A method for optimizing the scheduling of power supply resources in distribution networks under severe weather conditions

Abstract

This invention relates to the field of power distribution network resource scheduling technology, and discloses a method for optimizing the scheduling of power supply resources in power distribution networks under severe weather conditions. The method includes: acquiring real-time disaster forecast data and equipment topology data to generate a dynamic disaster feature sequence with spatiotemporal labels; inputting the sequence into a spatiotemporal attention neural network to obtain the fault probability distribution of each grid area, retrieving historical similar disaster scenarios and extracting actual fault patterns, and quantifying and fusing the two to generate a dynamic fault risk level map; based on the map and combined with real-time resource status, constructing an optimization function with the objective of minimizing power outage time, and introducing an urgency penalty factor according to the real-time change rate of risk level; using iterative optimization to obtain a resource allocation scheme with multiple time windows and converting it into scheduling instructions for issuance, and tracking execution and disaster evolution in real time for closed-loop feedback. This invention achieves real-time adaptation between resource scheduling and disaster evolution, effectively solving the resource mismatch problem and minimizing power outage time and scope.

Description

Technical Field

[0001] This invention relates to the field of power distribution network resource scheduling technology, and in particular to a method for optimizing the scheduling of power supply resources in power distribution networks under severe weather conditions. Background Technology

[0002] Distribution networks play a crucial role in ensuring reliable power supply for residents, industrial and agricultural production, and users. Extreme weather events such as typhoons, torrential rains, and ice storms pose serious threats to distribution network facilities, often leading to widespread power outages. How to quickly restore power, minimize the scope of outages, and shorten outage time under such extreme conditions, and how to optimize the scheduling of power resources in distribution networks during severe weather, have become one of the most important core research topics in the field of power system emergency management.

[0003] Currently, power resource dispatching for disaster weather mainly relies on two modes: one is a fixed emergency plan based on historical experience, and the other is manual resource allocation by dispatchers based on on-site feedback. These methods can play a role in routine fault handling and predictable maintenance scenarios, but they often prove inadequate when facing real disaster impacts. The reason for this is the high degree of uncertainty in the occurrence and development of disasters—different types, intensities, and movement paths of disasters cause vastly different damage patterns and fault distribution characteristics to the distribution network. Typhoons may cause widespread pole collapses and line breaks, torrential rains may cause flooding leading to equipment submersion, and ice storms may cause lines to become covered in ice and dance. Existing fixed plans or manual dispatching methods are difficult to adjust quickly and accurately according to the real-time evolution of disasters, often resulting in a serious mismatch between resource allocation and actual demand, and the contradiction of insufficient repair forces in severely affected areas while resources are idle in peripheral areas frequently occurs.

[0004] A deeper analysis reveals that disasters themselves are complex, multi-dimensional, and dynamically changing systems. The severity of the disaster directly determines the extent of equipment damage, the affected area determines the number of lines and transformer substations that need to be covered, the duration determines the urgency of fault repair and power restoration, and the development trend further influences the future trajectory of the above three dimensions. These four dimensions are interconnected and dynamically evolve, jointly determining the true scale and priority of power supply resource needs. If these dimensions cannot be comprehensively considered and dynamically quantified as an organic whole, it is impossible to accurately determine how many repair teams, emergency generators, and material reserves are needed at a given moment and in a given area. If the dispatch mechanism continues to operate according to the initial large-scale average allocation method, a resource misallocation dilemma will inevitably arise. This problem of inaccurate resource demand judgment due to the insufficient identification and real-time tracking of the dynamic correlation between the multi-dimensional characteristics of disasters has become the most prominent technical bottleneck in the current power supply dispatch field. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology that relies on fixed plans or human experience for resource scheduling, which cannot be adjusted in real time according to the dynamic changes of multi-dimensional characteristics such as disaster intensity, scope, duration and development trend, resulting in a serious mismatch between resource allocation and actual fault demand. The present invention provides a method for optimizing the scheduling of power distribution network resources under severe weather conditions, which realizes the real-time adaptation of resource scheduling plan to disaster evolution and dynamic tilting towards high-risk areas, effectively solves the problem of resource mismatch, and can minimize the power outage time and scope.

[0006] To address the aforementioned technical problems, this invention provides a method for optimizing the scheduling of power distribution network resources during severe weather, comprising the following steps:

[0007] The system acquires real-time disaster forecast data from meteorological monitoring systems and equipment topology data from power grid geographic information systems, integrates multiple dimensions such as disaster intensity, impact range, duration, and development trend, and generates a dynamic disaster feature sequence with timestamps and spatial coordinates.

[0008] The disaster dynamic feature sequence is input into a neural network based on a spatiotemporal attention mechanism for processing, and the fault probability distribution of each grid area of ​​the distribution network within a preset time period is calculated.

[0009] Based on the disaster dynamic feature sequence, the system retrieves matching historical similar disaster scenarios from the historical fault database and extracts the actual fault distribution patterns corresponding to the historical similar disaster scenarios.

[0010] The fault probability distribution is quantitatively integrated with the actual fault distribution pattern to generate a dynamic fault risk level map of each line and transformer area in the distribution network in the future period.

[0011] Based on the dynamic fault risk level map, and combined with the real-time adjustable status information of the repair team, emergency power generation vehicle and material reserves, a resource scheduling optimization function is constructed with the goal of minimizing power outage time. Among them, the resource scheduling optimization function dynamically introduces the time urgency penalty factor according to the real-time change rate of the fault risk level of each area, so that areas with rapidly rising risk levels receive higher priority in resource allocation.

[0012] The resource scheduling optimization function is solved by iterative optimization. By simulating the resource allocation scheme multiple times to approximate the minimum power outage time, the optimal scheduling scheme for the target deployment location and action sequence of each repair team, emergency power generation vehicle and materials in multiple future time windows is finally obtained.

[0013] The optimal scheduling scheme is converted into scheduling instructions, which are then sent to the corresponding mobile terminals and resource management systems. The execution status of the instructions and the evolution of the disaster are tracked in real time for closed-loop feedback.

[0014] In one embodiment of the present invention, the disaster dynamic feature sequence includes:

[0015] Extract core indicators characterizing disaster intensity, boundary coordinates of the forecasted affected area, expected duration, and forecast path or evolution direction from disaster forecast data released by meteorological monitoring systems; extract line type, tower density, and equipment commissioning years from power grid geographic information systems as basic data on equipment vulnerability.

[0016] Spatially aligned according to the same grid number, and normalized to map the numerical range to the interval of 0 to 1 for various types of data; spliced ​​in chronological order to form a feature vector sequence containing timestamp, grid number, intensity normalized value, influence range marker, duration, evolution direction angle, moving speed, line type code, tower density normalized value, and commissioning year normalized value.

[0017] In one embodiment of the present invention, the neural network based on the spatiotemporal attention mechanism includes: an input layer, a spatial attention layer, a temporal attention layer, and an output layer, wherein:

[0018] The input layer receives the dynamic feature sequence of the disaster;

[0019] The spatial attention layer calculates the spatial attention weights between each grid region and all other grid regions. The weights are calculated using the softmax function based on the Euclidean distance and geographic distance between the feature vectors of the two regions. This layer is used to capture the mutual transmission relationship of disaster impacts between different regions.

[0020] The time attention layer calculates the time attention weights between each time window and multiple time windows before and after it. The weights are calculated using the softmax function based on the rate of change of the feature vector over time, and are used to capture the evolution trend of disaster intensity.

[0021] The spatial attention weights and temporal attention weights are multiplied by the original features and then added together to obtain the enhanced feature representation, which is then input into the output layer to calculate the fault probability of each grid region.

[0022] In one embodiment of the present invention, retrieving matching historical similar disaster scenarios from a historical fault database based on a disaster dynamic feature sequence specifically includes:

[0023] Calculate the cosine similarity between the current dynamic feature sequence of disasters and the feature sequence of each historical disaster record in the historical fault database;

[0024] The cosine similarity scores were sorted from largest to smallest, and the top three historical disaster records with the highest similarity scores were selected as the matching historical similar disaster scenarios.

[0025] Extract the location, type, and number of line faults that actually occurred within 24 hours after the disaster from these three historical disaster records to obtain the actual fault distribution pattern.

[0026] In one embodiment of the present invention, generating a dynamic fault risk level map specifically includes:

[0027] Sort all grid regions in descending order of their failure probability distribution values ​​to obtain the failure probability distribution value for each grid region.

[0028] Sort all grid regions from largest to smallest according to the total number of failures in each historical similar disaster scenario to obtain the historical experience risk value corresponding to each grid region;

[0029] The fault probability distribution value of each grid area is added to the historical risk value to obtain the comprehensive risk value of the grid area;

[0030] According to the comprehensive risk value from largest to smallest, the lines and transformer areas corresponding to the top 10% of the grid areas are marked as high-risk, those between 10% and 30% are marked as medium-risk, and those in the bottom 70% are marked as low-risk.

[0031] On the geographic information system map of the power distribution network, high-risk areas are rendered in red, medium-risk areas in yellow, and low-risk areas in green, forming a visualized dynamic fault risk level map.

[0032] In one embodiment of the present invention, the real-time adjustable status information of the repair team, emergency power generation vehicle, and material reserves includes:

[0033] The vehicle-mounted GPS terminal collects the current latitude and longitude coordinates of each repair team and each emergency power generation vehicle in real time; the electronic ledger of the resource management system obtains the current warehouse location, material type and available quantity of each batch of materials; the location information is associated with the dynamic fault risk level map using the same grid numbering system to establish a mapping relationship between resource location and risk area.

[0034] In one embodiment of the present invention, a resource scheduling optimization function is constructed with the objective of minimizing power outage time, including:

[0035] The objective is to minimize the sum of the estimated outage times for all fault areas, where the estimated outage time for each fault area is equal to the travel time of the repair team from its current location to the fault area plus the estimated repair time. The constraints of the resource scheduling optimization function include: each repair team can only be dispatched to one fault area at a time, each emergency generator can only support one transformer substation at a time, and the amount of materials transferred from each warehouse cannot exceed the available inventory of the warehouse.

[0036] In one embodiment of the present invention, the resource scheduling optimization function dynamically introduces a time urgency penalty factor based on the real-time change rate of the fault risk level of each region, including:

[0037] The comprehensive risk value of each grid area is calculated periodically according to a preset time interval, and the change in risk value between adjacent time intervals is recorded.

[0038] When the risk value of a certain grid area shows a positive increase over multiple consecutive time intervals, and the cumulative increase exceeds 10% of the original comprehensive risk value, the area is marked as an area of ​​accelerated risk deterioration.

[0039] When allocating resources to areas where risks are rapidly deteriorating, an urgency multiplier greater than 1 is introduced into the calculation of their resource needs. The specific value is positively correlated with the cumulative growth rate of the risk value; the greater the growth rate, the higher the value of the urgency multiplier.

[0040] By introducing a urgency multiplier, the optimization function tends to allocate available resources preferentially to regions where the risk is accelerating during the solution process.

[0041] In one embodiment of the present invention, the resource scheduling optimization function is solved using an iterative optimization method, including:

[0042] First, an initial solution is generated, which means randomly assigning a target area to each repair team and each emergency power generation vehicle;

[0043] Fine-tune the current solution by swapping the target areas of the two teams or changing the drop location of a certain material, and calculate the weighted power outage time corresponding to the new solution;

[0044] If the power outage time of the new solution is shorter than that of the original solution, the new solution is accepted; otherwise, the new solution is accepted with a certain probability.

[0045] Repeat the above fine-tuning and comparison process until the power outage time no longer decreases after multiple iterations. Then output the current optimal solution as the final scheduling scheme.

[0046] In one embodiment of the present invention, real-time tracking of instruction execution status and disaster evolution for closed-loop feedback includes:

[0047] The system monitors whether each scheduling instruction has been executed by uploading confirmation information through mobile terminals. If no execution confirmation is received within 30 minutes after an instruction is issued, the resource status is marked as abnormal and it re-participates in the next round of scheduling calculation.

[0048] The system reacquires the latest meteorological monitoring data every 10 minutes, compares the newly generated optimal scheduling plan with the previous plan, and only sends the changed instructions to the resource terminal, while keeping the unchanged instructions running.

[0049] The technical solution of the present invention has the following advantages compared with the prior art:

[0050] The power distribution network resource optimization scheduling method described in this invention generates a refined dynamic fault risk map by real-time fusion and dynamic analysis of multi-dimensional disaster characteristics, combined with dual-path fusion of deep neural network prediction and historical fault mode matching. Furthermore, it introduces an urgency weight mechanism based on the risk change rate in resource scheduling optimization, enabling the scheduling scheme to adapt in real time as the disaster evolves. This achieves dynamic tilting of resources towards areas with rapidly rising risks, effectively solving the problem of resource mismatch and minimizing power outage time and scope. Attached Figure Description

[0051] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:

[0052] Figure 1 This is a flowchart illustrating the steps of the method for optimizing the scheduling of power supply resources in a power distribution network under severe weather conditions, as described in this invention.

[0053] Figure 2 This is a flowchart illustrating the steps involved in constructing the dynamic disaster feature sequence of this invention.

[0054] Figure 3 This is a flowchart illustrating the steps involved in constructing a spatiotemporal attention neural network according to the present invention.

[0055] Figure 4 This is a flowchart of the steps for searching and matching historical similar disaster scenarios in this invention;

[0056] Figure 5 This is a flowchart of the steps involved in generating the dynamic fault risk level map of this invention.

[0057] Figure 6 This is a flowchart illustrating the steps involved in constructing and solving the resource scheduling optimization function of this invention.

[0058] Figure 7 This is a flowchart illustrating the steps involved in introducing a time urgency penalty factor in this invention.

[0059] Figure 8 This is a flowchart of the closed-loop feedback and instruction update steps of the present invention. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0061] Reference Figure 1 As shown, this invention proposes a method for optimizing the scheduling of power distribution network resources under severe weather conditions. First, real-time disaster forecast data released by the meteorological monitoring system is acquired, and simultaneously, equipment topology data from the power grid geographic information system is accessed. The four core dimensions of disaster intensity, impact range, duration, and development trend are deeply integrated to generate a dynamic disaster feature sequence with precise timestamps and spatial coordinates. This dynamic disaster feature sequence organically integrates previously isolated meteorological and power grid data, laying a data foundation for subsequent precise analysis.

[0062] After obtaining the dynamic feature sequence of disasters, this invention employs a two-pronged parallel analysis approach to assess future fault risks. The first approach is intelligent prediction based on deep learning: the dynamic feature sequence of disasters is input into a neural network based on a spatiotemporal attention mechanism, which automatically captures the dynamic correlation between disaster characteristics in different regions at different points in time, thereby calculating the fault probability distribution of each grid area of ​​the distribution network within a specific future period. This spatiotemporal attention mechanism allows the neural network to focus on areas that are rapidly deteriorating, capturing the dynamic trend of disaster evolution, rather than simply making a static assessment. The second approach is pattern matching based on historical experience: the current dynamic feature sequence of disasters is simultaneously retrieved and compared with a historical fault database to identify historical disaster scenarios most similar to the current disaster characteristics, and the distribution patterns of actual faults in these scenarios are extracted. By mining real fault data under similar scenarios, empirical references can be provided for the current assessment.

[0063] Next, this invention organically integrates the above two analysis results—quantitatively fusing the intelligently predicted fault probability distribution with the historically matched actual fault distribution patterns to generate a dynamic fault risk level map for each line and transformer substation in the distribution network for future periods. This dynamic fault risk level map can be updated in real time as the disaster evolves, clearly indicating which areas and lines are most likely to experience faults in the future, and what the risk level is.

[0064] Based on a dynamic risk map, the system acquires the real-time location and status information of all dispatchable repair teams, emergency power generation vehicles, and material reserves. Then, with the goal of minimizing power outage time, a resource scheduling optimization function is constructed. This function dynamically incorporates a time urgency penalty factor based on the real-time rate of change of fault risk levels in each region. Simply put, if the fault risk level in a certain region is rapidly increasing (e.g., a sudden intensification of a typhoon causing a sharp rise in risk in a certain area), the system automatically adds an urgency weight to this region, causing the optimization solution to prioritize resource allocation to that region. This mechanism effectively solves the problem of resource response lagging behind disaster changes in traditional methods, achieving dynamic adaptation of resource scheduling to disaster evolution.

[0065] Finally, this invention employs an iterative optimization approach to solve the aforementioned resource scheduling optimization function. By simulating different resource allocation schemes multiple times, it continuously approaches the optimal solution that minimizes the power outage time. Ultimately, it outputs the target deployment locations and action sequences for each repair team, each emergency power generation vehicle, and each batch of supplies within multiple future time windows. This scheduling scheme is then converted into specific scheduling instructions, issued to the corresponding mobile terminals and resource management systems, and continuously tracks the instruction execution status and the actual evolution of the disaster, forming a closed-loop system of perception-prediction-decision-execution-feedback to ensure that the scheduling scheme can be continuously optimized and adjusted according to the actual situation.

[0066] The beneficial effects of this invention through the above technical solutions include: First, by integrating multiple dimensions of disaster intensity, scope, duration, and development trend, it breaks through the limitations of traditional methods that rely solely on a single dimension or static data for judgment, making the understanding of disaster impact more comprehensive and accurate. Second, the dual-path prediction method combining a spatiotemporal attention mechanism neural network with historical fault pattern matching leverages the advantages of deep learning in capturing dynamic trends while incorporating real fault patterns from historical experience. These two approaches mutually reinforce and complement each other, significantly improving the accuracy and reliability of fault risk prediction. Third, the introduction of an urgency penalty factor based on the risk change rate in resource scheduling optimization enables the scheduling scheme to have dynamic perception and real-time response capabilities, effectively avoiding resource misallocation caused by sudden disaster changes and ensuring that resources always flow to the most urgently needed areas. Finally, the entire method forms a complete closed loop from data collection, risk prediction, optimization decision-making to command execution and effect feedback, allowing the scheduling scheme to continuously optimize and iterate as the disaster evolves and command execution progresses. This truly achieves dynamic and precise scheduling of power distribution network resources under severe weather conditions, providing solid technical support for minimizing power outage time and reducing the scope of power outages.

[0067] During the data acquisition process, in order to ensure that the constructed disaster dynamic feature sequence can comprehensively and accurately capture the multidimensional dynamic information of the disaster and the inherent vulnerability of power grid equipment, reference was made to... Figure 2 As shown, this application proposes a method for constructing disaster dynamic feature sequences, specifically including:

[0068] Extracting core indicators characterizing disaster intensity, boundary coordinates of the forecasted affected area, expected duration, and forecast path or evolution direction from disaster forecast data released by meteorological monitoring systems aims to comprehensively capture the dynamic characteristics of disaster events. For example, for typhoon disasters, core indicators could be maximum wind speed or wind force level; for snow and ice disasters, they could be snow depth or ice thickness. The boundary coordinates of the forecasted affected area precisely define the geographical range that the disaster may affect, providing a spatial basis for subsequent grid division and risk assessment. The expected duration provides the temporal dimension information of the disaster event, helping to assess its long-term impact. The forecast path or evolution direction reveals the movement trend of the disaster, which is crucial for predicting future risk areas.

[0069] Meanwhile, extracting line types, tower density, and equipment service life within each grid area from the power grid geographic information system serves as fundamental data for equipment vulnerability. This is to quantify the inherent vulnerability of distribution network equipment in the face of disasters. Line type (such as overhead lines, cables, and lines of different voltage levels) determines its disaster resistance capability; tower density reflects the exposure degree and damage risk of overhead lines; and equipment service life serves as an indicator of equipment aging, generally considered to have a higher failure rate and weaker disaster resistance capability with longer service life. This data provides important static background information for assessing the potential failure risk of power grid equipment under disasters.

[0070] To ensure that data from different sources and with different dimensions can be processed and analyzed uniformly, this application spatially aligns the extracted disaster dynamics information and equipment vulnerability data according to the same grid number. This means mapping all data to predefined geographic grid cells, ensuring that each grid cell contains its corresponding disaster characteristics and power grid equipment vulnerability information. Based on this, various types of data are normalized to map the numerical range to the interval between 0 and 1. This processing method eliminates the dimensional differences between different features, preventing certain features with large numerical ranges from dominating subsequent model training, thereby ensuring the fairness of data input and the convergence of the model.

[0071] Finally, the spatially aligned and normalized data are concatenated in chronological order to form a feature vector sequence containing timestamps, grid numbers, intensity normalized values, impact range markers, duration, evolution direction angle, movement speed, line type codes, tower density normalized values, and commissioning year normalized values. This sequence not only contains dynamic evolution information of the disaster event at different points in time but also integrates the inherent vulnerability of power grid equipment within each geographical grid, forming a multi-dimensional, spatiotemporally correlated comprehensive feature representation. This provides rich and structured input for subsequent neural networks based on spatiotemporal attention mechanisms.

[0072] After obtaining the dynamic feature sequence of the disaster, this invention adopts two parallel analysis paths to assess the future failure risk, including: intelligent prediction based on deep learning and pattern matching based on historical experience. The two assessment processes are described below.

[0073] Reference Figure 3 As shown, this application proposes a neural network based on a spatiotemporal attention mechanism, comprising an input layer, a spatial attention layer, a temporal attention layer, and an output layer. This neural network architecture based on spatiotemporal attention mechanism aims to effectively process data with temporal and spatial dependencies. By introducing an attention mechanism, it enables the model to dynamically focus on the more important parts of the input data in the temporal and spatial dimensions, thereby more accurately understanding the evolution and impact of disasters. As the starting part of the neural network, the input layer is responsible for receiving the preprocessed disaster dynamic feature sequence. This sequence is usually structured data, containing various feature information of the disaster at different time points and spatial grids. The input layer may perform preliminary linear transformations or embedding operations on these raw features to adapt them to the processing of subsequent attention layers.

[0074] The spatial attention layer functions to identify and quantify the interrelationships of disaster impacts between different grid areas in a power distribution network. Specifically, for any target grid area, the spatial attention layer calculates its spatial attention weights with all other grid areas. These weights are calculated based on the Euclidean distance between the feature vectors of each area and their geographic distances, and are normalized using a softmax function. The Euclidean distance reflects the similarity at the feature level, while the geographic distance reflects physical proximity or connectivity. Combining the two allows for a comprehensive assessment of the spatial propagation and mutual influence intensity of disasters.

[0075] The temporal attention layer focuses on capturing the dynamic trend of disaster intensity changing over time. It calculates temporal attention weights for each specific time window relative to multiple preceding and subsequent time windows. These weights are calculated based on the rate of change of the feature vector along the time axis and are also normalized using the softmax function. In this way, the temporal attention layer can highlight key time points where disaster intensity changes rapidly, or identify the persistence or periodicity of disaster impacts.

[0076] After calculating the attention weights for the spatial and temporal attention layers, these weights are used to weight the original disaster dynamic features. Specifically, the spatial attention weights are multiplied by the original features, and the temporal attention weights are also multiplied by the original features. These two weighted feature representations are then summed. This fusion mechanism ensures that the final feature representation not only includes the original disaster information but also incorporates the spatial correlations and temporal evolution trends learned by the model, resulting in a richer and more insightful feature vector. Finally, the output layer receives the feature representation enhanced by the spatial and temporal attention mechanisms. This layer typically consists of one or more fully connected layers, and its task is to map the high-dimensional enhanced features to the final prediction result, i.e., the failure probability of each grid area in the distribution network. The output layer may use a sigmoid activation function to output a probability value between 0 and 1, representing the likelihood of a failure occurring in each grid area within a predetermined future time period.

[0077] Reference Figure 4 As shown, this application also proposes a specific method for retrieving matching historical similar disaster scenarios from a historical fault database based on disaster dynamic feature sequences. This method includes calculating the cosine similarity between the current disaster dynamic feature sequence and the feature sequence of each historical disaster record in the historical fault database. Cosine similarity is a measure of the directional difference between two non-zero vectors, with a value between -1 and 1. The closer the value is to 1, the more consistent the directions of the two vectors, i.e., the higher the similarity. Here, both the current disaster dynamic feature sequence and the feature sequences of historical disaster records can be considered as multi-dimensional vectors. The dimensions of these vectors can include key indicators such as disaster intensity, impact range, duration, evolution direction, movement speed, and vulnerability of power grid equipment. By calculating the cosine similarity, the similarity between the current disaster scenario and historical disaster events in multi-dimensional features can be quantified, thus providing an objective basis for subsequent similar scenario selection. In specific implementation, the disaster dynamic feature sequence and the feature sequences of historical disaster records are first vectorized, and then the standard cosine similarity calculation formula is applied, i.e., the dot product of the two vectors divided by the product of their respective moduli.

[0078] The cosine similarity scores are sorted from highest to lowest, and the top three historical disaster records with the highest similarity scores are selected as the matching historical similar disaster scenarios. After calculating the cosine similarity between the current disaster dynamic feature sequence and all historical disaster records in the historical fault database, these similarity values ​​need to be sorted in descending order. This step aims to identify the historical events most similar to the current disaster scenario. Selecting the top three historical disaster records as the matching historical similar disaster scenarios is based on experience and practice, aiming to balance retrieval efficiency and the representativeness of the results. Selecting the top three ensures that a sufficient number of similar scenarios are covered, avoiding the randomness of a single historical event, while also avoiding the introduction of too much irrelevant historical data, thereby improving the accuracy and reliability of subsequent fault distribution pattern extraction. Of course, in practical applications, the number of selected scenarios or the similarity threshold can be flexibly adjusted according to specific needs and the amount of historical data.

[0079] The actual location, type, and number of line faults occurring within 24 hours of the disaster in these three historical disaster records are extracted to represent the actual fault distribution pattern. After identifying the top three historically similar disaster scenarios, specific fault information needs to be extracted from these historical records. Here, the actual location of the line fault refers to the specific line segment, transformer substation, or equipment node where the fault occurred; the fault type refers to the nature of the fault, such as short circuit, grounding, or open circuit; and the number of faults refers to the number of fault events occurring at a specific location or type. Limiting the extraction of fault data to within 24 hours of the disaster focuses on the direct impact of the initial and peak periods of the disaster, which are typically the times when the distribution network is most severely damaged and requires emergency response. By extracting this specific, quantified fault data, an actual fault distribution pattern reflecting the actual vulnerability and fault modes of the distribution network under similar historical disaster scenarios can be formed.

[0080] Furthermore, after analyzing the failure probability distribution and the actual failure distribution patterns, in order to assess the failure risk, reference is made to... Figure 5 As shown, this application proposes a specific method for generating a dynamic fault risk level map, including: sorting all grid regions according to their fault probability distribution values ​​from largest to smallest, obtaining the fault probability distribution value corresponding to each grid region. This step aims to quantify the probability of each grid region experiencing a fault within a preset future time period, and the value is derived from the calculation results of a neural network based on a spatiotemporal attention mechanism. By sorting, it is possible to intuitively understand which regions are considered to have a higher fault risk in the prediction model, thereby providing a basic predictive basis for subsequent risk assessment.

[0081] Simultaneously, all grid areas are sorted from highest to lowest based on the total number of failures in various historical similar disaster scenarios, yielding a historical empirical risk value for each grid area. This step introduces historical data as a crucial component of risk assessment, reflecting the actual vulnerability and failure frequency of different grid areas under similar disaster conditions. By statistically analyzing the total number of historical failures, areas exhibiting higher failure rates in past disasters can be identified, providing empirical, operationally-based support for risk assessment.

[0082] Based on this, the fault probability distribution value of each grid area is added to the historical experience risk value to obtain the comprehensive risk value of that grid area. This step achieves a quantitative fusion of predictive and empirical risks. The comprehensive risk value can more comprehensively reflect the potential fault risks of each grid area, taking into account both the prediction results under current disaster forecasts and power grid topology data, as well as the actual operating experience accumulated in historical disaster scenarios, making the risk assessment results more robust and reliable.

[0083] Furthermore, based on the comprehensive risk value from highest to lowest, the lines and transformer substations corresponding to the top 10% of grid areas are marked as high-risk, those between 10% and 30% are marked as medium-risk, and the bottom 70% are marked as low-risk. This step discretizes the continuous comprehensive risk value into clear risk levels, facilitating quick understanding and decision-making by dispatchers. By setting different percentage thresholds, the distribution network is divided into high, medium, and low-risk zones, allowing risk management strategies to be differentiated for different zones. For example, high-risk zones prioritize resource allocation, medium-risk zones require early warning and preparation, and low-risk zones maintain routine monitoring. These percentage thresholds can be adjusted based on actual operational experience and risk tolerance.

[0084] Finally, on the distribution network geographic information system map, high-risk areas are rendered in red, medium-risk areas in yellow, and low-risk areas in green, forming a visualized dynamic fault risk level map. This step provides an intuitive and real-time risk visualization interface. By presenting areas of different risk levels on the geographic information system map using standardized colors (e.g., red for high risk, yellow for medium risk, and green for low risk), dispatchers can clearly grasp the risk distribution of the entire distribution network, quickly identify areas requiring key attention and priority handling, and greatly improve situational awareness and decision-making efficiency.

[0085] Reference Figure 6As shown, based on the fault risk level map constructed above, this embodiment further discloses the process of constructing and solving the resource scheduling optimization function. First, it is necessary to take stock of the available emergency resources in real time. This application further proposes real-time adjustable status information of repair teams, emergency power generation vehicles, and material reserves, including: real-time collection of the current latitude and longitude coordinates of each repair team and each emergency power generation vehicle through the vehicle-mounted global positioning system terminal; obtaining the current warehouse location, material type, and available quantity of each batch of materials through the electronic ledger of the resource management system; and associating the location information with the dynamic fault risk level map using the same grid numbering system to establish a mapping relationship between resource location and risk area.

[0086] Specifically, real-time adjustable status information is the foundation for optimized resource scheduling. It encompasses all physical resources available for disaster response, including manpower (repair teams), critical equipment (emergency power generation vehicles), and consumables (material reserves). Real-time adjustability emphasizes that this information must be currently valid, and that the resources themselves must be capable of adjusting according to dispatch instructions. Accurately acquiring this real-time status is a prerequisite for ensuring the feasibility and effectiveness of the dispatch plan, avoiding the dispatch of resources to inaccessible or uncapable areas. To achieve this goal, vehicle-mounted GPS terminals are used to collect the current latitude and longitude coordinates of each repair team and each emergency power generation vehicle in real time. A vehicle-mounted GPS terminal is a device that uses satellite signals for positioning. Installed on the vehicles of repair teams or emergency power generation vehicles, it can continuously and automatically acquire the precise geographical location information, i.e., latitude and longitude coordinates, of these mobile resources. Real-time acquisition means that the system can update this location data at a preset frequency, ensuring that the dispatch center always has the latest resource dynamics, thus achieving high accuracy in calculating resource arrival times. Simultaneously, the electronic ledger of the resource management system obtains the current warehouse location, material type, and available quantity of each batch of materials. A resource management system is a software platform used to manage various resources within an enterprise or organization. Its electronic ledger function is a digital record system used to meticulously register and track information such as the receipt, issuance, inventory levels, types, and storage locations of materials. Through this system, users can easily find out which warehouse holds a specific material, the exact available quantity, and the warehouse's geographical location. This ensures the accuracy and timeliness of material allocation, preventing shortages or waste caused by information delays or errors.

[0087] Based on this, the obtained location information is associated with the dynamic fault risk level map using the same grid numbering system to establish a mapping relationship between resource locations and risk areas. The grid numbering system is a method of dividing a geographical area into regular grids and assigning a unique identifier to each grid. By adopting the same grid numbering system as the dynamic fault risk level map, the latitude and longitude coordinates of repair teams, emergency power generation vehicles, and material warehouses can be converted into corresponding grid numbers. This unified geographic reference system allows for seamless integration of resource location information with the geographic information of risk areas, thereby establishing a direct mapping relationship between resources and risk areas. This mapping relationship is crucial for the optimization scheduling algorithm to quickly determine which resources are closest to which high-risk areas and which resources can effectively cover which risk areas.

[0088] Secondly, the aforementioned emergency resources need to be allocated rationally. This application proposes a resource scheduling optimization function aimed at minimizing power outage time. Specifically, it aims to minimize the sum of the estimated power outage times for all affected areas through resource scheduling. This objective concretizes the abstract goal of minimizing power outage time into a calculable value, providing a clear direction for the optimization algorithm. Specifically, for each potential affected area, the estimated power outage time can be decomposed into the travel time required for the repair team to reach the affected area from its current location, and the estimated repair time required for troubleshooting and repair after arrival. Travel time can be estimated using a geographic information system combined with real-time traffic conditions, road capacity, and road damage data in the disaster-affected area. For example, the shortest path algorithm can be used to calculate the shortest travel time from the repair team's current location to the affected area. The estimated repair time can be predicted based on historical fault data, fault type, equipment damage level, required materials and quantities, and the repair team's professional skills. For example, the repair time will vary significantly for different types of line faults. The optimization function sums up the estimated power outage times for all fault areas assigned to emergency repair tasks, and uses the minimum of this sum as the optimization objective.

[0089] The constraints of the resource scheduling optimization function include the following: each repair team can only be dispatched to one fault area at a time; each emergency power generator can only support one transformer substation at a time; and the amount of materials transferred from each warehouse cannot exceed the available inventory of that warehouse. These constraints ensure that the scheduling scheme generated during the solution process conforms to actual operational limitations and the physical properties of resources. They reflect the exclusivity, capacity limitations, and availability of resources, preventing the generation of unrealistic or unexecutable scheduling instructions. For example, the exclusivity constraint of repair teams ensures that within any given time window, a repair team can only be assigned to one and only one fault area for repair tasks. This can be achieved through binary variables or unique allocation rules in integer programming. Similarly, the exclusivity constraint of emergency power generators ensures that each emergency power generator can only provide power support to a designated transformer substation at a time. For the material inventory constraint, it is necessary to obtain the material ledger information of the warehouses in real time and use it as an upper limit constraint in the optimization model to ensure that the total amount of materials transferred from each warehouse cannot exceed the current available inventory of that warehouse.

[0090] Finally, the resource scheduling optimization function needs to be solved. In this embodiment, iterative solution is adopted, which includes: first, generating an initial solution, that is, randomly assigning a target area to each repair team and each emergency power generation vehicle; fine-tuning the current solution by swapping the target areas of two teams or changing the placement location of a certain material, and calculating the weighted outage time corresponding to the new solution; if the outage time of the new solution is lower than that of the original solution, the new solution is accepted; otherwise, the new solution is accepted with a certain probability; repeat the above fine-tuning and comparison process until the outage time no longer decreases after multiple iterations, and output the current optimal solution as the final scheduling scheme.

[0091] Specifically, in the initial solution generation step, the system randomly assigns a target area or deployment location to each repair team, each emergency power generation vehicle, and various supplies. For example, for repair teams and emergency power generation vehicles, the target area could be a high-risk or medium-risk grid area identified in the dynamic fault risk level map; for supplies, the deployment location could be a temporary assembly point near a high-risk area or a damaged power station area. This random initialization strategy aims to provide a diverse starting point for subsequent iterative optimization, preventing the algorithm from prematurely getting trapped in local optima.

[0092] In the fine-tuning step of the current solution, the system makes local adjustments based on the current resource scheduling scheme. For example, it may randomly select two repair teams and exchange their target areas, or randomly select a material and change its deployment location. The purpose of the fine-tuning operation is to explore new solutions within the neighborhood of the current solution. After each fine-tuning, the system calculates the weighted outage time corresponding to the new solution based on the resource scheduling optimization function. This weighted outage time not only considers the expected outage duration but also incorporates a time urgency penalty factor to reflect the real-time rate of change of fault risk in different areas.

[0093] In the step of accepting a new solution, the system compares the weighted downtime of the new solution with that of the original solution. If the weighted downtime of the new solution is shorter than that of the original solution, it indicates that the new solution is better, and the new solution is accepted unconditionally and adopted as the new current solution. If the weighted downtime of the new solution is longer than that of the original solution, meaning that the new solution is inferior to the original solution, the system will still accept the new solution with a certain probability. This probabilistic acceptance mechanism is a characteristic of heuristic algorithms such as simulated annealing, designed to help the algorithm escape local optima and continue exploring a wider solution space, thereby increasing the probability of finding the global optimum. This acceptance probability usually gradually decreases with the increase of the number of iterations to ensure that the algorithm eventually converges.

[0094] During the repeated fine-tuning and comparison process, the aforementioned cyclical process of fine-tuning and accepting / rejecting new solutions continues. The system continuously monitors whether the weighted outage time of the optimal solution decreases. When, after a preset number of consecutive iterations, the weighted outage time of the optimal solution no longer decreases significantly, it indicates that the algorithm has converged, and the currently found optimal scheduling scheme is output as the final scheduling scheme.

[0095] In practical applications, risk levels are constantly changing. Accurately identifying and quantifying rapidly escalating risk levels, and effectively mapping this dynamic change to resource allocation priorities, are key challenges in ensuring timely responses to evolving disasters. Failure to effectively distinguish between the static level of risk and its dynamic deterioration trend may result in resource allocation failing to adequately reflect priority responses to areas experiencing urgent deterioration.

[0096] In this regard, refer to Figure 7As shown, this application proposes a specific implementation method for dynamically introducing a time urgency penalty factor into the resource scheduling optimization function based on the real-time change rate of the fault risk level of each region. This method includes periodically calculating the comprehensive risk value of each grid region at preset time intervals and recording the change in risk value between adjacent time intervals. This process aims to achieve continuous dynamic monitoring of the risk status of each grid region in the distribution network. During the evolution of severe weather, the risk level of each region is not static but changes in real time with factors such as meteorological conditions and equipment status. The system can set a fixed time interval, such as every 5 minutes, 10 minutes, or 15 minutes, to re-execute the step of generating a dynamic fault risk level map, thereby obtaining the comprehensive risk value of each grid region at the current moment. After each calculation, the comprehensive risk value at the current moment is compared with the comprehensive risk value calculated in the previous time interval, and the difference between the two is calculated, which is the change in risk value. These changes are recorded to form a time series for subsequent trend analysis.

[0097] Furthermore, when the risk value of a grid area shows a positive increase over multiple consecutive time intervals, and the cumulative increase exceeds 10% of the original comprehensive risk value, the area is marked as a region of accelerated risk deterioration. This feature defines the identification criteria for "regions of accelerated risk deterioration," aiming to filter out areas from numerous risk areas where the risk level is rapidly and continuously rising; these areas typically require more urgent resource intervention. The system continuously monitors the risk value change sequence of each grid area. When it detects that the risk value of a grid area shows a positive increase over N consecutive time intervals, and the cumulative increase of the risk value over these N time intervals exceeds 10% of the comprehensive risk value of the area before those N time intervals, the grid area is automatically marked as a region of accelerated risk deterioration. This mechanism ensures that only areas where the risk is indeed deteriorating rapidly and significantly are identified.

[0098] Building upon this, when allocating resources to areas where risk is rapidly deteriorating, an urgency multiplier greater than 1 is introduced into the calculation of their resource requirements. The specific value of this multiplier is positively correlated with the cumulative increase in risk value; the larger the increase, the higher the urgency multiplier. This feature is the core mechanism for prioritizing resource allocation to areas where risk is rapidly deteriorating, quantitatively increasing the weight of these areas in the resource scheduling optimization function. When a grid area is marked as an area where risk is rapidly deteriorating, an urgency multiplier is introduced for the resource requirements of that area when constructing the resource scheduling optimization function. This multiplier is a value greater than 1, and its specific magnitude is proportional to the cumulative increase in risk value of the area. For example, a linear mapping relationship can be established, such as: Urgency Multiplier = 1 + k * (Cumulative Increase / Original Overall Risk Value), where k is an adjustment coefficient. This means that the faster the risk deteriorates in an area, the larger its urgency multiplier, thus assigning a higher priority to the resource requirements of that area in the optimization function.

[0099] By introducing an urgency multiplier, the optimization function tends to prioritize allocating available resources to areas where the risk is accelerating. Resource scheduling optimization functions typically aim to minimize the total outage time. With the urgency multiplier, the resource demands of areas with accelerating risk are amplified. This causes the optimization function to consider that failing to respond to these areas in a timely manner will result in a greater penalty (i.e., a higher equivalent outage time) when calculating the total outage time. Therefore, during the iterative optimization process, the optimization algorithm naturally tends to prioritize meeting the demands of these areas with amplified accelerating risks, thereby prioritizing the allocation of available resources such as repair teams, emergency generators, and supplies to these areas to achieve the goal of minimizing the overall outage time.

[0100] In actual disaster response, disaster situations can evolve rapidly, and resource execution instructions may also deviate or be delayed. Without an effective real-time tracking and feedback mechanism, issued dispatch instructions may fail to adapt to the new situation, leading to reduced resource allocation efficiency and even unnecessary extensions of power outage time.

[0101] In this regard, refer to Figure 8 As shown. This application proposes a method for real-time tracking of command execution status and disaster evolution to achieve closed-loop feedback. Specifically, the method includes the following steps: monitoring whether each dispatch command has been executed via confirmation information uploaded by a mobile terminal. This means that after receiving a dispatch command, the repair team or emergency power generation vehicle sends confirmation information with a timestamp and status identifier to the resource management system via its equipped mobile terminal. This confirmation information can include feedback on key execution nodes such as command reception, departure, arrival at the target location, and task completion, to ensure that the system can monitor the execution progress of the commands in real time.

[0102] If no execution confirmation is received within 30 minutes of an instruction being issued, the resource's status is marked as abnormal and it re-enters the next round of scheduling calculations. This means the resource management system sets an expected response time window for each issued instruction, such as 30 minutes. Within this time window, if no execution confirmation is received from the corresponding mobile terminal, the system will automatically determine that the instruction execution is abnormal and update the status of the involved resources to abnormal or unavailable. Subsequently, the abnormal resource will be excluded from subsequent resource scheduling optimization calculations to avoid including resources that cannot perform tasks in new scheduling decisions, thereby ensuring the effectiveness and feasibility of the scheduling scheme.

[0103] Furthermore, the system periodically and automatically retrieves the latest meteorological monitoring data every 10 minutes, including disaster intensity, impact range, duration, and development trend. This latest data serves as input, driving the entire optimized scheduling process to re-run and generate the optimal scheduling plan based on the current disaster situation. Comparing the newly generated optimal scheduling plan with the previous plan means that, driven by new meteorological data, the system recalculates and generates a completely new optimal scheduling plan. To minimize interference with on-site personnel, the system performs a detailed comparison between this new plan and the scheduling plan generated in the previous time window, identifying which instructions have changed. Only sending changed instructions to resource terminals while maintaining unchanged instructions means that, based on the comparison results, the system only reissues scheduling instructions that have changed compared to the previous plan to the corresponding mobile terminals. Instructions that remain unchanged in both the old and new plans are not reissued but are allowed to continue execution by default, thus reducing unnecessary communication burdens and frequent operational requirements for on-site personnel.

[0104] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for optimizing the scheduling of power supply resources in a distribution network under severe weather conditions, characterized in that, Includes the following steps: The system acquires real-time disaster forecast data from meteorological monitoring systems and equipment topology data from power grid geographic information systems, integrates multiple dimensions such as disaster intensity, impact range, duration, and development trend, and generates a dynamic disaster feature sequence with timestamps and spatial coordinates. The disaster dynamic feature sequence is input into a neural network based on a spatiotemporal attention mechanism for processing, and the fault probability distribution of each grid area of ​​the distribution network within a preset time period is calculated. The neural network based on the spatiotemporal attention mechanism includes an input layer, a spatial attention layer, a temporal attention layer, and an output layer. The input layer receives a sequence of dynamic disaster features. The spatial attention layer calculates spatial attention weights for each grid region relative to all other grid regions, based on the Euclidean distance and geographic distance between the feature vectors of the two regions using a softmax function. This weight is used to capture the mutual transmission relationship of disaster impacts between different regions. The temporal attention layer calculates temporal attention weights for each time window relative to multiple preceding and following time windows, based on the rate of change of the feature vector over time using a softmax function. This weight is used to capture the evolution trend of disaster intensity. The spatial attention weights and temporal attention weights are multiplied by the original features and then summed to obtain an enhanced feature representation, which is input to the output layer to calculate the failure probability of each grid region. Based on the disaster dynamic feature sequence, the system retrieves matching historical similar disaster scenarios from the historical fault database and extracts the actual fault distribution patterns corresponding to the historical similar disaster scenarios. The fault probability distribution is quantitatively integrated with the actual fault distribution pattern to generate a dynamic fault risk level map of each line and transformer area in the distribution network in the future period. Based on the dynamic fault risk level map, and combined with the real-time adjustable status information of the repair team, emergency power generation vehicle and material reserves, a resource scheduling optimization function is constructed with the goal of minimizing power outage time. Among them, the resource scheduling optimization function dynamically introduces the time urgency penalty factor according to the real-time change rate of the fault risk level of each area, so that areas with rapidly rising risk levels receive higher priority in resource allocation. The resource scheduling optimization function is solved by iterative optimization. By simulating the resource allocation scheme multiple times to approximate the minimum power outage time, the optimal scheduling scheme for the target deployment location and action sequence of each repair team, emergency power generation vehicle and materials in multiple future time windows is finally obtained. The optimal scheduling scheme is converted into scheduling instructions, which are then sent to the corresponding mobile terminals and resource management systems. The execution status of the instructions and the evolution of the disaster are tracked in real time for closed-loop feedback.

2. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: The dynamic characteristic sequence of disasters includes: Extract core indicators characterizing disaster intensity, boundary coordinates of the forecasted affected area, expected duration, and forecast path or evolution direction from disaster forecast data released by meteorological monitoring systems; extract line type, tower density, and equipment commissioning years from power grid geographic information systems as basic data on equipment vulnerability. Spatially aligned according to the same grid number, and normalized to map the numerical range to the interval of 0 to 1 for various types of data; spliced ​​in chronological order to form a feature vector sequence containing timestamp, grid number, intensity normalized value, influence range marker, duration, evolution direction angle, moving speed, line type code, tower density normalized value, and commissioning year normalized value.

3. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: Based on the dynamic feature sequence of disasters, matching historical similar disaster scenarios are retrieved from the historical fault database, specifically including: Calculate the cosine similarity between the current dynamic feature sequence of disasters and the feature sequence of each historical disaster record in the historical fault database; The cosine similarity scores were sorted from largest to smallest, and the top three historical disaster records with the highest similarity scores were selected as the matching historical similar disaster scenarios. Extract the location, type, and number of line faults that actually occurred within 24 hours after the disaster from these three historical disaster records to obtain the actual fault distribution pattern.

4. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: Generate a dynamic fault risk level map, specifically including: Sort all grid regions in descending order of their failure probability distribution values ​​to obtain the failure probability distribution value for each grid region. Sort all grid regions from largest to smallest according to the total number of failures in each historical similar disaster scenario to obtain the historical experience risk value corresponding to each grid region; The fault probability distribution value of each grid area is added to the historical risk value to obtain the comprehensive risk value of the grid area; According to the comprehensive risk value from largest to smallest, the lines and transformer areas corresponding to the top 10% of the grid areas are marked as high-risk, those between 10% and 30% are marked as medium-risk, and those in the bottom 70% are marked as low-risk. On the geographic information system map of the power distribution network, high-risk areas are rendered in red, medium-risk areas in yellow, and low-risk areas in green, forming a visualized dynamic fault risk level map.

5. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: Real-time adjustable status information of repair teams, emergency power generation vehicles, and material reserves, including: The vehicle-mounted GPS terminal collects the current latitude and longitude coordinates of each repair team and each emergency power generation vehicle in real time; the electronic ledger of the resource management system obtains the current warehouse location, material type and available quantity of each batch of materials; the location information is associated with the dynamic fault risk level map using the same grid numbering system to establish a mapping relationship between resource location and risk area.

6. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: Construct a resource scheduling optimization function with the objective of minimizing power outage time, including: The objective is to minimize the sum of the estimated outage times for all fault areas, where the estimated outage time for each fault area is equal to the travel time of the repair team from its current location to the fault area plus the estimated repair time. The constraints of the resource scheduling optimization function include: each repair team can only be dispatched to one fault area at a time, each emergency generator can only support one transformer substation at a time, and the amount of materials transferred from each warehouse cannot exceed the available inventory of the warehouse.

7. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: The resource scheduling optimization function dynamically introduces a time urgency penalty factor based on the real-time change rate of the fault risk level in each region, including: The comprehensive risk value of each grid area is calculated periodically according to a preset time interval, and the change in risk value between adjacent time intervals is recorded. When the risk value of a certain grid area shows a positive increase over multiple consecutive time intervals, and the cumulative increase exceeds 10% of the original comprehensive risk value, the area is marked as an area of ​​accelerated risk deterioration. When allocating resources to areas where risks are rapidly deteriorating, an urgency multiplier greater than 1 is introduced into the calculation of their resource needs. The specific value is positively correlated with the cumulative growth rate of the risk value; the greater the growth rate, the higher the value of the urgency multiplier. By introducing a urgency multiplier, the optimization function tends to allocate available resources preferentially to regions where the risk is accelerating during the solution process.

8. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: The resource scheduling optimization function is solved using an iterative optimization approach, including: First, an initial solution is generated, which means randomly assigning a target area to each repair team and each emergency power generation vehicle; Fine-tune the current solution by swapping the target areas of the two teams or changing the drop location of a certain material, and calculate the weighted power outage time corresponding to the new solution; If the power outage time of the new solution is shorter than that of the original solution, the new solution is accepted; otherwise, the new solution is accepted with a certain probability. Repeat the above fine-tuning and comparison process until the power outage time no longer decreases after multiple iterations. Then output the current optimal solution as the final scheduling scheme.

9. The method for optimizing the scheduling of power distribution network resources under severe weather conditions according to claim 1, characterized in that: Real-time tracking of command execution status and disaster evolution for closed-loop feedback, including: The system monitors whether each scheduling instruction has been executed by uploading confirmation information through mobile terminals. If no execution confirmation is received within 30 minutes after an instruction is issued, the resource status is marked as abnormal and it re-participates in the next round of scheduling calculation. The system reacquires the latest meteorological monitoring data every 10 minutes, compares the newly generated optimal scheduling plan with the previous plan, and only sends the changed instructions to the resource terminal, while keeping the unchanged instructions running.

Images