceeded; 4) Analysis of the changing trend of power grid equipment operation dynamic information based on the information time sequence, marking the evolution direction of equipment operation status, and analysis of the changing trend of natural disaster occurrence dynamic information, marking the evolution direction of disaster impact; 5) Integration of the associated and bound information, marked evolution directions, and temporarily stored information to be matched, sorted by area identifier, to generate a multi-source power grid and disaster information evolution association set, and adding a timestamp identifier to each piece of information in the multi-source power grid and disaster information evolution association set.

3. The power grid natural disaster early warning method based on artificial intelligence according to claim 1, characterized in that, The step involves performing a time-series risk transmission analysis on the multi-source power grid and disaster information evolution association set to capture the time-series characteristics of risk transmission for different equipment types under the influence of disasters, and obtaining the time-series association results between power grid equipment risk and disaster transmission. This includes: classifying the dynamic information of power grid equipment operation in the multi-source power grid and disaster information evolution association set by equipment type to obtain dynamic subsets of transmission equipment operation, substation equipment operation, distribution equipment operation, and dispatching equipment operation, each subset being sorted by timestamp; and classifying the dynamic information of natural disaster occurrence in the multi-source power grid and disaster information evolution association set by disaster type to obtain dynamic subsets of rainstorm disaster occurrence, strong wind disaster occurrence, and snow and ice disaster occurrence. The system extracts dynamic subsets of power transmission equipment operation and lightning disaster occurrence, each sorted by timestamp. For the power transmission equipment operation dynamic subset and the rainstorm disaster occurrence dynamic subset, it extracts power transmission equipment operation status parameters and rainstorm disaster intensity parameters at different timestamps, analyzes the time-series correspondence between changes in rainstorm intensity and changes in power transmission equipment insulation performance, and generates the first risk transmission time-series data. For the substation equipment operation dynamic subset and the lightning disaster occurrence dynamic subset, it extracts substation equipment operation status parameters and lightning disaster intensity parameters at different timestamps, analyzes the time-series correspondence between changes in lightning intensity and changes in substation circuit fault frequency, and generates the second risk transmission time-series data. For the distribution equipment operation dynamic subset and the snow and ice disaster occurrence dynamic subset, it extracts different... The operating status parameters of power distribution equipment at different timestamps and the intensity parameters of snow and ice disasters are analyzed to determine the time-series correspondence between changes in snow and ice thickness and changes in the frequency of overload on power distribution equipment, generating third risk transmission time-series data. For the dynamic subsets of dispatching equipment operation and the dynamic subsets of strong wind disaster occurrence, the operating status parameters of dispatching equipment at different timestamps and the intensity parameters of strong wind disasters are extracted to determine the time-series correspondence between changes in wind speed and changes in the frequency of poor heat dissipation of dispatching equipment, generating fourth risk transmission time-series data. Based on the first, second, third, and fourth risk transmission time-series data, equipment risk transmission time-series curves are plotted for each equipment type. The horizontal axis of the time-series curve represents the timestamp, and the vertical axis represents the frequency of equipment risk occurrence. Disaster impact area expansion parameters are extracted at different timestamps, and disaster impact diffusion time-series curves corresponding to each disaster type are plotted. The horizontal axis of these curves represents the timestamp, and the vertical axis represents the area of ​​the disaster-affected region. The correlation between the equipment risk transmission time-series curve and the corresponding disaster impact diffusion time-series curve is analyzed. The similarity of the changing trends of the equipment risk transmission time-series curve and the corresponding disaster impact diffusion time-series curve at the same timestamp is calculated, and time intervals where the similarity meets a preset threshold are marked. All equipment risk transmission time-series curves, disaster impact diffusion time-series curves, and marked similarity time intervals are integrated to generate the time-series correlation results between power grid equipment risk and disaster transmission.

4. The power grid natural disaster early warning method based on artificial intelligence according to claim 1, characterized in that, The process involves calling a pre-trained intelligent hierarchical prediction model for power grid disasters to perform multi-stage feedback prediction processing on the time-series correlation results. This process corrects prediction deviations through an internal model feedback mechanism, generating hierarchical prediction results for power grid natural disasters. The process includes: performing feature parsing on the time-series correlation results to extract equipment risk transmission time-series features, disaster impact diffusion time-series features, and curve similarity features, generating multi-dimensional prediction input features arranged by time dimension; performing a first-stage prediction process based on these multi-dimensional prediction input features to predict equipment risk levels, weighting the equipment risk transmission time-series features and curve similarity features in the multi-dimensional prediction input features to generate a comprehensive equipment risk time-series feature value, with the weighting coefficient dynamically adjusted based on historical prediction errors; matching the comprehensive equipment risk time-series feature value with a preset risk level mapping table to determine a preliminary predicted equipment risk level value, where the risk level mapping table is divided into different mapping rules according to time intervals; and comparing the preliminary predicted equipment risk level value with a preset historical equipment risk database. If the predicted value deviates from the actual risk level in similar historical time-series scenarios by more than a preset value, the weighting coefficient is adjusted, and the equipment risk is recalculated. The process involves analyzing time-series comprehensive feature values ​​and risk level prediction values ​​until the deviation meets the requirements, resulting in dynamic prediction information for equipment risk levels. A second-stage prediction process is then performed based on the multi-dimensional prediction input features to predict regional disaster impacts. First, disaster impact diffusion time-series features and curve similarity features are extracted from the multi-dimensional prediction input features to generate a disaster diffusion time-series comprehensive feature vector. Based on this comprehensive feature vector, the diffusion process of the disaster within a preset time period is simulated to generate preliminary regional disaster impact prediction values. The simulation process incorporates trend changes in the time-series features. The preliminary regional disaster impact prediction values ​​are compared with a preset historical disaster impact database. If the deviation between the predicted value and the actual disaster impact range under similar historical time-series scenarios exceeds a preset value, the simulation parameters are adjusted, and the simulation process is re-executed until the deviation does not exceed the preset value, resulting in dynamic prediction information for regional disaster impacts. Finally, the dynamic prediction information for equipment risk levels and the dynamic prediction information for regional disaster impacts are linked and integrated, with prediction time and prediction deviation identifiers added to generate a tiered prediction result for power grid natural disasters. Based on the time-series correspondence between equipment risk levels and regional disaster impacts in the tiered prediction result for power grid natural disasters, the coordinated change range of the two is marked.

5. The power grid natural disaster early warning method based on artificial intelligence according to claim 4, characterized in that, The step of weighting and calculating the equipment risk transmission time-series features and curve similarity features in the multi-dimensional prediction input features to generate a comprehensive equipment risk time-series feature value includes: calling a pre-stored feature weight configuration table in the intelligent hierarchical prediction model for power grid disasters, wherein the feature weight configuration table contains weight coefficients for equipment risk transmission time-series features and curve similarity features, and the initial weight coefficients are determined based on historical data training; extracting the feature quantization values ​​corresponding to the equipment risk transmission time-series features in the multi-dimensional prediction input features, wherein the feature quantization values ​​are calculated by the slope and amplitude of the time-series curves; extracting the feature quantization values ​​corresponding to the curve similarity features in the multi-dimensional prediction input features, wherein the feature quantization values ​​are calculated by the correlation coefficient of the two curves; standardizing the equipment risk transmission time-series feature quantization values ​​and the curve similarity feature quantization values ​​respectively, converting them into dimensionless standard scores to make different feature values ​​comparable; and multiplying the standardized equipment risk transmission time-series feature quantization values ​​by the equipment risk transmission time-series feature weight coefficients to obtain... The process involves: 1) Obtaining a weighted value for the time-series features of equipment risk transmission; 2) Multiplying the standardized quantized value of the curve similarity feature with the weight coefficient of the curve similarity feature to obtain a weighted value for the curve similarity feature; 3) Summing the weighted value of the time-series features of equipment risk transmission with the weighted value of the curve similarity feature to obtain an initial comprehensive time-series feature value for equipment risk; 4) Calling the historical prediction error database in the intelligent hierarchical prediction model for power grid disasters to extract prediction error values ​​corresponding to historical scenarios similar to the current prediction scenario; 5) If the prediction error value of a historical scenario exceeds a preset error threshold, adjusting the weight coefficients in the feature weight configuration table according to the magnitude of the error value. The larger the error value, the larger the adjustment. Prioritizing the adjustment of feature weight coefficients whose influence on the prediction result exceeds a preset proportion verified by historical data; 6) Recalculating the standardized weighted value of the time-series features of equipment risk transmission and the weighted value of the curve similarity feature using the adjusted weight coefficients to obtain the adjusted comprehensive time-series feature value for equipment risk, and recording the weight coefficients used in this weighting calculation and the basis for adjustment.

6. The power grid natural disaster early warning method based on artificial intelligence according to claim 4, characterized in that, The method simulates the diffusion process of a disaster within a preset time period based on the comprehensive feature vector of disaster diffusion time series, generating preliminary regional disaster impact predictions. The simulation process incorporates trend changes in the time series features, including: extracting disaster type information from the comprehensive feature vector of disaster diffusion time series; calling the corresponding disaster diffusion simulation algorithm based on the disaster type information (e.g., liquid diffusion simulation algorithm for rainstorm disasters, gas flow diffusion simulation algorithm for strong wind disasters, solid accumulation diffusion simulation algorithm for snow and ice disasters, and pulse diffusion simulation algorithm for lightning disasters); extracting disaster diffusion time series features from the comprehensive feature vector of disaster diffusion time series, determining the initial velocity, acceleration, and trend change direction of disaster diffusion, with the trend change direction determined by the slope change of the time series curve; extracting curve similarity features from the comprehensive feature vector of disaster diffusion time series, determining the synergistic relationship between equipment risk and disaster diffusion, and using this synergistic relationship as a constraint condition for the simulation process; and setting the time step for disaster diffusion simulation, the time step being based on data acquisition from the time series features. The intervals are determined to ensure consistency between the simulation results and the time dimension of the time series data. Based on the disaster diffusion simulation algorithm, initial velocity, acceleration, trend change direction, and constraints, the boundary coordinates of the disaster-affected area are calculated at each time step. Based on the boundary coordinates at each time step, the area of ​​the disaster-affected area at the corresponding time step is calculated, generating a sequence of changes in the area of ​​the disaster-affected region within a future preset time period. Based on the sequence of changes in the area of ​​the disaster-affected region, combined with the impact intensity parameters corresponding to the disaster type, the degree of impact of the disaster on equipment within the region at each time step is calculated. The area of ​​the disaster-affected region and the degree of impact at each time step are integrated to generate preliminary predicted values ​​of the regional disaster impact, which are arranged by time step. Based on the trend changes in the disaster diffusion time series characteristics, the preliminary predicted values ​​of the regional disaster impact are smoothed to ensure that the trend of the predicted values ​​of the regional disaster impact is consistent with the trend in the time series characteristics. The algorithms, parameters, and constraints used in the disaster diffusion simulation process are recorded.

7. The power grid natural disaster early warning method based on artificial intelligence according to claim 1, characterized in that, The step of dynamically iteratively optimizing the early warning strategy based on the stratified prediction results of natural disasters in the power grid, adjusting the early warning parameters based on the temporal changes of the prediction information, and generating dynamic iterative adjustment parameters for the early warning strategy includes: extracting dynamic prediction information of equipment risk levels from the stratified prediction results of natural disasters in the power grid, identifying the risk level of each type of equipment at different prediction time steps, wherein the risk level includes the risk level of the core function of the equipment, the risk level of the auxiliary function of the equipment, and the risk level of the cascading impact of the equipment; and calling a preset early warning intensity iterative adjustment rule library, wherein the early warning intensity iterative adjustment rule library contains early warning intensity adjustment schemes corresponding to the temporal changes of different risk levels, and each early warning intensity adjustment scheme includes an iterative adjustment method for the early warning signal frequency and early warning information. The document outlines several methods for adjusting push frequency and early warning response priority. For each device, the risk level at each predicted time step is matched with rules in the early warning intensity adjustment rule base to determine the corresponding early warning intensity adjustment parameters, including frequency adjustment magnitude, push frequency adjustment magnitude, and priority adjustment magnitude. Based on these parameters, a device early warning intensity iteration sequence is generated, arranged by time step and labeled with the adjustment basis for each time step. All device early warning intensity iteration sequences are then categorized and integrated by device type, with device type and time step identifiers added to generate the device early warning intensity iteration adjustment mechanism. The process involves: extracting dynamic prediction information of regional disaster impacts from the hierarchical prediction results of natural disasters in the power grid; identifying the disaster impact range of each region at different prediction time steps, whereby the disaster impact range includes core impact areas, secondary impact areas, and potential impact areas; invoking a preset early warning range iterative adjustment rule library, which contains early warning range adjustment schemes corresponding to different temporal changes in impact ranges, with each scheme including range expansion iteration, range contraction iteration, and range hierarchical iteration; and for each region, matching the disaster impact range at each time step with the rules in the early warning range iterative adjustment rule library according to the prediction time step order to determine the corresponding disaster impact range at that time step. The system includes parameters for adjusting the warning range, such as the adjustment range of the expansion boundary, the adjustment range of the contraction boundary, and the adjustment method of the hierarchical identifier. Based on the adjustment parameters for the warning range at different time steps, a regional warning range iterative sequence is generated. This regional warning range iterative sequence is arranged by time step, and the adjustment basis for each time step is marked. The warning range iterative sequences of all regions are classified and integrated according to regional identifiers, and regional identifiers and time step identifiers are added to generate regional warning range iterative adjustment content. A correlation is established between the equipment warning intensity iterative adjustment content and the regional warning range iterative adjustment content, so that the warning intensity adjustment of equipment in the same region at the same time step is consistent with the warning range adjustment of that region.The iterative adjustment content of the associated device warning intensity and the iterative adjustment content of the regional warning range are integrated, and time identifiers and rule matching identifiers are added to the adjustment parameters to generate dynamic iterative adjustment parameters for the warning strategy. Based on the adjustment trend of different time steps in the dynamic iterative adjustment parameters of the warning strategy, the direction of iterative optimization of the parameters is marked.

8. The power grid natural disaster early warning method based on artificial intelligence according to claim 7, characterized in that, For each type of equipment, the risk level at each predicted time step is matched with the rules in the early warning intensity iterative adjustment rule base to determine the corresponding early warning intensity adjustment parameters for that time step. This includes: for each type of equipment, extracting the risk levels at different predicted time steps from its dynamic prediction information, arranging them in order of predicted time steps to form a time series sequence of equipment risk levels; encoding the risk level corresponding to each predicted time step in the time series sequence to generate a risk level time series code, wherein the risk level time series code includes a time step encoding segment. The system includes: core function risk level coding segment, auxiliary function risk level coding segment, and cascading impact risk level coding segment; it calls the rule coding table in the early warning intensity iterative adjustment rule base, which contains the rule code corresponding to each rule. The structure of the rule code is consistent with the structure of the risk level time sequence coding, and different adjustment parameters are associated according to the time step; the risk level time sequence coding of the equipment is matched segment by segment with the rule codes in the rule coding table. First, the time step coding segment is matched, and the rule subset corresponding to the current time step is filtered out; within the rule subset, the core function risk level coding segment, ... For the risk level coding segments of auxiliary functions and the risk level coding segments of cascading effects, identify the rule entries that are completely matched or have the most matching segments. If a completely matched rule entry exists, directly extract the warning intensity adjustment parameters corresponding to that rule entry, including the adjustment range of warning signal frequency, the adjustment range of warning information push frequency, and the adjustment range of warning response priority. If no completely matched rule entry exists, select the rule entry with the most matching segments as the reference rule. Combine this with the warning adjustment data of the device under the same historical time step, and modify the adjustment parameters of the reference rule to generate warning intensity adjustment parameters adapted to the current time step. Record the matching rule entries, adjustment parameters, and correction basis under the current time step. Repeat the above matching, extraction, or correction process according to the predicted time step order to obtain the warning intensity adjustment parameters of the device under all predicted time steps. Arrange the warning intensity adjustment parameters under all predicted time steps in chronological order, and label the risk level code and rule matching result corresponding to each parameter to form the warning intensity adjustment parameter sequence of the device. Analyze the parameter change trend in the warning intensity adjustment parameter sequence, and label the increasing or decreasing range of the parameters to ensure that the parameter changes are consistent with the risk level change trend.

9. The power grid natural disaster early warning method based on artificial intelligence according to claim 7, characterized in that, For each region, the process involves matching the disaster impact range at each predicted time step with the rules in the early warning range iterative adjustment rule base to determine the corresponding early warning range adjustment parameters. This includes: for each region, extracting the disaster impact range at different predicted time steps from its regional disaster impact dynamic prediction information, arranging them in order of predicted time steps to form a regional impact range time series; extracting the boundary coordinates of the disaster impact range corresponding to each predicted time step in the regional impact range time series to obtain the core impact area boundary coordinate set, the secondary impact area boundary coordinate set, and the potential impact area boundary coordinate set, with each coordinate set labeled according to the time step; and calling... The warning range iterative adjustment rule base includes a range description table. This table contains warning range boundary coordinate templates for each rule, categorized by time step. Each template contains standard boundary coordinate structures for core areas, secondary areas, and potential areas. The similarity is calculated between the core influence area boundary coordinate set at the current time step and the core area boundary coordinate templates at the same time step in the range description table. The sum of coordinate point distance deviations is used as the similarity index. A subset of rules corresponding to core area templates whose similarity index meets a preset minimum threshold is selected. Within this subset, the secondary influence area boundary coordinate set at the current time step is sequentially compared with... The similarity between the secondary region template, the boundary coordinate set of the potential impact area, and the potential region template is calculated to further filter rules and determine rule entries that meet the preset minimum threshold for similarity index with the disaster impact range at the current time step. If the selected rule entry is a range expansion scheme, the adjustment range of the expansion boundary is extracted from the scheme, including the expansion range of the core area, the expansion range of the secondary area, and the expansion range of the potential area, and the adjustment method of the hierarchical identifier is also extracted. If the selected rule entry is a range contraction scheme, the adjustment range of the contraction boundary is extracted from the scheme, including the contraction range of the core area, the contraction range of the secondary area, and the contraction range of the potential area, and the adjustment method of the hierarchical identifier is also extracted. The current time step is recorded. Matching rule entries, adjusting parameters, and similarity calculation results; repeating the boundary extraction, similarity calculation, rule filtering, and parameter extraction processes according to the prediction time step order to obtain the warning range adjustment parameters for the region under all prediction time steps; arranging the warning range adjustment parameters under all prediction time steps in chronological order, labeling the disaster impact range boundary coordinate set and rule matching results corresponding to each parameter to form the warning range adjustment parameter sequence for the region; analyzing the parameter change trend in the warning range adjustment parameter sequence, labeling the increasing or decreasing intervals of the parameters to ensure that the parameter changes are consistent with the disaster impact range change trend; integrating the region identifier, warning range adjustment parameter sequence, and trend labeling.

10. An artificial intelligence-based power grid natural disaster early warning system, characterized in that, It includes a processor and a readable storage medium storing a program that, when executed by the processor, implements the artificial intelligence-based power grid natural disaster early warning method according to any one of claims 1-9.

Images