A three-stage multi-resource coordinated power distribution network resilience improvement method and system

Abstract

This invention discloses a three-stage, multi-resource collaborative method and system for improving the resilience of distribution networks. The method includes: generating typical disaster scenarios and assessing load priorities; constructing a pre-positioning model and solving for an emergency site pre-positioning scheme, and pre-deploying it; acquiring real-time status datasets; combining a multi-resource collaborative constraint system to decompose the dynamic scheduling optimization problem during a disaster into a discrete decision master problem and a continuous optimization sub-problem, and iteratively solving them; dynamically generating and executing scheduling strategies; prioritizing the repair of damaged components; solving a multi-objective distribution network reconfiguration model to obtain the optimal distribution network reconfiguration scheme and implementing it; quantitatively evaluating the scheme implementation effect; archiving the entire process data and optimizing the parameters of the integrated model, pre-positioning model, and decomposition algorithm; and constructing a knowledge base to achieve rapid response to subsequent disasters. This invention breaks through the limitations of traditional segmented responses and solves the problem that traditional models ignore cross-domain impacts, leading to schemes that are detached from reality.

Description

Technical Field

[0001] This invention belongs to the field of power distribution system resilience enhancement technology, and relates to a three-stage multi-resource collaborative method and system for enhancing the resilience of power distribution networks. Background Technology

[0002] Global climate anomalies have led to frequent natural disasters. As the end point of power supply, the distribution network is vulnerable to large-scale power outages caused by disasters such as typhoons and floods. Rapid power restoration has become a core requirement for improving the resilience of the power grid. Existing research on post-disaster recovery and resilience enhancement of distribution networks focuses on resource optimization and system recovery under disaster scenarios: some studies focus on the generation and reduction of disaster fault scenarios, optimizing the pre-positioning of emergency sites based on historical data; some explore the joint allocation of flexible resources such as repair teams, mobile energy storage, and distributed power sources to shorten load recovery time; some studies focus on the cross-domain coupling of power and transportation networks, integrating road network accessibility constraints to improve scheduling feasibility; and some works divide the recovery order based on indicators such as load importance and fault impact range to ensure priority power supply to critical loads. All of these studies revolve around the efficient allocation of post-disaster resources and rapid system recovery.

[0003] Although existing solutions cover key aspects of pre-disaster, during-disaster, and post-disaster recovery, significant shortcomings remain: They often employ segmented optimization, lacking a closed-loop collaborative mechanism across the entire "pre-disaster - during-disaster - post-disaster" process; scenario generation does not fully consider the coupling effects of multiple disasters; cross-domain coupling modeling is one-sided, focusing primarily on power-transportation coupling while neglecting deep collaboration among power flow, traffic flow, and information flow, and failing to fully integrate dynamic characteristics such as distributed power generation output fluctuations; multi-resource collaboration dimensions are limited, with insufficient consideration for the coordinated configuration of various emergency resources, and an inadequate balance between algorithm solution efficiency and optimization objectives; post-disaster recovery lacks a comprehensive data review and strategy iteration system, static solutions are ill-suited to sudden situations during disasters, and load priority assessment does not fully incorporate temporal characteristics and dynamic operating conditions, making it difficult to support accurate decision-making. Therefore, existing post-disaster power restoration technologies for distribution networks mainly focus on single-resource scheduling, isolated phase optimization, or neglect network coupling. There is an urgent need for a method to enhance the resilience of distribution networks after disasters, which covers the entire life cycle, integrates multi-domain coupling relationships and dynamic characteristics, and achieves full-process collaborative optimization of meta-emergency multi-resources. This method can enable rapid power restoration and control, and achieve an upgrade in disaster response from passive response to proactive prevention and control, and from segmented optimization to closed-loop iteration. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a three-stage, multi-resource collaborative method and system for enhancing the resilience of distribution networks. It integrates typical disaster scenario generation, multi-resource collaborative scheduling, post-disaster recovery optimization, and iterative effect optimization. Through typical disaster scenario generation and load priority assessment, it optimizes emergency site pre-location and resource pre-deployment. Combined with a multi-resource collaborative constraint system, it employs a decomposition algorithm to achieve dynamic scheduling during disasters. Post-disaster recovery is completed through priority ranking of damaged component repairs and multi-objective distribution network reconstruction. Finally, it builds a knowledge base based on full-process data archiving and model parameter optimization, thereby achieving enhanced distribution network resilience and rapid power restoration.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of this invention proposes a three-stage, multi-resource collaborative method for improving the resilience of distribution networks, comprising: Disaster prevention resource coordination and pre-positioning stage: Generate typical disaster scenarios and evaluate load priority based on LSTM-AdaBoost ensemble model, construct a pre-positioning model based on the typical disaster scenarios and load priority, solve for the emergency site pre-positioning scheme, and pre-deploy it; In the disaster dynamic collaborative scheduling phase: real-time data on distribution network faults, traffic network status, and emergency resource status are acquired and fused to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, a decomposition algorithm is used to decompose the disaster dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization sub-problem, and the solutions are iteratively obtained. The scheduling strategy is dynamically generated and executed. Post-disaster recovery optimization and iteration phase: Prioritize the repair of damaged components, solve the multi-objective distribution network reconstruction model to obtain the optimal distribution network reconstruction scheme and implement it, quantify and evaluate the implementation effect of the scheme, archive the whole process data and optimize the parameters of the integrated model, pre-positioning model and decomposition algorithm, and build a knowledge base to achieve rapid response to subsequent disasters.

[0007] Preferably, the step of generating typical disaster scenarios and evaluating load priorities based on the LSTM-AdaBoost ensemble model, constructing a pre-location model based on the typical disaster scenarios and load priorities, solving for an emergency site pre-location scheme, and pre-deploying the scheme includes: Based on historical disaster data, power grid infrastructure information, and equipment vulnerability parameters, typical disaster scenarios are generated using Bayesian networks and the Monte Carlo method. During the generation of these scenarios, the probability of line faults involving multi-flow coupling is dynamically adjusted, as shown in the following formula:

[0008] In the formula, For the line l Dynamic failure probability at time t in disaster scenario s; For the line lThe baseline failure rate under normal operating conditions; α, β, and γ are weighting coefficients. Disaster intensity The corresponding fault impact function; Traffic network congestion coefficient at time t The corresponding reachability influence function; For the line l Related distributed power generation output fluctuation coefficient The corresponding electrical energy flow coupling effect function; Feature vectors are obtained based on load baseline data, historical load curves, and disaster impact coefficients and input into the LSTM-AdaBoost ensemble model to evaluate load priority and obtain the load priority weights of each node in the distribution network. Based on typical disaster scenarios and the load priority weights of each node in the distribution network, a mixed integer linear programming pre-positioning model is constructed with the goal of minimizing pre-deployment costs and expected scheduling time to optimize resource allocation, obtain emergency site pre-positioning schemes, and carry out pre-deployment.

[0009] Preferably, the step of constructing a mixed-integer linear programming pre-positioning model based on typical disaster scenarios and load priority weights of each node in the distribution network, with the goal of minimizing pre-deployment costs and expected scheduling time, to optimize resource allocation, obtain an emergency site pre-positioning scheme, and perform pre-deployment, specifically includes: Define the decision variables for a mixed-integer linear programming prepositioning model, including binary variables. , , , and continuous variables , , ;in, This indicates that a charging station is deployed at node i in the distribution network; a value of 0 indicates otherwise. Indicates on traffic roads p If a maintenance station is deployed nearby, the result is 0; otherwise, it is not. This indicates that node j is normal in scene s; a value of 0 indicates otherwise. Indicates the line l Normal in scenario s, 0 indicates the opposite; This represents the shortest travel time from node i to node j at the charging station; Indicates that the maintenance station is located from the road. p to the damaged line l Corresponding roads ql The shortest travel time; Indicates the line l Weights: , Let be the load priority weight for node j; Based on the aforementioned decision variables, a mixed-integer linear programming pre-positioning model is constructed with the goal of minimizing pre-deployment cost and expected scheduling time as its objectives:

[0010] in: This represents the weighted travel time from the charging station to the damaged node; This represents the weighted travel time from the repair site to the damaged line; S is a set of typical disaster scenarios. Let be the probability of scenario s occurring. Let R be the load priority weight of node i, and R be the set of traffic roads. This is the set of roads corresponding to the damaged routes. A set of nodes; Set constraints for the mixed-integer linear programming pre-positioning model; solve the objective function under the constraints, output the emergency site pre-positioning scheme, and pre-deploy it.

[0011] Preferably, the step of acquiring and fusing data on distribution network faults, traffic network status, and emergency resource status to obtain a real-time status dataset, and combining this with a multi-resource collaborative constraint system, employing a decomposition algorithm to decompose the dynamic scheduling optimization problem during a disaster into a discrete decision master problem and a continuous optimization subproblem, and iteratively solving these subproblems, dynamically generating and executing a scheduling strategy, includes: Real-time data on distribution network faults, traffic network status, and emergency resource status are acquired and outlier removal, data alignment, and scenario matching are performed to obtain a standardized real-time status dataset. Using the connection status of mobile energy storage, the route of emergency repair teams, and the status of line switches as decision variables, a multi-resource collaborative constraint system is constructed, which includes constraints on node power balance, mobile energy storage charging and discharging and SOC, the accessibility of emergency repair teams, and the safety constraints on line topology, capacity, and voltage. Based on a standardized real-time status dataset and a multi-resource collaborative constraint system, a decomposition algorithm is used to decompose the disaster dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization sub-problem, and solve them iteratively to dynamically generate and execute a scheduling scheme. By combining a rolling window mechanism, the scheduling plan is adjusted based on real-time feedback to adapt to unexpected scenarios.

[0012] Preferably, the step of using a decomposition algorithm based on a standardized real-time state dataset and a multi-resource collaborative constraint system to decompose the disaster-stricken dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization subproblem, and then iteratively solving them to dynamically generate and execute a scheduling scheme, specifically includes: The discrete decision master problem is constructed, with the objective function being to minimize the load shedding cost and resource scheduling cost. The constraints include resource quantity constraints, traffic accessibility constraints for emergency repair teams, line topology and capacity constraints, as well as optimization cut and feasibility cut constraints from subproblems. Construct a continuous optimization subproblem, whose input is the discrete variable output of the main problem. The objective function is to minimize node voltage deviation and line power flow exceedance. The constraints include the dual form of node power balance constraints, mobile energy storage charging and discharging and SOC constraints, voltage safety constraints, and line topology and capacity constraints. The discrete decision-making main problem and the continuous optimization subproblem are solved iteratively.

[0013] Preferably, the step of combining a rolling window mechanism to adjust the scheduling scheme based on real-time feedback to adapt to unexpected scenarios specifically includes: Set the rolling window parameters and adopt a window mode of 15-minute update and 1-hour prediction. Generate a scheduling plan for the next hour in each window. After executing the plan for the first 15 minutes, re-optimize the plan for the next 45 minutes based on the new real-time data. At the end of each 15-minute cycle, data on the execution status of the plan is collected, including resource arrival status, load recovery status, and voltage over-limit status. Deviation analysis is performed based on the collected data on the execution status of the plan, and the scheduling plan is adjusted based on the results of the deviation analysis.

[0014] Preferably, the process of prioritizing the repair of damaged components, solving the multi-objective distribution network reconfiguration model to obtain the optimal distribution network reconfiguration scheme and implementing it, quantitatively evaluating the implementation effect of the scheme, archiving the entire process data and optimizing the parameters of the integrated model, pre-positioning model and decomposition algorithm, and building a knowledge base to achieve rapid response to subsequent disasters includes: The Analytic Hierarchy Process (AHP) is used to prioritize the repair of damaged components. Then, the optimal distribution network reconfiguration scheme is obtained by solving the dual-objective distribution network reconfiguration model that minimizes network loss and maximizes load recovery rate. The scheme is then implemented to obtain the corresponding load recovery rate for emergency resource recovery and reuse planning, thereby enabling the distribution network to smoothly transition to normal operation. An evaluation index system is constructed from four dimensions: power restoration efficiency, emergency resource utilization efficiency, economic cost, and user satisfaction, to quantify the effectiveness of the implementation plan. Archive data from the entire process and optimize the parameters of the integrated model, pre-positioning model, and decomposition algorithm. Based on the evaluation results, extract the optimal response strategies for typical scenarios and build a knowledge base to support rapid response to subsequent disasters.

[0015] Preferably, the step of prioritizing the repair of damaged components using the analytic hierarchy process (AHP) and then solving the dual-objective distribution network reconfiguration model of minimizing network loss and maximizing load recovery rate to obtain the optimal distribution network reconfiguration scheme and implement it, thereby obtaining the corresponding load recovery rate for emergency resource recovery and reuse planning, and realizing a smooth transition of the distribution network to normal operation, specifically includes: An evaluation system was constructed using the analytic hierarchy process (AHP). The evaluation criteria layer contained four dimensions: scope of failure impact, load importance, repair difficulty, and material availability. Each damaged component was scored according to these four dimensions, and a weighted score was calculated. The scores were then sorted to generate a corresponding repair priority list. A dual-objective optimization model is constructed to minimize network losses and maximize load recovery rate. The decision variables are line switch status and DG output. The constraints include: main grid power supply capacity limit, DG output upper limit constraint, and line radial constraint. Based on the repair priority list, solve the bi-objective optimization model, output the optimal distribution network reconfiguration scheme, and execute it; Obtain the load recovery rate after the implementation of the optimal distribution network reconfiguration scheme, determine the timing of resource recovery based on the load recovery rate for emergency resource recovery, formulate a recovery route based on the shortest path from the current location to the warehouse, detect the resource status and carry out reuse planning, update the resource database data, and record the status and next available time of each resource. Output a repair priority list, a power distribution network reconstruction execution plan, and an emergency resource recovery list.

[0016] Preferably, the indicators of the power supply restoration efficiency dimension include the system average outage time, the system average number of outages, the critical load restoration time, and the total load restoration rate; The indicators for the emergency resource utilization efficiency dimension include mobile energy storage utilization rate, effective working rate of emergency repair teams, and resource idle rate. The indicators in the economic cost dimension include total repair cost, power outage loss cost, and cost saving rate; The user satisfaction dimension is a weighted average of scores from three dimensions: timely recovery, power supply stability, and communication transparency.

[0017] Preferably, the process of archiving the entire data process and optimizing the integrated model, pre-positioning model, and decomposition algorithm parameters, extracting the optimal response strategies for typical scenarios based on the evaluation results, and constructing a knowledge base to support rapid disaster response, specifically includes: Archive all process data and use a time-series database to store real-time monitoring data, and a relational database to store the structured solution and evaluation results; Based on the archived data, adjust the weight parameters of the prepositioning model; retrain the ensemble model and adjust the iterative convergence threshold of the decomposition algorithm; Based on the assessment results, the optimal response strategies for typical scenarios are extracted and a strategy knowledge base is built to form a problem-solution mapping table, so that when similar problems are detected in subsequent disasters, solutions can be automatically pushed.

[0018] A second aspect of this invention proposes a three-stage, multi-resource collaborative distribution network resilience enhancement system, the system comprising: The disaster prevention resource coordination and pre-positioning module is used to generate typical disaster scenarios and evaluate load priorities based on the LSTM-AdaBoost ensemble model. Based on the typical disaster scenarios and load priorities, a pre-positioning model is constructed and the emergency site pre-positioning scheme is obtained and pre-deployed. The disaster dynamic collaborative scheduling module is used to acquire and merge data on distribution network faults, traffic network status and emergency resource status in real time to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, the decomposition algorithm is used to decompose the disaster dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization sub-problem and solve them iteratively, dynamically generating and executing scheduling strategies. The post-disaster recovery optimization and iteration module is used to prioritize the repair of damaged components, solve the multi-objective distribution network reconstruction model to obtain the optimal distribution network reconstruction scheme and implement it, quantitatively evaluate the implementation effect of the scheme, archive the full-process data and optimize the parameters of the integrated model, pre-positioning model and decomposition algorithm, and build a knowledge base to achieve rapid response to subsequent disasters.

[0019] A third aspect of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0021] Compared with the prior art, the beneficial effects of the present invention include at least the following: (1) This invention constructs a closed-loop system covering the entire lifecycle of "disaster prevention-disease in-disaster-post-disaster", breaking the limitations of traditional segmented response. In the disaster prevention stage, resource allocation is optimized through accurate scenario prediction. During the disaster, the dynamic scheduling mechanism is used to quickly adapt to the evolution of faults and changes in the external environment. In the post-disaster stage, the system recovery is accelerated through recovery optimization and resource reuse, comprehensively improving the disaster resistance resilience and power supply stability of the distribution network. At the same time, through post-disaster data review and knowledge base construction, the typical scenario response strategies are accumulated and the model parameters are continuously iterated, providing intelligent decision support for subsequent disaster response and forming a long-term disaster resistance capability. (2) This invention incorporates the concept of multi-flow coordination of power flow, traffic flow, and information flow throughout the entire process and multiple stages. Addressing the deep coupling characteristics of power distribution networks, transportation networks, and distributed power sources, it integrates multi-source sensing data for precise correction, effectively solving the problem of traditional models neglecting cross-domain influences and resulting in solutions that are detached from reality. Through a multi-dimensional constraint system and multi-step collaborative optimization, it adapts to different power distribution network operating conditions and energy structure characteristics, ensuring the safety of power supply to critical loads while improving the absorption capacity of distributed energy, thus balancing power supply reliability and power quality. (3) This invention achieves a dual improvement in resource optimization and operational efficiency through efficient algorithms and intelligent control technology, thereby reducing operation and maintenance and emergency repair costs; it significantly reduces the total life cycle cost of the power distribution network by reducing power outage losses and optimizing resource utilization; and it can prioritize the power supply of key loads for people's livelihood and public services. Attached Figure Description

[0022] Figure 1 This is a flowchart of the three-stage multi-resource collaborative distribution network resilience enhancement method of the present invention.

[0023] Figure 2 This is the optimal location for charging stations and maintenance stations.

[0024] Figure 3 For comparison of load recovery efficiency. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0026] Embodiment 1 of this invention provides a three-stage, multi-resource collaborative method for improving the resilience of distribution networks. This involves a three-stage architecture: a closed-loop architecture encompassing "pre-disaster resource collaborative pre-location - dynamic collaborative scheduling during disaster - post-disaster recovery optimization." The load priority assessment method uses an LSTM-AdaBoost ensemble model, inputting six features and integrating them through ten LSTM weak learners. The emergency site pre-location model considers the coupling between traffic and the distribution network, minimizing the objective function of "pre-deployment cost + expected scheduling time," with constraints including resource quantity, coverage, and traffic accessibility. The solution employs Gurobi+parallel computation to output the optimal site location. Multi-resource collaborative constraint modeling integrates three types of constraints: power balance, MESS SOC, and CT routing, ensuring the feasibility of the solution. The post-disaster optimization method utilizes AHP damaged component sorting, NSGA-II distribution network reconstruction, and a "disaster-scheduling-assessment" database + strategy knowledge base for iterative optimization. A cross-network coupling mechanism calculates the shortest resource path using the Dijkstra algorithm, establishing a "traffic status-resource routing-distribution network recovery" association to improve the practicality of the solution. Figure 1 As shown, the method includes the following steps: S1: Disaster Prevention Resource Coordination and Pre-positioning Phase: Generate typical disaster scenarios and evaluate load priorities based on the LSTM-AdaBoost ensemble model. Construct a pre-positioning model based on the typical disaster scenarios and load priorities, solve for the emergency site pre-positioning scheme, and pre-deploy it. Specifically, this includes: S11: Disaster scenario generation and quantitative modeling, the specific methods are as follows: First, three types of data are collected, including historical disaster data such as wind speed, rainfall, and duration, network topology (basic information of the power grid), and equipment vulnerability parameters such as line fault wind speed threshold and substation flooding depth threshold. Then, typical disaster scenario sets are generated through Bayesian network construction and Monte Carlo scenario sampling. When constructing the Bayesian network, the nodes represent disaster intensity, such as wind speed and rainfall, equipment status, and the scope of fault impact. Taking wind speed as an example, a conditional probability table is used to fit the mapping of "wind speed → line failure rate".

[0027]

[0028] in: Let l be the failure rate of line l in scenario s. The baseline failure rate for the line. The real-time wind speed in scenario s. This is the critical wind speed for the line. The wind speed at which the power line collapsed. Let μ be the cumulative function of the standard normal distribution, μ be the historical mean wind speed, and σ be the historical standard deviation of wind speed.

[0029] The formula for correcting the line fault probability for multi-flow coupling is as follows:

[0030] In the formula, For the line l Dynamic failure probability at time t in disaster scenario s; For the line l The baseline failure rate under normal operating conditions; α, β, and γ are weighting coefficients (satisfying α+β+γ=1). Disaster intensity (e.g., wind speed) The corresponding fault effect function; Traffic network congestion coefficient at time t The corresponding reachability influence function; For the line l Related distributed power generation output fluctuation coefficient The corresponding electrical flow coupling effect function.

[0031] Subsequently, the Monte Carlo method is used for scene sampling. The failure rate calculated by the above formula is used as the conditional probability parameter of each node (such as line and device) in the Bayesian network, thereby establishing a joint probability distribution based on the Bayesian network, as shown in the following formula:

[0032] in, This represents the joint probability of n variables (such as the fault states of each line and node) in a typical disaster scenario of a power distribution network. This represents the conditional probability of variable Xi under the conditions of its parent node Pa(Xi) (such as antecedent factors such as wind speed and rainfall that affect line faults).

[0033] Generate 1000-2000 initial fault scenarios; By using the K-means clustering algorithm, which uses the "fault location similarity" of fault status data associated with the basic data of distribution network equipment and the "power outage loss similarity" of load and economic loss data as clustering indicators, the initial scenarios are reduced to 10-15 typical scenarios, and the final output is a set of typical disaster scenarios.

[0034] S12: Dynamic load priority assessment, which quantifies the load importance of different users through an ensemble learning model to ensure priority recovery of critical loads during a disaster. The process is as follows: Based on load fundamentals such as user type, power, and reliability requirements, historical load curves, and disaster impact coefficients, The input data is collected, where the disaster impact coefficient is a quantitative parameter characterizing the impact of disasters on the electricity demand of different types of loads. Its value needs to be determined in combination with the disaster type (such as typhoon, flood), disaster intensity level, disaster exposure degree of the area where the load is located, the disaster resistance characteristics of the load itself (such as whether it has emergency backup power supply), and historical load electricity fluctuation data during disasters. Specifically, it can be obtained by statistically analyzing the ratio of the actual electricity consumption change of different loads under similar historical disasters to the normal electricity demand before the disaster, and then calculating the weight of each influencing factor by combining the analytic hierarchy process. The value range is usually between 0.8 and 1.5 (less than 1 indicates a decrease in load demand, and greater than 1 indicates an increase in load demand).

[0035] Next, we construct the LSTM-AdaBoost ensemble model. The first step is to train the LSTM sub-model, whose core formula is as follows:

[0036]

[0037]

[0038] in: Forgot the door at time t, The input gate is at time t. The output gate is at time t. The cell state at time t. Output at time t. These are the weight matrices for the forget gate, input gate, cell state, and output gate, respectively. These are the corresponding biases. Here, sigmoid is the activation function, and tanh is the hyperbolic tangent activation function. Output at time t-1 The input feature vector is given at time t.

[0039] The second step involves AdaBoost ensemble optimization, initializing 10 LSTM sub-models, with weights for each sub-model. (K=10 represents the number of sub-models). For k LSTM sub-models, predictor variables are obtained from the LSTM model using the training dataset. The computation step calculates the prediction error of the model's predictor variables on the training dataset. The global error of all training samples is calculated to evaluate the overall performance of the predictor variables. The coefficients of the current predictor variables are determined based on the global error; these coefficients are used to adjust their weights in the final ensemble predictor. The weight distribution of the training samples is updated accordingly based on the updated model weights. The calculation formulas for each variable are shown below:

[0040] The third step is to construct the final prediction model. By integrating the above predictor variables, a powerful composite predictor variable is developed, as shown in the following formula: Divided into 3 levels: (Critical load) (Important load) (Ordinary load) The final output is a load priority weight table for each node in the distribution network.

[0041]

[0042]

[0043] S13: Emergency site pre-location. Using a mixed-integer linear programming (MILP) model, the locations of charging stations (CS) and maintenance stations (RS) are optimized to minimize the "pre-deployment cost + expected scheduling time," as detailed below: First, define the decision variables, including binary variables. , , , ,like This indicates that a charging station is deployed at node i in the distribution network; if it is 0, then the opposite is true. This indicates that a maintenance station is deployed next to traffic road p, with a value of 0; otherwise, if... This indicates that node j is functioning normally in scene s; a value of 0 indicates otherwise. Indicates the line l Normal in scenario s, 0 indicates the opposite; continuous variable , representing the shortest travel time from node i to node j of the charging station; This indicates the repair station is located from road p to the damaged line. l The shortest travel time for the corresponding road ql; Indicates the line l Weights:

[0044] Next, we construct the objective function, aiming to minimize the weighted sum of "pre-deployment cost + expected scheduling time", as shown in the following formula.

[0045] Part 1: Weighted travel time from charging stations to damaged nodes; Part Two: Weighted travel time from the repair site to the damaged line; Where: S represents a set of typical scenarios. Let be the probability of scenario s occurring. Let R be the load priority weight of node j (from S12), and R be the set of traffic roads. This is the set of roads corresponding to the damaged routes.

[0046] Then set constraints, including resource quantity constraints:

[0047] Capacity constraint: The maximum number of MESS devices that can be connected to each charging station is ≤

[0048] Coverage constraint: The distance from the critical load node j to the nearest charging station is ≤

[0049] Traffic accessibility constraint: Expected travel time from the maintenance site to the damaged line ≤

[0050] Distribution network radiation constraints:

[0051] Site location constraints:

[0052] Solve the objective function under the given constraints, and finally output the emergency site pre-location scheme, including charging station deployment nodes, maintenance station deployment roads, and resource configuration for each site. The optimal locations of charging stations and maintenance stations are as follows: Figure 2 As shown.

[0053] S2: Disaster-Related Dynamic Cooperative Scheduling Phase: Real-time acquisition and fusion of data on distribution network faults, traffic network status, and emergency resource status to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, a decomposition algorithm is used to decompose the disaster-related dynamic scheduling optimization problem into a discrete decision master problem and continuous optimization sub-problems, which are then solved iteratively. Scheduling strategies are dynamically generated and executed, specifically including: S21: Real-time Status Awareness and Data Fusion. This step involves acquiring and fusing multi-source data to obtain real-time information on distribution network faults, transportation network status, and emergency resources, providing dynamic input for dispatching. The specific process is as follows: First, determine the data source and collection method. Distribution network fault data comes from SCADA systems used to collect line current and voltage, fault indicators used to collect line fault status, and smart meters used to collect user power outage status. The data content includes fault line number, fault occurrence time, damage level, and a list of power outage nodes. Traffic network data comes from traffic management department APIs to obtain road traffic status, road surveillance cameras to obtain congestion coefficients, and drone inspections to obtain water depth. The data also includes travel time correction factors. ; Emergency resource status data comes from resource controllers, GPS positioning, and handheld terminal reports. The data includes MESS remaining capacity, MEG remaining fuel, and the number of CT-repaired lines. Next, data fusion processing is performed, and outlier removal is carried out using 3... The criteria are as follows: the abnormal congestion coefficient is corrected based on historical traffic data; data alignment is based on the timestamp of the distribution network SCADA system, and traffic and resource data are synchronized by timestamp interpolation to ensure that the time deviation is ≤1 second; the fault scenario matching compares the real-time fault data, such as the number and location of faulty lines, with the typical scenarios generated by the disaster prevention and control, and calculates the matching degree γ = (number of faulty lines in common between real-time and typical scenarios) / (total number of faulty lines in real-time), and takes the typical scenarios with γ≥80% as the input of basic scheduling parameters; The final output is a real-time status dataset, including a fault list, traffic status table, resource status table, and scene matching results.

[0054] S22: Multi-resource collaborative constraint modeling. A multi-constraint model incorporating "power security, transportation accessibility, and resource characteristics" is constructed to ensure the feasibility of the scheduling scheme, as detailed below: First, define the decision variables, including binary variables. , , , This indicates that MESS m connects to distribution network node i at time t, and vice versa; This indicates that CT n arrives at traffic road p at time t, and vice versa; This indicates that line ij is closed at time t, and 0 indicates that it is not closed at time t; otherwise, it is a continuous variable. This represents the active power output by the MEG to node i at time t. This represents the charging / discharging power of MESS m at time t. This represents the active power flow of line ij at time t. This represents the reactive power flow of line ij at time t. This represents the voltage magnitude of node i at time t; Next, we construct the core constraints: Node power balance constraints: The power balance constraints of node i at time t ensure the conservation of active power, as shown in the following formula:

[0055] in: The power supplied from the main grid to the distribution network. For the output of the distributed power source at node i, Let MESS be the discharge power at node i. The charging power of MESS at node i. For the active power flow of the line from node i to j. Let i be the load power. Let be the load cut-off power for node i.

[0056] The formula for reactive power is as follows:

[0057] in: Reactive power of the main grid For reactive power output of distributed power sources, Unnecessarily contributing to MEG For reactive power flow of the line, For node i, the reactive load, The reactive power cut-off for node i.

[0058] Mobile energy storage charging and discharging and SOC constraints: MESS characteristic constraints ensure that its charging and discharging behavior conforms to the device characteristics. The SOC (State of Charge) change formula is as follows:

[0059] in: Let m be the state of charge of MESS at time t. Let SOC be at time t-1. For charging efficiency, The discharge efficiency is given by Δt, where Δt is the time step. For charging power, This refers to the discharge power; the SOC range is constrained to 10% ≤ ≤90%, charging and discharging power constraint is , Simultaneous charging and discharging are not allowed; charging and discharging are mutually exclusive. ; Accessibility constraints for emergency repair teams: CT routing constraints ensure that CT travel time complies with traffic network limitations, as shown in the following formula:

[0060] in: Let p be the base travel time from road p to q. Let τ be the travel time correction factor at time t, and τ be the time interval. This constraint indicates that after CT n arrives at road p at time t, it must travel at least... It takes hours to reach road q; CT working constraints: each CT can only be located on one road at a time, and each road can accommodate a maximum of [number missing] vehicles at a time. Support CT; load priority constraints ensure that critical loads are not removed, the formula is as follows: The power cutoff of critical load i is 0. ; Line topology and capacity constraints: Limiting line power flow to prevent exceeding limits, the formula is as follows:

[0061] in: The rated active power capacity of line ij is... When the line is disconnected, the power flow is 0. Voltage safety constraints: Ensure that node voltages are within the allowable range, as shown in the following formula:

[0062] in: , , Calculated by the DistFlow equation; This will ultimately form a complete constraint system, covering three categories of constraints: power operation, resource characteristics, and transportation accessibility.

[0063] S23: Dynamic scheduling scheme generation, using the Benders decomposition algorithm, decomposes the complex multi-constraint optimization problem into a main problem (discrete decision-making) and sub-problems (continuous optimization), improving solution efficiency. The specific process is as follows: First, we construct the main problem (discrete decision layer), with the objective function being to minimize the "load shedding cost + resource scheduling cost", as shown in the following formula:

[0064] in: A vector of discrete variables (including binary variables) , , , This indicates that MESS m connects to distribution network node i at time t, and vice versa; This indicates that CT n arrives at traffic road p at time t, and vice versa; This indicates that line ij is closed at time t, and 0 indicates that it is closed at time t (otherwise). Let θ be the discrete variable coefficient vector of the components related to the load shedding decision, and let θ be the lower bound of the resource scheduling-related optimal cost objective function value output by the subproblem. The main problem constraints include resource quantity constraints, CT routing constraints (accessibility constraints for emergency repair teams), line topology and capacity constraints, as well as optimization cuts and feasibility cuts from sub-problems. Among them, resource quantity constraints are the basic prerequisite constraints of the multi-resource collaborative constraint system, limiting the available resource scale of constraints such as node power balance, mobile energy storage charging and discharging, and SOC constraints. The core of the emergency repair team accessibility constraints is to limit the travel time, road capacity, and route path of the emergency repair team. The line topology and capacity constraints cover the requirements for line radiation and the limit of power flow exceeding the limit. The optimization cut constraints and feasibility cut constraints are iterative constraints of the Benders decomposition algorithm, which do not directly correspond to physical constraints. They correct the discrete decision of the main problem by feeding back the satisfaction of node power balance, mobile energy storage charging and discharging and SOC constraints, and line capacity and voltage safety constraints in the sub-problems. The node power balance constraints, mobile energy storage charging and discharging and SOC constraints, and node voltage safety constraints in the multi-resource collaborative constraint system are only indirectly guaranteed by the main problem through algorithm iteration in the sub-problems below.

[0065] Next, subproblems (continuous optimization layers) are constructed, with the discrete variables output by the main problem as input. The objective function is to minimize the node voltage deviation plus the line power flow limit, where the node voltage deviation corresponds to... The terms related to voltage constraints in the middle correspond to the limits of line power flow. Items related to line topology and capacity constraints in the middle and The formula for the tidal current equilibrium correlation term is as follows:

[0066] in: , Let them be the dual variable vectors, corresponding to the dual variables of equality constraints and inequality constraints, respectively. This is a vector of equality constraint coefficients (such as power balance constraint coefficients). G is the vector on the right side of the inequality constraint (such as the upper limit of voltage and the upper limit of power), and G is the correlation matrix between discrete variables and continuous variables (including node voltage magnitude, line active / reactive power flow, mobile energy storage charging and discharging power, mobile emergency generator output power, and load shedding power). The sub-problem constraints include the dual forms of node power balance constraints, MESS SOC constraints (mobile energy storage charging and discharging and SOC constraints), voltage safety constraints, and line topology and capacity constraints. Since the accessibility constraints of emergency repair teams in the multi-resource collaborative constraint system do not have a direct dual form, they indirectly affect the satisfaction of sub-problem constraints through the discrete decision variables of the main problem (the route of emergency repair teams).

[0067] Then, an iterative solution is performed, with the following steps: Step 1 Initialization: Set the lower bound Upper Realm The number of iterations k=0, and the initial discrete variables are... ,like This indicates that the MESS is initially located at a prepositioned node. In the mobile energy storage charging and discharging constraints, this initial connection state limits the initial power supply node of the mobile energy storage and is the basis for calculating the initial SOC (State of Charge). In the spatiotemporal routing constraints, this initial position defines the scheduling starting point of the mobile energy storage and determines the range of nodes that can be covered and the travel time constraints. Step 2 Solve the subproblem: convert the discrete variables Substitute the subproblem into the main problem. If the subproblem is infeasible, generate a feasibility cut to feed back to the main problem and correct the discrete decision. If it is feasible, calculate the optimal value of the subproblem (i.e., the optimal result that minimizes the node voltage deviation and the line power flow limit). Update the upper bound UB = ∑ + Generate an optimized cut (included in the main problem); Step 3: Solve the main problem: After adding cut constraints, solve the main problem to obtain the discrete variables. The optimal value of the main problem (i.e., the optimal result that minimizes the cost of load shedding and the cost of resource scheduling). Update the Nether ; Step 4 Convergence check: If If the iteration stops, then the iteration stops. The threshold for iterative convergence is set; otherwise, k = k + 1, and the process returns to Step 2. To improve the efficiency of the solution, a preheating startup strategy, time step optimization, and parallel computing are adopted. The final output is the dynamic scheduling scheme at time t, which includes the resource output plan, CT routing table, line switch operation instructions, and load restoration list.

[0068] S24: Dynamic updating and closed-loop adjustment of the scheduling plan. Based on real-time status feedback, a rolling window mechanism is used to update the scheduling plan to adapt to fault evolution and traffic changes. The specific process is as follows: (1) First, set the rolling window parameters and adopt the window mode of "15-minute update + 1-hour prediction". In each window, generate the scheduling plan for the next hour. After executing the plan for the first 15 minutes, re-optimize the plan for the next 45 minutes based on the new real-time data. (2) Next, feedback data collection is carried out. At the end of each 15-minute cycle, data on the execution status of the scheme are collected, including resource arrival status, load recovery status, and voltage over-limit status. (3) Then perform deviation analysis: If the actual arrival time of the resources deviates from the planned arrival time If the time is measured in minutes, analyze whether the cause is increased traffic congestion or the closure of new roads. If the deviation between the actual load restoration power and the planned restoration power is greater than the planned restoration power of node i at time t, that is... If so, analyze whether the cause is insufficient output of mobile energy storage or a new fault; If the voltage exceeds the limit, analyze whether the cause is excessive power flow in the line or fluctuation in DG output. (4) Finally, formulate a plan and adjust strategies: (3.1) Traffic congestion adjustment to address worsening traffic congestion: If the traffic congestion correction coefficient of road p increases by a certain amount at time t... >1.5, recalculate the shortest path for CT / MESS using Dijkstra's algorithm and adjust the routing scheme; (3.2) Handling newly added faults: If a newly added faulty line is detected... Add it to the fault list, rematch typical scenarios, and revise the MESS output plan; (3.3) Resource fault replacement for insufficient mobile energy storage output: If MESS m fails, the pre-deployed backup MESS m' is invoked, and the discharge power of the backup mobile energy storage m' to node i at time t is adjusted to be consistent with the discharge power of the original failed mobile energy storage m at that time and node. To ensure continuous power supply; (3.4) For newly closed roads: If road p is detected to be newly closed due to water accumulation or damage, the alternative routes of the repair team (CT) / mobile energy storage (MESS) will be recalculated using the Dijkstra algorithm, and non-closed roads will be given priority; if the closed road is a critical path, emergency resources (such as backup CT / MESS) pre-deployed in backup transportation hubs will be called, and the resource scheduling priority will be adjusted to ensure that the repair of damaged lines and emergency power supply are not interrupted.

[0069] (3.5) Adjustment for line power flow exceeding limits and DG output fluctuation: If the voltage exceeding limit is caused by line power flow exceeding limits If the overloaded line is disconnected via a remote control switch, the distribution network topology is reconstructed, and the power flow is transferred to a backup line; if the overload is caused by fluctuations in DG output (such as a sudden drop in photovoltaic output / a sudden increase in wind power), the charging and discharging power of the mobile energy storage is adjusted. Upward adjustment (Downward adjustment) to smooth out fluctuations in DG output.

[0070] (3.6) Dynamic adjustment of load priority: If the medical load A sudden increase will lead to an increase in its priority weight. Priority will be given to dispatching emergency repair teams to repair the associated lines of the node using CT scanners, and mobile energy storage devices (MESS) will prioritize power supply to the node. (5) Finally, output the updated scheduling scheme, including the reasons for the adjustment, a comparison table before and after the adjustment, and the next window optimization parameters.

[0071] S3: Post-disaster recovery optimization and iteration phase: Prioritize the repair of damaged components, solve the multi-objective distribution network reconfiguration model to obtain the optimal distribution network reconfiguration scheme and implement it, quantitatively evaluate the implementation effect of the scheme, archive the entire process data and optimize the parameters of the integrated model, pre-positioning model and decomposition algorithm, and build a knowledge base to achieve rapid response to subsequent disasters, specifically including: S31: Refined management and control of the post-disaster recovery process. By optimizing long-term recovery strategies, a smooth transition of the power distribution network from "emergency power supply" to "normal operation" is achieved. The specific process is as follows: First, the repair priorities of damaged components are ranked. An evaluation system is constructed using the Analytic Hierarchy Process (AHP). The criteria layer contains four dimensions: scope of impact of the fault, load importance, repair difficulty, and material availability. For each damaged component (equipment, line, node), a score of 1-10 is assigned and a weighted score is calculated to generate a repair priority list: scope of impact of the fault, weight 0.4, the more nodes affected, the higher the score. Load importance, weighted at 0.3, with higher scores for critical load-related components; Repair difficulty, weighted at 0.2, the less time required, the higher the score; Material availability, weighted at 0.1, with higher scores indicating sufficient material inventory; For each damaged component, score it on a scale of 1 to 10 and calculate a weighted score. ,according to A repair priority list is generated in descending order; the repair priority list clarifies the repair order of the damaged components and is the core basis for the next step of allocating emergency repair resources (such as emergency repair teams and repair materials) and arranging the repair sequence, that is, resources are scheduled to carry out repair work in order of priority from high to low according to the list.

[0072] The relationship between repair priority ranking and the optimal distribution network reconstruction scheme is as follows: First, the repair priority ranking provides a topology feasibility basis for distribution network reconstruction. The repair sequence of damaged components determined by it clarifies the range of available topologies. Distribution network reconstruction needs to optimize the switch status and power flow based on the topology formed by the repaired components. If key components are not repaired in priority, reconstruction cannot achieve dual-objective optimization. Second, the priority of repair is dynamically adjusted based on the demand for distribution network reconstruction. Unrepaired bottleneck components identified during the reconstruction process will be given higher priority, helping the reconstruction to achieve optimization goals based on the complete topology. Third, the two work together to serve the dual objectives of minimizing network loss and maximizing load recovery rate. Repair priority shortens the recovery time of critical loads, while distribution network reconstruction expands load coverage and reduces network loss.

[0073] Next, distribution network reconstruction and optimization are carried out, constructing a dual-objective optimization model of "minimum network loss + maximum load recovery", with the decision variable being the state of line switches. Contributing with DG (Understandably, it solves the problem through the coupling relationship between constraints and the objective function), thus reducing network loss. Minimize and load recovery rate The objective function is as follows:

[0074] in: For line resistance, , For line flow, For node voltage, To restore load power; A set of nodes; For the line ( i , j The resistance of ) For nodes i Load priority weights; Represents a set of routes; Constraints include main grid power supply capacity limitations. , DG output upper limit Line radiation constraints ; in, Main grid power supply capacity and main grid power supply capacity limits; Power output for DG, maximum power output for DG; As an auxiliary variable, it is used to characterize the topological association state of line (i,j) at time t; N is the total number of distribution network nodes. Let be the number of independent power source nodes in the distribution network at time t;

[0075] Where: the numerator T is the set of statistical time periods. It is the actual time spent on repair work by the h-th repair team at time t, and the denominator is... It is the total working time of the h-th repair team at time t.

[0076] The NSGA-II algorithm is used to solve the bi-objective optimization model, and the Pareto optimal solution is output. The scheme that "reduces network loss by 10% and achieves a load recovery rate of ≥95%" is selected for execution. Then, emergency resource recovery and reuse planning is carried out, and the recovery timing is determined based on the load recovery rate: when the load recovery rate of a certain area is ≥98%, the MESS / CT recovery of that area is initiated; recovery routes are formulated: Dijkstra's algorithm is used to calculate the shortest path from the current location of the resource to the warehouse; resource status detection: MESS checks the SOC, SOC ≥80% is marked "can be reused immediately", 30%≤SOC<80% is marked "requires charging before reuse", SOC<30% is marked "awaiting maintenance"; CT checks the equipment integrity, tool integrity rate ≥90% is marked "can be reused immediately"; update the resource database data, and record the status and next available time of each resource; The final output includes a repair plan for damaged components, an implementation scheme for power distribution network reconstruction, and an emergency resource recovery list.

[0077] S32: Multi-dimensional quantitative evaluation of power restoration effectiveness. The effectiveness of the technical solution is quantitatively evaluated from four dimensions: "power restoration, resource utilization, economic cost, and user experience," as detailed below: First, an evaluation index system is constructed. The power restoration efficiency index includes System Average Outage Time (SAIDI), System Average Outage Frequency (SAIFI), and Critical Load Restoration Time, with the following formulas:

[0078]

[0079]

[0080]

[0081] in: Let be the power outage duration of node i. Let i be the number of users at node i. Let be the number of power outages at node i. For critical load 100% recovery time, The end time of the disaster. To restore the load power of node i at time T, The pre-disaster load power of node i; emergency resource utilization efficiency indicators include MESS utilization rate, CT effective working rate, and resource idle rate, with the following formulas:

[0082]

[0083]

[0084] in: This represents the maximum charge and discharge power of MESS. Total resource deployment time, Let n be the actual repair time of CT at time t. Total CT scan duration For the duration of resource idle time; Economic cost indicators include total repair costs and emergency resource allocation costs. + Repair costs Costs of power outages ( For user type power outage loss coefficient), cost saving rate = (cost of traditional solution - cost of this solution) / cost of traditional solution × 100%; User satisfaction was assessed through a questionnaire survey, with scores ranging from 1 to 5 points across three dimensions: "Timeliness of power restoration (weight 0.4)," "Power supply stability (weight 0.4)," and "Communication transparency (weight 0.2)." The satisfaction score was calculated as: Satisfaction = ∑(Timeliness score × 0.4 + Stability score × 0.4 + Transparency score × 0.2) / (Total sample size × 5) × 100%. Next, an evaluation process was implemented. Data collection involved extracting outage data from the SAIDI / SAIFI software database, cost data from the financial system, and satisfaction questionnaire data from the user feedback platform. Benchmark comparisons were made between the evaluation results and "traditional solutions without multi-party collaboration and traffic coupling" and "industry average levels." Finally, a power restoration effectiveness evaluation report and a list of key improvement points were output.

[0085] S33: Data review and strategy iteration optimization, accumulating experience in post-disaster power restoration, optimizing software model parameters and algorithms, and improving subsequent disaster response capabilities. The specific process is as follows: First, full-process data archiving and structured storage are implemented, establishing a "Disaster-Dispatch-Assessment" database. This database categorizes data into three types: disaster data (including disaster type: typhoon / flood, intensity level: 1-5, affected area: list of damaged lines / nodes, duration, etc.), dispatch data (including pre-positioning plans: site locations, resource quantities; in-disaster dispatch instructions: MESS charging / discharging plans, CT routing tables; execution feedback: resource arrival deviations, load recovery deviations; resource status logs: MESSSOC time series, CT working duration time series), and assessment data (including results of various indicators, deviation cause analysis reports, and original user feedback). A time-series database is used to store real-time monitoring data, while a relational database stores structured plans and assessment results, ensuring data traceability. Next, model parameter optimization was performed. For the MILP pre-positioning model of S13, the weight parameters were adjusted based on the review data. For example, the weight of traffic in a certain area was adjusted. Too low a value will cause resource scheduling delays. Increase the proportion of traffic factors in the objective function; For the LSTM-AdaBoost model of S12, load data of this disaster was added, such as adding samples of "medical load suddenly increased by 20% during the typhoon", and the model was retrained to reduce the load priority assessment error. For the Benders decomposition algorithm of S23, the iterative convergence threshold is adjusted to improve the accuracy of the scheduling scheme; Then, a strategy knowledge base is built to extract the optimal response strategies for typical scenarios, such as "when a typhoon causes multiple line failures, MESS should be prioritized to medical nodes to ensure power supply, and CT should be prioritized to repair the main network connection line". Each strategy includes applicable scenarios, such as disaster type, number of failures, resource configuration, such as number of MESS / CT, and scheduling steps. Establish a "problem-solution" mapping table, such as "Problem: Traffic congestion causes CT delay >30 minutes → Solution: Pre-deploy backup CT to traffic hubs and activate backup resources during congestion", embed the "intelligent recommendation" module in the software, and automatically push solutions when similar problems are detected in subsequent disasters; The final output includes a structured database, an optimized model parameter configuration file, and a knowledge base of typical scenario coping strategies.

[0086] This invention can be applied to post-disaster resilience enhancement of distribution networks with a high proportion of distributed power sources connected to them. It focuses on scenario-based applications for distribution networks characterized by a high proportion of distributed power sources, diverse load types, network topology with multiple interconnecting lines, and the need for emergency resource scheduling to adapt to power output fluctuations.

[0087] First, using the disaster prevention resource collaborative pre-location method described in S1, detailed information on distribution network nodes, lines, towers, and distributed power source access points is collected. A "power source-load" coupling correction model is established by combining historical output curves of distributed power sources and load time-series characteristic data to ensure that the disaster scenario generation accurately reflects the superimposed effect of distributed power source output fluctuations and load peak-valley differences on fault impacts. The reliability requirements and outage loss coefficients of different types of loads, such as industrial production lines, commercial complexes, and residential communities, are collected. The LSTM-AdaBoost ensemble model is used to enhance the priority differentiation of diverse loads, highlighting the impact of "power source-load" matching degree on load importance. When pre-locating emergency sites based on the MILP model, charging stations are prioritized in areas with dense distributed power source access. Maintenance site selection considers both tie-line distribution and resource scheduling radius, while reserving spare resource capacity adapted to the output characteristics of distributed power sources to ensure that the pre-deployment scheme can support a collaborative power supply mode of "distributed power source + emergency resources."

[0088] Secondly, based on the disaster-prone dynamic collaborative scheduling process described in S2, and considering the voltage fluctuations, frequency shifts, and islanding formation that are easily caused by the high proportion of distributed power sources connected to the distribution network, data from the SCADA system, the distributed power source monitoring platform, and tie-line fault indicators are integrated to perceive the location of distribution network faults, the real-time output of distributed power sources, and the status of tie-line switches in real time. During data fusion, the focus is on correcting the distributed power source output prediction error coefficient and the tie-line fault isolation time coefficient to improve scenario matching accuracy. In multi-constraint modeling, the coupling constraints of "distributed power source output - emergency resource scheduling - power balance" are strengthened, the impact of distributed power source output fluctuations on load power supply is considered, and the mobile energy storage (MESS) charging and discharging strategy is optimized to smooth out fluctuations. At the same time, tie-line switching timing constraints are incorporated to avoid secondary faults caused by blind operation. When using the Benders decomposition algorithm to generate scheduling schemes, priority is given to building islanded power supply systems through the collaboration of distributed power sources and MESS to cover critical loads in the fault area and reduce the repair pressure on emergency repair teams (CT). Relying on the rolling window mechanism of "15-minute update + 1-hour prediction", the scheduling strategy is dynamically adjusted to adapt to the distributed power source output fluctuations and tie-line recovery status. Comparing the recovery rate time series of three scenarios: multi-resource collaboration, fixed-sequence repair, and CT-only repair, such as... Figure 3 The plans 1-3 shown represent three scenarios: multi-resource collaboration, fixed-sequence repair, and CT-only repair. The curves intuitively present the time-series progress and efficiency differences of load recovery in each scenario, demonstrating the advantages of the multi-resource collaboration strategy in terms of load recovery speed and final recovery rate.

[0089] Given the high penetration rate of distributed generation (DG), this study utilizes the post-disaster recovery optimization and iterative method described in S3, combined with the topological characteristics of the distribution network ("ring network + multiple tie lines"), and employs the Analytic Hierarchy Process (AHP) to prioritize the repair of damaged components. A new "distributed generation tie line weight" is added to the existing criteria layer, prioritizing the repair of tie lines connecting the DG cluster to the main network and key nodes. The NSGA-II algorithm is used to solve a three-objective distribution network reconfiguration model of "minimum network loss + maximum DG absorption + maximum load recovery," fully leveraging the flexibility of the ring network topology and the output characteristics of DG to optimize line switch states and power output allocation, achieving optimal synergy between network loss, power absorption, and load recovery. During resource recovery planning, the timing of MESS and CT recovery is optimized based on DG output prediction results, prioritizing the recovery of emergency resources in DG coverage areas to avoid resource idleness. In multi-dimensional evaluation, key additions include "distributed generation utilization rate," "island power supply duration," and "voltage deviation compliance rate." These distinctive indicators enhance the quantification of the synergistic effect of "power supply-load-resources." Based on this, the distributed power output weight coefficient of the prepositioning model and the proportion of multivariate load samples in the load assessment model are optimized through data review. Typical strategies such as "MESS dynamic charging and discharging in high-fluctuation scenarios of distributed power supply" and "priority repair of tie lines and coordinated power supply recovery" are extracted. A knowledge base for post-disaster resilience enhancement adapted to high proportion of distributed power supply access to the distribution network is constructed to achieve accurate identification of weak links in the distribution network and optimized allocation of emergency resources.

[0090] Based on the same overall concept as Embodiment 1, Embodiment 2 of the present invention provides a three-stage multi-resource collaborative distribution network resilience enhancement system, which consists of the following functional modules: A disaster prevention resource coordination and pre-positioning module is used to generate typical disaster scenarios and evaluate load priorities based on an LSTM-AdaBoost ensemble model. Based on the typical disaster scenarios and load priorities, a pre-positioning model is constructed and solved to obtain an emergency site pre-positioning scheme for pre-deployment. More preferably, based on historical disaster data, power grid infrastructure information, and equipment vulnerability parameters, typical disaster scenarios are generated using Bayesian networks and Monte Carlo methods. An LSTM-AdaBoost ensemble model is used to evaluate load priorities, and a mixed-integer linear programming model is constructed to optimize the deployment locations of charging stations and maintenance stations, as well as the allocation of resources such as mobile energy storage and emergency repair teams. The pre-positioning scheme is then output to support pre-disaster preparedness. The construction process of the LSTM-AdaBoost ensemble model includes: (1) Train the LSTM sub-model, where the LSTM sub-model formula includes:

[0091]

[0092]

[0093] in: Forgot the door at time t, The input gate is at time t. The cell state at time t. Output gate at time t; Output at time t. These are the weight matrices for the forget gate, input gate, cell state, and output gate, respectively. These are the biases for the forget gate, input gate, cell state, and output gate, respectively. Here, sigmoid is the activation function, and tanh is the hyperbolic tangent activation function. Output at time t-1 Input the feature vector at time t; (2) Perform AdaBoost ensemble optimization on the LSTM sub-model to output accurate load forecasts for each node of the distribution network at different time scales after the disaster, as well as a load importance quantification weight table, including: Initialize the weights of each LSTM sub-model , Number of sub-models; For the LSTM sub-model, the predictor variables are obtained from the LSTM sub-model using the training dataset; Calculate the prediction error of the model's predictor variables on the training dataset to obtain the global error for all training samples; The coefficients of the current predictor variable are determined based on the global error, and these coefficients are used to update the model weights. Based on updating the model weights, the weight distribution of the training samples is updated accordingly.

[0094] The formulas for calculating each predictor variable are as follows:

[0095] in: For prediction error, This represents the actual load priority value. This is the upper bound of the error. For global error, Let i be the weight of the i-th sample in the k-th round. For prediction coefficients, For weight adjustment factor, As a normalization factor, ensure It represents a probability distribution.

[0096] (3) Based on the predictive variables and weight coefficients obtained in step (2), the final prediction model is constructed by integrating the predictive variables described in (2), resulting in the LSTM-AdaBoost ensemble model to evaluate load priority and obtain the load priority weights of each node in the distribution network:

[0097]

[0098] in, For LSTM weights, For prediction coefficients, For the final priority weight, The prediction results for each model. The disaster-prone dynamic collaborative scheduling module is used to acquire and fuse real-time data on distribution network faults, transportation network status, and emergency resource status to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, it employs a decomposition algorithm to break down the disaster-prone dynamic scheduling optimization problem into a discrete main decision problem and continuous optimization sub-problems, solving them iteratively. It then dynamically generates and executes scheduling strategies. Specifically, it involves: The real-time status perception and data fusion unit is used to collect distribution network fault data through SCADA system, fault indicator, and smart meter, obtain traffic network status data by relying on traffic management department API, road monitoring, and drones, and collect emergency resource status data by combining resource controller and GPS positioning. After outlier removal, time synchronization calibration and scene matching, a standardized real-time status dataset is generated.

[0099] The multi-resource collaborative constraint modeling unit is used to define decision variables such as mobile energy storage connection status, emergency repair team routing, and line switch status. It constructs a multi-dimensional constraint system that includes node power balance, mobile energy storage charging and discharging and SOC constraints, emergency repair team accessibility, line capacity and voltage safety, etc., to ensure the feasibility and safety of the dispatching scheme.

[0100] The disaster dynamic scheduling and solution unit is used to decompose the optimization problem into a discrete decision master problem and a continuous optimization subproblem for iterative solution using the Benders decomposition algorithm. Combined with the rolling window mechanism of "15-minute update + 1-hour prediction", it dynamically generates resource output plan, emergency repair route table and load recovery list, and adjusts the scheduling strategy according to real-time feedback to adapt to sudden scenarios such as traffic congestion and new faults.

[0101] The post-disaster recovery optimization and iteration module is used to prioritize the repair of damaged components, solve the multi-objective distribution network reconfiguration model to obtain the optimal distribution network reconfiguration scheme, implement it, quantitatively evaluate the implementation effect, archive the entire process data, optimize the parameters of the integrated model, pre-positioning model, and decomposition algorithm, and build a knowledge base to achieve rapid response to subsequent disasters; specifically involving: The post-disaster recovery optimization unit is used to prioritize the repair of damaged components using the analytic hierarchy process (AHP). It solves the dual-objective distribution network reconfiguration model of "minimum network loss + maximum load recovery" using the NSGA-II algorithm. Based on the load recovery rate, it plans emergency resource recovery routes and reuse classifications to achieve a smooth transition of the distribution network to normal operation.

[0102] The power restoration effect evaluation and iteration unit is used to construct an evaluation index system from four dimensions: power restoration, resource utilization, economic cost, and user experience. It quantifies the implementation effect of the solution, archives the data of the whole process and optimizes the model parameters, extracts the response strategies for typical scenarios and builds a knowledge base to support rapid response to subsequent disasters.

[0103] Embodiment 3 of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0104] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0105] This invention constructs a closed-loop system covering the entire lifecycle of disaster prevention, during disaster, and post-disaster recovery, breaking through the limitations of traditional segmented responses. In the pre-disaster phase, resource allocation is optimized through precise scenario prediction; during disaster, a dynamic scheduling mechanism rapidly adapts to fault evolution and changes in the external environment; and in the post-disaster phase, recovery optimization and resource reuse accelerate system restoration, comprehensively enhancing the disaster resilience and power supply stability of the distribution network. Simultaneously, through post-disaster data review and knowledge base construction, typical scenario response strategies are accumulated and model parameters are continuously iterated, providing intelligent decision support for subsequent disaster responses and forming a long-term disaster resistance capability. This invention incorporates a multi-flow collaborative concept encompassing power flow, traffic flow, and information flow across the entire process and multiple stages. Addressing the deep coupling between distribution networks, transportation networks, and distributed power sources, it integrates multi-source sensing data to construct a precise correction model. This model (corresponding to the S11 typical disaster scenario generation stage, where an initial scenario is generated based on historical disaster data, basic grid information, and equipment vulnerability parameters, followed by scenario correction; the S12 load priority assessment stage, where load prediction deviations are corrected through an integrated model combining basic load data, historical load curves, and disaster impact coefficients; the S2.1 real-time data fusion stage, which integrates distribution network fault, transportation network status, and emergency resource status data, performing outlier removal, time synchronization, and scenario matching to correct data accuracy; and the S2.2 multi-resource collaborative constraint stage, which corrects relevant constraint parameters based on the coupling characteristics of distribution networks, transportation networks, and distributed power sources. The core is to improve the model's adaptability to actual operating conditions through multi-source data fusion and cross-domain parameter correction), effectively solving the problem of traditional models neglecting cross-domain influences and resulting in solutions being detached from reality. Through a multi-dimensional constraint system and a multi-step collaborative optimization strategy (corresponding to the S13 emergency site pre-positioning stage, which combines typical disaster scenarios and load priorities to optimize the location and resource allocation of emergency sites to minimize pre-deployment costs and expected scheduling time; the S2.3 disaster dynamic scheduling stage, which decomposes the dynamic scheduling optimization problem into discrete decision master problems and continuous optimization sub-problems for iterative solution, dynamically generating a scheduling scheme containing resource output plans, repair routes and load recovery lists; the S2.4 stage, which adjusts the scheduling strategy based on real-time feedback according to the rolling window mechanism to adapt to sudden scenarios; and the S3.1 post-disaster recovery stage, which uses repair priority ranking and multi-objective distribution network reconstruction model to collaboratively optimize the repair order of damaged components and distribution network topology to minimize network losses and maximize load recovery rate, the core is to adapt the coupling characteristics of distribution network and transportation network and distributed power sources through multi-stage, multi-resource, and multi-constraint collaborative adaptation), adapting to different distribution network operating conditions and energy structure characteristics, ensuring the power supply safety of critical loads, improving the distributed energy absorption capacity, and taking into account power supply reliability and power quality; This invention achieves a dual improvement in resource optimization and operational efficiency through efficient algorithms and intelligent control technology, reducing operation and maintenance and emergency repair costs; by reducing power outage losses and optimizing resource utilization, it significantly reduces the total life-cycle cost of the distribution network; and it can prioritize the power supply of critical loads for people's livelihood and public services.

[0106] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0107] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0108] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0109] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A three-stage, multi-resource collaborative method for improving the resilience of a distribution network, characterized in that, include: Before a disaster, typical disaster scenarios are generated and load priorities are evaluated based on the LSTM-AdaBoost ensemble model. Based on the typical disaster scenarios and load priorities, a pre-location model is constructed, and the emergency site pre-location scheme is solved and pre-deployed. During a disaster, real-time data on distribution network faults, traffic network status, and emergency resource status are acquired and fused to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, a decomposition algorithm is used to decompose the dynamic scheduling optimization problem during a disaster into a discrete decision master problem and a continuous optimization sub-problem, and the solutions are iteratively obtained. The scheduling strategy is then dynamically generated and executed. After a disaster, the repair priorities of damaged components are ranked, and based on this, a multi-objective distribution network reconfiguration model is solved to obtain the optimal distribution network reconfiguration scheme and implemented. The implementation effect of the scheme is quantitatively evaluated, the whole process data is archived, and the parameters of the integrated model, pre-positioning model and decomposition algorithm are optimized. A knowledge base is built to achieve rapid response to subsequent disasters.

2. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration as described in claim 1, characterized in that: The process of generating typical disaster scenarios and evaluating load priorities based on the LSTM-AdaBoost ensemble model, constructing a pre-location model based on the typical disaster scenarios and load priorities, solving for the emergency site pre-location scheme, and pre-deploying the solution includes: Based on historical disaster data, power grid infrastructure information, and equipment vulnerability parameters, typical disaster scenarios are generated using Bayesian networks and the Monte Carlo method. During the generation of these scenarios, the probability of line faults involving multi-flow coupling is dynamically adjusted, as shown in the following formula: In the formula, For the line l Dynamic failure probability at time t in disaster scenario s; For the line l The baseline failure rate under normal operating conditions; α, β, and γ are weighting coefficients. Disaster intensity The corresponding fault impact function; Traffic network congestion coefficient at time t The corresponding reachability influence function; For the line l Related distributed power generation output fluctuation coefficient The corresponding electrical energy flow coupling effect function; Feature vectors are obtained based on load baseline data, historical load curves, and disaster impact coefficients and input into the LSTM-AdaBoost ensemble model to evaluate load priority and obtain the load priority weights of each node in the distribution network. Based on typical disaster scenarios and the load priority weights of each node in the distribution network, a mixed-integer linear programming pre-positioning model is constructed with the goal of minimizing pre-deployment costs and expected scheduling time, so as to obtain an emergency site pre-positioning scheme and carry out pre-deployment.

3. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 2, characterized in that: Based on typical disaster scenarios and the load priority weights of each node in the distribution network, a mixed-integer linear programming pre-positioning model is constructed with the goal of minimizing pre-deployment costs and expected scheduling time. This model yields an emergency site pre-positioning scheme and enables pre-deployment. Specifically, this includes: Define the decision variables for a mixed-integer linear programming prepositioning model, including binary variables. , , , and continuous variables , , ;in, This indicates that a charging station is deployed at node i in the distribution network; a value of 0 indicates otherwise. Indicates on traffic roads p If a maintenance station is deployed nearby, the result is 0; otherwise, it is not. This indicates that node j is normal in scene s; a value of 0 indicates otherwise. Indicates the line l Normal in scenario s, 0 indicates the opposite; This represents the shortest travel time from node i to node j at the charging station; Indicates that the maintenance station is located from the road. p to the damaged line l Corresponding roads ql The shortest travel time; Indicates the line l Weights: , Let be the load priority weight for node j; Based on the aforementioned decision variables, a mixed-integer linear programming pre-positioning model is constructed with the goal of minimizing pre-deployment cost and expected scheduling time as its objectives: in: This represents the weighted travel time from the charging station to the damaged node; This represents the weighted travel time from the repair site to the damaged line; S is a set of typical disaster scenarios. Let be the probability of scenario s occurring. Let R be the load priority weight of node i, and R be the set of traffic roads. This is the set of roads corresponding to the damaged routes. A set of nodes; Set constraints for the mixed-integer linear programming pre-positioning model; solve the objective function under the constraints, output the emergency site pre-positioning scheme, and pre-deploy it.

4. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 1, characterized in that: The process involves acquiring and fusing data on distribution network faults, traffic network status, and emergency resource status to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, a decomposition algorithm is used to decompose the disaster-related dynamic scheduling optimization problem into a discrete decision master problem and continuous optimization sub-problems, which are then iteratively solved. This process dynamically generates and executes scheduling strategies, including: Real-time data on distribution network faults, traffic network status, and emergency resource status are acquired and outlier removal, data alignment, and scenario matching are performed to obtain a standardized real-time status dataset. Using the connection status of mobile energy storage, the route of emergency repair teams, and the status of line switches as decision variables, a multi-resource collaborative constraint system is constructed, which includes node power balance constraints, mobile energy storage charging and discharging and SOC constraints, resource quantity constraints, emergency repair team accessibility constraints, line topology and capacity constraints, and voltage safety constraints. Based on a standardized real-time status dataset and a multi-resource collaborative constraint system, a decomposition algorithm is used to decompose the disaster dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization sub-problem, and solve them iteratively to dynamically generate and execute a scheduling scheme. By combining a rolling window mechanism, the scheduling plan is adjusted based on real-time feedback to adapt to unexpected scenarios.

5. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 4, characterized in that: Based on a standardized real-time status dataset and a multi-resource collaborative constraint system, the method employs a decomposition algorithm to decompose the disaster-stricken dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization subproblem, and solves them iteratively. This dynamically generates and executes a scheduling scheme. Specifically, this includes: The discrete decision master problem is constructed, with the objective function being to minimize the load shedding cost and resource scheduling cost. The constraints include resource quantity constraints, traffic accessibility constraints for emergency repair teams, line topology and capacity constraints, as well as optimization cut and feasibility cut constraints from subproblems. Construct a continuous optimization subproblem, whose input is the discrete variable output of the main problem. The objective function is to minimize node voltage deviation and line power flow exceedance. The constraints include the dual form of node power balance constraints, mobile energy storage charging and discharging and SOC constraints, voltage safety constraints, and line topology and capacity constraints. The discrete decision-making main problem and the continuous optimization subproblem are solved iteratively.

6. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 4, characterized in that: The aforementioned combination of a rolling window mechanism, which adjusts the scheduling scheme based on real-time feedback to adapt to unexpected scenarios, specifically includes: Set the rolling window parameters and adopt a window mode of 15-minute update and 1-hour prediction. Generate a scheduling plan for the next hour in each window. After executing the plan for the first 15 minutes, re-optimize the plan for the next 45 minutes based on the new real-time data. At the end of each 15-minute cycle, data on the execution status of the plan is collected, including resource arrival status, load recovery status, and voltage over-limit status. Deviation analysis is performed based on the collected data on the execution status of the plan, and the scheduling plan is adjusted based on the results of the deviation analysis.

7. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 1, characterized in that: The process involves prioritizing the repair of damaged components, solving a multi-objective distribution network reconfiguration model to obtain the optimal reconfiguration scheme, implementing it, quantitatively evaluating the scheme's effectiveness, archiving all process data, optimizing the parameters of the integrated model, pre-positioning model, and decomposition algorithm, and constructing a knowledge base to enable rapid disaster response. This includes: The Analytic Hierarchy Process (AHP) is used to prioritize the repair of damaged components. Then, the optimal distribution network reconfiguration scheme is obtained by solving the dual-objective distribution network reconfiguration model that minimizes network loss and maximizes load recovery rate. The scheme is then implemented to obtain the corresponding load recovery rate for emergency resource recovery and reuse planning, thereby enabling the distribution network to smoothly transition to normal operation. An evaluation index system is constructed from four dimensions: power restoration efficiency, emergency resource utilization efficiency, economic cost, and user satisfaction, to quantify the effectiveness of the implementation plan. Archive data from the entire process and optimize the parameters of the integrated model, pre-positioning model, and decomposition algorithm. Based on the evaluation results, extract the optimal response strategies for typical scenarios and build a knowledge base to support rapid response to subsequent disasters.

8. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 7, characterized in that: The method employs the analytic hierarchy process (AHP) to prioritize the repair of damaged components, then solves a dual-objective distribution network reconfiguration model that minimizes network loss and maximizes load recovery rate to obtain the optimal distribution network reconfiguration scheme, which is then implemented. The corresponding load recovery rate is obtained for emergency resource recovery and reuse planning, enabling the distribution network to smoothly transition to normal operation. Specifically, this includes: An evaluation system was constructed using the analytic hierarchy process (AHP). The evaluation criteria layer contained four dimensions: scope of failure impact, load importance, repair difficulty, and material availability. Each damaged component was scored according to these four dimensions, and a weighted score was calculated. The scores were then sorted to generate a corresponding repair priority list. A dual-objective distribution network reconfiguration model is constructed to minimize network losses and maximize load recovery rate. The decision variables are line switch status and distributed generation (DG) output. The constraints include: main grid power supply capacity limit, DG output upper limit constraint, and line radial constraint. Based on the repair priority list, solve the dual-objective distribution network reconfiguration model, output the optimal distribution network reconfiguration scheme and execute it; Obtain the load recovery rate after the implementation of the optimal distribution network reconfiguration scheme, determine the timing of resource recovery based on the load recovery rate for emergency resource recovery, formulate a recovery route based on the shortest path from the current location to the warehouse, detect the resource status and carry out reuse planning, update the resource database data, and record the status and next available time of each resource. Output a repair priority list, a power distribution network reconstruction plan, and an emergency resource recovery list.

9. The method for improving the resilience of a distribution network through three-stage multi-resource collaboration according to claim 7, characterized in that: The indicators for power restoration efficiency include average system outage time, average number of system outages, critical load restoration time, and total load restoration rate. The indicators for the emergency resource utilization efficiency dimension include mobile energy storage utilization rate, effective working rate of emergency repair teams, and resource idle rate. The indicators in the economic cost dimension include total repair cost, power outage loss cost, and cost saving rate; The user satisfaction dimension is a weighted average of scores from three dimensions: timely recovery, power supply stability, and communication transparency.

10. A three-stage multi-resource collaborative distribution network resilience enhancement system, operating the method described in any one of claims 1-9, characterized in that, The system includes: The disaster prevention resource coordination and pre-positioning module is used to generate typical disaster scenarios and evaluate load priorities based on the LSTM-AdaBoost ensemble model. Based on the typical disaster scenarios and load priorities, a pre-positioning model is constructed and the emergency site pre-positioning scheme is obtained and pre-deployed. The disaster dynamic collaborative scheduling module is used to acquire and merge data on distribution network faults, traffic network status and emergency resource status in real time to obtain a real-time status dataset. Combined with a multi-resource collaborative constraint system, the decomposition algorithm is used to decompose the disaster dynamic scheduling optimization problem into a discrete decision master problem and a continuous optimization sub-problem and solve them iteratively, dynamically generating and executing scheduling strategies. The post-disaster recovery optimization and iteration module is used to prioritize the repair of damaged components, solve the multi-objective distribution network reconstruction model to obtain the optimal distribution network reconstruction scheme and implement it, quantitatively evaluate the implementation effect of the scheme, archive the full-process data and optimize the parameters of the integrated model, pre-positioning model and decomposition algorithm, and build a knowledge base to achieve rapid response to subsequent disasters.

Images