Electric power emergency resource cooperative scheduling method and device

Abstract

The invention provides an electric power emergency resource cooperative scheduling method and device, and relates to the technical field of electric power system emergency management and intelligent optimization scheduling, and the method comprises the steps: firstly, fusing the multi-source data of meteorology, hydrology, electric power and the like, predicting disaster evolution, and quantifying the power failure influence; secondly, constructing an urban integrated network integrating electric power facilities and roads, and dividing emergency service subareas based on node importance; thirdly, establishing a pre-stationing optimization model with minimization of response time, power failure loss and operation cost as targets before a disaster, and solving the optimal positions, the number and the inventory ratio of the storage points and the mobile power supply gathering points; and finally, updating information in real time during a disaster period, solving the scheduling model considering multiple constraints, and generating an optimal scheduling scheme. According to the method and the device, through cooperation of pre-distribution before the disaster and intelligent scheduling in the disaster, fine management of emergency resources is realized, power failure loss can be effectively reduced, and emergency response efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of power system emergency management and intelligent optimization scheduling technology, specifically to a method and apparatus for collaborative scheduling of power emergency resources. Background Technology

[0002] With the continuous expansion of urban power grids and the high concentration of loads, sudden natural disasters such as typhoons, rainstorms, and floods can easily trigger large-scale power outages, seriously threatening urban operational safety, public safety, and economic and social stability. To address such risks, power companies currently generally adopt an emergency management model that combines static material deployment with fixed-quota reserves, supplemented by a single-objective dispatching method based on vehicle routing problems. This involves pre-setting fixed storage points and configuring conventional emergency supplies in several power supply areas based on historical experience or load levels. During a disaster, the deployment of mobile power vehicles and the distribution of supplies are abstracted into a path planning problem with the goal of "shortest path" or "minimum transportation cost," relying on offline solutions to generate dispatching instructions. While this model has a certain degree of operability, its core logic is based on the assumptions of known disaster conditions, static demand, and closed boundaries. It does not incorporate real-time disaster evolution information, nor does it link urban spatial structure and user attribute characteristics, making it unable to adapt to rapid changes in disaster conditions. Summary of the Invention

[0003] This application provides a method and apparatus for collaborative scheduling of power emergency resources, which enables the scheduling of power emergency resources to adapt to rapid changes in disaster situations and enhances the resilience of the power grid in the face of sudden disasters.

[0004] To achieve the above objectives, this application provides the following technical solution: The first aspect of this application provides a method for coordinated scheduling of power emergency resources, including the following steps: S10 acquires and integrates meteorological data, hydrological data, power operation data, on-site sensor data, remote sensing image data, and historical disaster event database, and performs time alignment, spatial registration, missing value completion, and outlier removal to form a unified data warehouse; S20, based on the data warehouse, predict the evolution of disasters within a preset time period in the future, and map the disaster prediction results to the failure probability and power outage probability of power equipment; based on the failure probability and power outage probability combined with the power distribution network topology, determine the expected number of users experiencing power outages, the amount of power outage, and the duration of the power outage; S30. Construct an integrated urban network with substations, switching stations, transformer substations, and pre-set important users as nodes and power lines and roads as edges. Assign each node attributes such as population, load level, user type, and historical power outage loss. Determine the importance of each node based on its attributes and perform clustering based on the importance of each node to divide the integrated urban network into several emergency service sub-regions. S40. Based on the expected power outage impact and the node attributes of each emergency service sub-area, determine the disaster risk weight and material demand weight of each emergency service sub-area. Based on the disaster risk weight and material demand weight of each emergency service sub-area, construct a pre-deployment optimization model with the goal of minimizing response time, power outage losses, and construction and operation costs. Solve the pre-deployment optimization model to obtain the location, quantity, and inventory ratio of the optimal storage point and mobile power supply aggregation point. S50, after a disaster occurs or is warned, the disaster prediction results and expected power outage impact are updated in real time, and the optimal scheduling scheme is obtained by solving the resource collaborative scheduling model; wherein, the resource collaborative scheduling model aims to minimize the total cost of power outage loss, travel time and scheduling response time, and emergency power capacity redundancy, and considers the constraints of emergency power capacity, emergency material inventory and road access.

[0005] In some embodiments, in S20, mapping the disaster prediction results to the failure probability and power outage probability of power equipment includes: Based on the tolerance parameters of power equipment, logistic regression or tree models are used to map the rainfall, wind speed and water level included in the predicted disaster evolution into the failure probability and power outage probability of power equipment.

[0006] In some embodiments, in step S30, the node importance I_i is calculated using a weighted sum formula: I_i = α × L_i + β × H_i + γ × D_i + δ × C_i Where L_i is the load level weight, H_i is the user type risk coefficient, D_i is the historical average power outage loss, C_i is the number of people associated with the node, and α, β, γ, and δ are adjustable empirical coefficients.

[0007] In some embodiments, in step S30, the clustering adopts the K-medoids clustering algorithm, using node importance and the shortest path distance between nodes as joint similarity measures, to divide the urban integrated network into several emergency service sub-regions. After iterative convergence, it ensures that the average power path between nodes in each sub-region is ≤ a preset distance and the average road travel time is ≤ a preset time, so that the nodes in each sub-region are closely coupled in terms of road travel and power supply structure; the shortest path distance between nodes is the weighted sum of power path length and road travel time.

[0008] In some embodiments, in step S40, determining the disaster risk weight and material demand weight of each emergency service sub-region based on the expected power outage impact and the node attributes of each emergency service sub-region includes: Taking the emergency service sub-zone as the evaluation object, historical disaster data, disaster prediction results and important user density are selected as indicators. By calculating the similarity coefficient between each indicator and the preset ideal plan, the disaster risk weight of the emergency service sub-zone is generated based on the similarity coefficient. The material demand weights for each emergency service sub-region are obtained based on the expected power outage volume and importance of each node in each emergency service sub-region.

[0009] In some embodiments, step S40, which involves constructing a pre-deployment optimization model based on the disaster risk weights and material demand weights of each emergency service sub-zone to minimize response time, power outage losses, and operating costs, includes: The pre-placement optimization model is as follows: MinΣf(x1)+ Σ{λ1×f(x2)+ λ2×f(x3)}; Where f(x1) represents the construction and operation costs corresponding to the storage point and assembly point setup scheme; f(x2) represents the weighted average response time from the emergency service sub-zone to the nearest storage point and assembly point corresponding to the storage point and assembly point setup scheme; f(x3) represents the expected power outage loss of the emergency service sub-zone corresponding to the storage point and assembly point setup scheme; λ1 represents the disaster risk weight of the emergency service sub-zone; and λ2 represents the material demand weight of the emergency service sub-zone.

[0010] In some embodiments, in S50, the real-time update refers to updating the disaster prediction results and the expected power outage impact at a time interval of 15 to 30 minutes, and resolving the resource collaborative scheduling model.

[0011] In some embodiments, in S50, the decision variables of the resource collaborative scheduling model include: the number and capacity of mobile power vehicles dispatched to each area to be rescued, the type and quantity of materials sent from each storage point to the repair site, and the specific driving routes and operation sequences of the corresponding vehicles.

[0012] In some embodiments, the constraints in S50 include at least: the upper limit of the capacity of each mobile power vehicle, that a single vehicle can only serve one repair point at a time, that the power shortage of each user awaiting rescue must be fully covered, that the inventory at the storage point does not exceed the upper limit, and that the driving route meets the road traffic conditions.

[0013] The second aspect of this application provides a power emergency resource collaborative scheduling device, including a module for performing the method described in the first aspect of this application.

[0014] The beneficial effects of the method and apparatus of this application are as follows: It completely overturns the traditional static and isolated emergency management model, constructing a dynamic and intelligent collaborative scheduling closed loop. Compared to the old model that relies solely on historical experience and single-path planning, the method and apparatus of this application, by integrating multi-source data and predicting disaster evolution in real time, transforms emergency decision-making from "passive response" to "proactive prediction"; by constructing a comprehensive urban network and assessing node importance, it achieves a resource allocation shift from "average distribution" to "precise tilting"; and by establishing a multi-objective pre-deployment and scheduling model, it achieves a comprehensive optimization from "pursuing efficiency" to "balancing response time, power outage losses, and operating costs." In summary, the full-chain optimization of this application, from pre-disaster preparation to in-disaster scheduling, can adapt to rapidly changing disaster situations, significantly improve the resilience and recovery capability of the power grid in the face of sudden disasters, and minimize the safety risks and economic losses of urban operations. Attached Figure Description

[0015] Figure 1 This is a flowchart of a power emergency resource collaborative scheduling method according to an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of an urban integrated network in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions in the embodiments of this invention / invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this invention / invention, and not all embodiments. Based on the embodiments of this invention / invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention / invention.

[0018] like Figure 1 As shown, one embodiment of this application provides a method for coordinated scheduling of power emergency resources, including the following steps: S10 acquires and integrates meteorological data, hydrological data, power operation data, on-site sensor data, remote sensing image data, and historical disaster event database, and performs time alignment, spatial registration, missing value completion, and outlier removal to form a unified data warehouse; Specifically, meteorological data includes rainfall, wind speed, wind direction, temperature, and humidity, collected in real time by automatic weather stations deployed in key river basins, low-lying coastal areas, and around substations, with a sampling frequency of up to 5 minutes; hydrological data includes water level, flow rate, and reservoir capacity, sourced from hydrological stations of the water resources department and IoT water level sensors deployed in rivers and pumping stations, with a data update interval of up to 10 minutes; power operation data includes switch status, load curves, voltage / current amplitude, and fault alarm signals, taken from the Supervisory Control and Data Acquisition (SCADA) system and the Distribution Automation System (DMS), with timestamp accuracy... The data is measured in seconds; on-site sensing data includes tower tilt angle, soil moisture content, and geological displacement, acquired by MEMS tilt sensors and TDR time domain reflectometers installed on the foundations of transmission towers and mountain slopes; remote sensing image data uses multispectral satellite imagery, and a deep learning segmentation model is used to identify water bodies / flooded areas, forming a high-resolution water / flood probability grid; the historical disaster event database contains the spatiotemporal range of typhoons, rainstorms, and floods in the past 10 years, a list of faulty equipment, statistics of users experiencing power outages, repair time, types and quantities of emergency supplies mobilized, mobile power vehicle dispatch records, and power restoration time.

[0019] Time alignment aims to address the issue of inconsistent data collection frequencies from different sources. Through techniques such as interpolation or resampling, all data is unified onto a common time reference to ensure joint analysis of multi-source data at the same point in time. Spatial registration unifies the spatial reference system of different data. By transforming coordinates, all geographically related data layers (such as power facilities, road networks, and disaster impact areas) are overlaid onto the same standard map, which is a prerequisite for spatial analysis. Missing value imputation addresses data loss caused by data transmission interruptions or sensor malfunctions. It uses methods such as historical mean, previous and next values, or predictions based on machine learning models to fill in missing values, ensuring the integrity of the dataset. Outlier removal identifies and handles data points that deviate significantly from the normal range due to equipment failure or transmission errors. These errors are typically discovered and corrected using statistical methods (such as the 3σ criterion) or physical rules to avoid interfering with the model.

[0020] A unified data warehouse is a centralized, standardized, high-quality data storage repository formed after all the above processing. It stores all data in association according to a unified model, providing stable and reliable data services for subsequent steps such as disaster prediction, network construction, and optimized scheduling.

[0021] S20, based on the data warehouse, predict the evolution of disasters within a preset time period in the future, and map the disaster prediction results to the failure probability and power outage probability of power equipment; based on the failure probability and power outage probability combined with the power distribution network topology, determine the expected number of users experiencing power outages, the amount of power outage, and the duration of the power outage; Specifically, predicting disaster evolution within a preset time period based on the data warehouse includes: using an incremental random forest model to make short-term predictions of rainfall intensity, water level changes, and wind speed and direction for the next 0.5 to 6 hours; and introducing an ARMA model to compensate for the intensity and duration of similar disasters in previous years. For example, suppose it is currently 2 PM, the center of Typhoon Haiyan is about 100 kilometers from the city's coastline, and the power emergency system has been activated. First, short-term predictions are made using incremental random forest, incremental support vector machine, or echo state network, receiving and analyzing the latest data entering the data warehouse in real time: for example, in the past 30 minutes, the wind speed at the coastal meteorological station has suddenly increased from 25 m / s to 35 m / s, and the radar image shows a strong rain cloud rapidly moving towards the substation in area A of the city. Based on these real-time changes, the model predicts: "Within the next 2 hours, the rainfall intensity in City A will reach 80 mm / h, the instantaneous wind speed may exceed 45 m / s, and the river level will rise by 1 meter." The prediction is highly detailed and dynamic; if the wind speed increase slows down after 15 minutes, the model will immediately absorb this new information and adjust subsequent predictions. However, this short-term prediction alone may not be sufficient to grasp the overall trend of the typhoon. Therefore, the ARMA model is also introduced, and a database of historical disaster events is consulted. It was found that Typhoon Matsa from 10 years ago is highly similar to the current Typhoon Haiyan in terms of path, intensity, and season. Based on historical data from Typhoon Matsa, the grey model derives a macro trend: "The average duration of the impact of this type of typhoon in the region is 18 hours, and it usually reaches its peak intensity 3 hours after landfall, then slowly weakens." The system then uses this "18-hour total impact duration" and "weakening after 3 hours" trend information as compensation to correct and calibrate the aforementioned short-term prediction. Ultimately, the system's comprehensive prediction is that the most severe period of the disaster will be from now until the next 6 hours (based on the incremental model), but the overall impact will last for about 18 hours, and the intensity will gradually decrease after 6 hours (based on the gray model compensation), thus providing a precise and forward-looking decision-making basis for the allocation of emergency resources.

[0022] Based on the power equipment vulnerability model, the predicted meteorological and hydrological factors are mapped to the failure probability and power outage probability of power equipment such as power lines and substations. Specifically, the power equipment vulnerability model is a nonlinear mapping function. The input is the predicted environmental factors (rainfall, water level, wind speed) and the inherent tolerance parameters of the equipment (such as the maximum allowable wind speed of 28 m / s for 110 kV overhead lines, the maximum allowable water depth of 0.3 m for box-type substations, and the maximum allowable humidity of 95% RH for GIS equipment). The output is the equipment failure probability and the downstream power outage probability caused by the failure of the equipment.

[0023] Combining the distribution network topology and the distribution of important users, the expected number of users experiencing power outages, the expected power outage amount, and the duration of power outages for each transformer substation / feeder are calculated during the predicted period. The distribution network topology data is obtained from a GIS geographic information system, including the hierarchical connection relationship between substations, feeders, transformer substations, and users, as well as line impedance, current carrying capacity, and protection action time limits. Monte Carlo simulation is used to perform 10,000 independent samplings: each sampling randomly determines whether a device has failed based on the failure probability of each device output by S20, and then traces the power supply path upstream along the topology and propagates the power outage range downstream. The start time, duration, number of affected users, and load of power outages for each transformer substation during the simulation period are statistically analyzed. Finally, the expected value (mean) is output as the expected number of users experiencing power outages, the power outage amount, and the duration of power outages.

[0024] S30. Construct an integrated urban network with substations, switching stations, transformer substations, and pre-set important users as nodes and power lines and roads as edges. Assign each node attributes such as population, load level, user type, and historical power outage loss. Determine the importance of each node based on its attributes and perform clustering based on the importance of each node to divide the integrated urban network into several emergency service sub-regions. Specifically, the node set encompasses four types of entities: substations (labeled with power supply radius, main transformer capacity, and number of important users), switching stations (labeled with the number of interconnecting switches and transfer capacity), transformer substations (labeled with substation number, number of residential households, number of commercial households, and a list of important users connected), and pre-defined important users (including hospitals, subway hubs, communication equipment rooms, military units, government agencies, large data centers, etc., labeled with user ID, load level, historical power outage loss amount, and life safety risk coefficient); the edge set contains two types of relationships: power edges represent physical electrical connections (e.g., "110 kVA substation—10 kV B feeder"), weighted as the normalized product of line length (km) and rated capacity (MVA); road edges represent vehicular accessibility (e.g., "Hospital A—Warehouse C"), weighted as the average travel time (min) or the real-time traffic delay index returned by the Gaode / Baidu Maps API. The urban integrated network, such as... Figure 2 As shown.

[0025] S40. Based on the expected power outage impact and the node attributes of each emergency service sub-area, determine the disaster risk weight and material demand weight of each emergency service sub-area. Based on the disaster risk weight and material demand weight of each emergency service sub-area, construct a pre-deployment optimization model with the goal of minimizing response time, power outage losses and operating costs. Solve the pre-deployment optimization model to obtain the location, quantity and inventory ratio of the optimal storage point and mobile power supply aggregation point. Specifically, the pre-location optimization model can be solved using either the Immune Algorithm or the Particle Swarm Optimization (PSO) algorithm. The Immune Algorithm simulates a biological immune response mechanism, encoding candidate site selection schemes as antibodies and iteratively evolving through affinity selection (fitness = reciprocal of the objective function), clonal amplification, and high-frequency mutation operations. The PSO algorithm encodes each particle position as a vector (x_i, y_j, z_i_k, w_j_l), guiding the search through individual and global extrema. Both algorithms support parallel computation and can solve a medium-sized problem with 50 candidate points, 8 sub-regions, and 12 types of materials within 2 hours. The output optimal solution set includes 5 storage points (e.g., distributed at the intersection of the outer ring expressway and logistics parks), 4 mobile power supply aggregation points (e.g., adjacent to main roads and emergency channels), safety stock thresholds for various materials, and a configuration scheme of 2–5 mobile power supply vehicles per point.

[0026] S50, after a disaster occurs or is warned, the disaster prediction results and expected power outage impact are updated in real time, and the optimal scheduling scheme is obtained by solving the resource collaborative scheduling model; wherein, the resource collaborative scheduling model aims to minimize the total cost of power outage loss, travel time and scheduling response time, and emergency power capacity redundancy, and considers the constraints of emergency power capacity, emergency material inventory and road access.

[0027] Specifically, real-time updates refer to rolling execution with a fixed time window of 15 to 30 minutes: at the end of each window, the latest meteorological radar echo map, real-time water level of hydrological station, newly added fault alarms, pole collapse and wire breakage points identified by drone inspection images, and power outage address tags collected by the citizen repair platform are simultaneously accessed, re-triggering step S20 prediction and step S30 network status refresh, generating a dynamically updated "set of users awaiting rescue" and its missing power, power outage loss weight, geographical coordinates and road traffic status (e.g., if a bridge is closed due to water accumulation, the corresponding road edge weight is reset to infinity). Decision variables for the resource collaborative scheduling model: a_i,j,k indicates whether the kth mobile power vehicle belonging to the i-th power supply company is dispatched to the j-th node awaiting rescue. v_m,n represents the path decision for the material vehicle from warehouse m to the emergency repair site n; The objective function of the resource collaborative scheduling model is: Minimize the sum of Σ power outage loss cost (weighted by node importance and power outage time), Σ (vehicle travel time + dispatch response time), and Σ (mobile power supply capacity redundancy); Minimize the total cost of power outage losses: Consider the power outage losses from the time the rescue request is received until the mobile power supply is in place and the supplies are delivered; Minimize travel time and dispatch response time: consider preparation time, travel time, on-site setup and dismantling time; Minimize the waste of emergency power capacity redundancy: Penalize the difference between "deployment capacity - actual shortfall" to make capacity configuration more closely match demand.

[0028] The constraints of the resource collaborative scheduling model include: The capacity of a single vehicle shall not exceed the rated capacity; each mobile power vehicle shall serve only one node at a time; the power shortage of each node shall be fully covered by one or more mobile power vehicles; the quantity of materials issued from the storage point shall not exceed the inventory; and the driving route shall meet the road traffic conditions (which can be restricted by road network weight).

[0029] The branch-and-bound algorithm can be used to solve the resource collaborative scheduling model, and obtain the vehicle dispatch plan, service node sequence and expected arrival time of each power supply company.

[0030] Let's use a concrete example to explain the solution process of the above scheduling model: Scenario: A powerful typhoon makes landfall, causing power outages at two key locations in the city. Node 1 is the city center hospital, which is of extremely high importance and has an estimated power shortage of 800kW. Node 2 is a large residential area, which is of medium importance and has an estimated power shortage of 900kW.

[0031] Available resources: Power company A has one 1000kW mobile power supply vehicle (MPS-A1) and one 500kW mobile power supply vehicle (MPS-A2); Power company B has one 800kW mobile power supply vehicle (MPS-B1); The central warehouse stores cables, fuel and other supplies needed for emergency repairs; The road condition is that the shortest road to the hospital is blocked due to flooding, and vehicles need to detour, which will increase the travel time by 30 minutes.

[0032] 1. Decision variables a_i,j,k: Should we send Company A's MPS-A1 (1000kW) to the hospital? Should we send MPS-A2 (500kW) to the residential area? Should we send Company B's MPS-B1 (800kW) to the hospital? These are all 0 or 1 decisions.

[0033] v_m,n: Should the supply trucks take a longer route to the hospital after leaving the warehouse, or wait for the floodwaters to recede? 2. Objective function The model aims to minimize a three-part "total cost," much like a scoring system. Minimizing the total cost of power outage losses: The "cost of loss" for a hospital is far greater than that for a residential area for every minute of power outage. Therefore, the model will prioritize restoring power to the hospital, even if it means that residential areas will have to wait longer.

[0034] Minimize travel time and dispatch response time: The model aims for all vehicles to arrive and start working as quickly as possible; preparation time, the 30 minutes added due to detours, and on-site setup time are all included, and the shorter the total time, the better.

[0035] Minimize the waste of emergency power capacity redundancy: If a 1000kW vehicle is sent to a hospital that only needs 800kW, the extra 200kW is a "waste". The model will therefore "deduct points" from this solution, striving for the perfect match.

[0036] 3. Constraints No matter how cost-effective or time-saving a solution may be, it cannot violate the following strict rules: Capacity rule: A 500kW vehicle cannot be sent to address a 900kW capacity shortage because it cannot handle the load.

[0037] One vehicle, one location rule: If an MPS-A1 is dispatched to a hospital, it cannot simultaneously go to a residential area.

[0038] The demand must be met: the 800kW gap in the hospital and the 900kW gap in the residential area must be fully covered, which can be accomplished by one or more vehicles working together.

[0039] Inventory rule: There are only 10 rolls of cable in the warehouse, so 12 rolls cannot be sent out.

[0040] Road condition rules: Vehicles cannot travel on the flooded road, and the path decision v_m,n must be feasible.

[0041] Model solution and final solution: The model quickly calculates thousands of possible combinations (e.g., A1 goes to the hospital, B1 goes to the residential area; or A1 goes to the residential area, B1 goes to the hospital, etc.) and calculates a "total cost" score for each option.

[0042] Option 1: Send MPS-A1 (1000kW) to the hospital, and MPS-A2 (500kW) and MPS-B1 (800kW) to the residential area. The demand in the residential area is met, but the hospital has a 200kW redundancy and the residential area has a 400kW redundancy, for a total redundancy of 600kW, which is a low score.

[0043] Option 2: Send an MPS-B1 (800kW) to the hospital and an MPS-A1 (1000kW) to the residential area. The hospital is a perfect match with zero redundancy! The residential area has 100kW of redundancy, for a total redundancy of 100kW. Although the detour increases the time, this option scores highly in terms of "power outage loss" and "capacity redundancy" because of the hospital's extremely high importance.

[0044] Ultimately, the model will determine that Scheme 2 is the "optimal scheduling scheme" through the branch and bound algorithm optimization tool.

[0045] The output is as follows: Power company B dispatches MPS-B1 to the hospital; power company A dispatches MPS-A1 to the residential area. The service node sequence is: MPS-B1's destination is the hospital, and MPS-A1's destination is the residential area. The estimated arrival times are: MPS-B1 is expected to arrive at the hospital in 45 minutes (due to detour); MPS-A1 is expected to arrive at the residential area in 20 minutes.

[0046] In some embodiments, in S20, mapping the disaster prediction results to the failure probability and power outage probability of power equipment includes: Based on the tolerance parameters of power equipment, logistic regression or tree models are used to map the rainfall, wind speed and water level included in the predicted disaster evolution into the failure probability and power outage probability of power equipment.

[0047] Specifically, the power equipment vulnerability model is a quantitative assessment model for key equipment on the distribution side (including but not limited to 10kV overhead line insulators, pole-mounted transformers, ring main units, cable intermediate joints, and underground substations). Its function is to transform multidimensional environmental stress variables into equipment-level failure risk indicators. This model does not rely on a single threshold criterion or empirical formula, but learns the nonlinear response relationship between the physical degradation mechanism of equipment and external disaster factors through a data-driven approach. The model output generates two parallel probability values: one is the probability of a single point of failure (P_fault ∈ [0,1]), which represents the possibility of mechanical damage, insulation breakdown, or structural instability of the equipment under a given environmental stress; the other is the probability of local power outage caused by this (P_outage ∈ [0,1]), which reflects the joint probability of continuous power outage of the transformer area / feeder supplied by the equipment after its failure. This value has been coupled with system-level factors such as downstream protection action logic, backup power supply switching success rate, and network reconfiguration capability.

[0048] The tolerance parameters of the equipment are the extreme operating boundary parameters provided by the equipment manufacturer or calibrated through type testing. Specifically, these include: the wet flashover voltage threshold and the maximum allowable wind deflection angle of the overhead line insulators; the protection level (IP code) of the pole-mounted transformer enclosure and the maximum allowable immersion depth; the lower limit of the air chamber pressure of the ring main unit and the tolerance threshold of SF6 leakage rate; the elevation of the ventilation openings and the height of the flood control baffle in the underground substation; and the sealing pressure resistance level and thermal cycling life decay coefficient of the cable intermediate joint. These parameters are embedded in the feature vector of the model training samples in the form of structured fields, serving as key identifiers for the model to distinguish the vulnerability differences of different equipment types. For example, under the same rainfall intensity, there is a significant difference in the failure probability prediction results between the IP65 protection level ring main unit and the IP44 level equipment. As another example, for underground substations with an elevation 0.8m lower than the historical highest flood level, the probability of power outage shows an S-shaped steep increase within the range of ±0.2m of the predicted water level. This characteristic is automatically captured as a split node by the tree model.

[0049] Logistic regression, as a generalized linear model, is configured as a binary classification probability estimator in this embodiment. Its input feature vector X includes standardized rainfall, wind speed, water level, and their interactions (e.g., rainfall × wind speed, water level²), equipment tolerance parameter encoding (One-Hot or ordinal encoding), and geographical weighting factors (e.g., terrain slope, distance from the river). The model solves for the weight vector W and bias term b through maximum likelihood estimation, resulting in an output... It directly approximates the true fault probability label; this implementation method has strong interpretability, the absolute value of each feature coefficient reflects its marginal contribution to fault risk, and the sign indicates the direction of positive and negative impact, which makes it easy for operation and maintenance personnel to understand "why the risk of a certain area has increased sharply"; at the same time, it supports online incremental updates: when new fault event data arrives, the stochastic gradient descent (SGD) algorithm only needs to traverse the new samples once to fine-tune W and b, without the need for full retraining.

[0050] Tree models generally refer to tree-based machine learning models such as Decision Tree, Random Forest, Gradient Boosting Tree (GBDT), or XGBoost. In this embodiment, XGBoost is selected as the backbone model because it has better generalization ability and robustness in scenarios with small samples and high noise power field data. The model constructs a training set with equipment ID as the granularity. Each sample contains: feature columns (rainfall, wind speed, water level, equipment type, years of operation, number of historical defects, surrounding vegetation density, soil resistivity) and label columns (whether a fault occurred, whether it caused a power outage). XGBoost constructs weak learners (CART trees) through multiple rounds of iteration. Each tree learns the residuals of the previous model, and finally, the weighted ensemble outputs P_fault and P_outage. This model naturally supports nonlinear relationship modeling and threshold effect recognition—for example, automatically detecting "when wind speed > 18.5". When the speed is m / s and the tower tilt angle is >3.2°, the probability of line breakage jumps to 0.73. Such rules do not need to be manually set and can be directly exported as an operation and maintenance early warning rule library. In addition, the model has a built-in feature importance ranking module, which can dynamically identify the main disaster-causing factors under the current disaster type (such as wind speed having the highest weight during typhoons and water level having the highest weight during rainstorms), providing a basis for differentiated reinforcement strategies.

[0051] In some embodiments, in step S30, the node importance I_i is calculated using a weighted sum formula: I_i = α × L_i + β × H_i + γ × D_i + δ × C_i Wherein, Li is the load level weight (e.g., Special Grade = 1.0, Grade 1 = 0.8), H_i is the user type risk coefficient (e.g., Hospital = 1.2, Subway = 1.1, Military Industry = 1.05, Data Center = 0.95), D_i is the historical average power outage loss (ten thousand yuan / hour), C_i is the number of people associated with this node (ten thousand people), and α, β, γ, δ are adjustable empirical coefficients (default values ​​are 0.25, 0.3, 0.3, 0.15), which can be dynamically adjusted according to the city's functional positioning.

[0052] In some embodiments, in step S30, the clustering adopts the K-medoids clustering algorithm, using node importance and the shortest path distance between nodes as joint similarity measures, to divide the urban integrated network into several emergency service sub-regions. After iterative convergence, it ensures that the average power path between nodes in each sub-region is ≤ a preset distance and the average road travel time is ≤ a preset time, so that the nodes in each sub-region are closely coupled in terms of road travel and power supply structure; the shortest path distance between nodes is the weighted sum of power path length and road travel time.

[0053] Specifically, the emergency service sub-zone is a functional spatial unit oriented towards power emergency response. Its boundary does not depend on the administrative jurisdiction or power supply business area division, but is naturally derived from the K-medoids clustering results. Each sub-zone contains at least one substation or switching station as a power support node, two or more transformer substations as load-bearing nodes, and no less than one pre-set important user (such as a tertiary hospital, a subway control center, or a provincial government cloud data center). The nodes within the sub-zone meet the following requirements: the maximum number of hops between any two nodes on the power path is ≤3, the average travel time on the road path is ≤15 minutes, and the overall power supply path dependence of the sub-zone (i.e., the probability that the failure of a critical node will cause ≥60% of the load in the sub-zone to lose power) is ≤0.15. K-medoids clustering is a partitioning clustering method that uses actual data points (medoids) as cluster centers. Unlike K-means, which uses virtual centroids, K-medoids selects k real nodes from the original node set as representative centers (medoids) for each cluster and optimizes the clustering results by minimizing the total distance from all nodes to their respective medoids. This algorithm is suitable for non-Euclidean distance metrics, especially for distances between nodes in urban networks defined by shortest path length, connectivity reliability, or topological similarity. In this embodiment, the distance between nodes is calculated using a weighted composite distance function. ,in This represents the importance distance between nodes i and j. This represents the shortest path distance between nodes i and j. , This is an adjustable weighting coefficient, with a value range of [0.3, 0.7], which can be selected. =0.5、 =0.5, to achieve a balanced representation of the power structure and transportation structure; this distance measurement method avoids the problem that simple geographical Euclidean distance cannot reflect the real-world constraints such as power grid islands, single-line power supply, and dead-end roads; The specific execution flow of the K-medoids clustering algorithm includes: initialization phase, randomly selecting k nodes from all N nodes as initial medoids; allocation phase, for each non-medoid node i, calculating its distance to the k medoids, and allocating it to the cluster corresponding to the nearest medoid; update phase, for each cluster, traversing all nodes in the cluster, trying each node as a candidate medoid, recalculating the total distance from all nodes in the cluster to the candidate medoid, and selecting the node with the smallest total distance as the new medoid; repeating the allocation and update until convergence (i.e., the medoid set no longer changes or the change in the objective function is less than the threshold); after the algorithm terminates, it outputs the k medoid nodes and the list of members of their respective subregions.

[0054] In some embodiments, in step S40, determining the disaster risk weight and material demand weight of each emergency service sub-region based on the expected power outage impact and the node attributes of each emergency service sub-region includes: Taking the emergency service sub-zone as the evaluation object, historical disaster data, disaster prediction results and important user density are selected as indicators. By calculating the similarity coefficient between each indicator and the preset ideal plan, the disaster risk weight of the emergency service sub-zone is generated based on the similarity coefficient. The material demand weights for each emergency service sub-region are obtained based on the expected power outage volume and importance of each node in each emergency service sub-region.

[0055] Specifically, to scientifically assess the comprehensive risks faced by each emergency service sub-region, this method treats each sub-region as an independent evaluation object and comprehensively considers three core indicators: first, "historical disaster data" reflecting its past disaster situation; second, "disaster prediction results" measuring the potential threat of the current disaster; and third, "density of important users" reflecting its social criticality. A "worst-case scenario" representing the worst-case situation is preset as a benchmark. The risk level of each actual sub-region is quantified by calculating the similarity coefficient between its indicators and this "worst-case scenario." The closer a sub-region's indicators are to this worst-case scenario, the higher its similarity coefficient, and the greater the resulting disaster risk weight, indicating that the region is a key target for emergency resource allocation and defense.

[0056] To quantify the actual scale and urgency of emergency resources needed by each emergency service sub-zone after a disaster, the system comprehensively calculates two key indicators for all nodes within that sub-zone: first, "expected power outage volume," which directly reflects the workload required to restore power; the larger the outage volume, the more mobile power supplies and repair materials are needed. Second, "importance," representing the social value and restoration priority of the node; the restoration needs of important nodes such as hospitals and transportation hubs are far higher than those of ordinary users. By combining the "expected power outage volume" of all nodes within the sub-zone with their corresponding "importance" for weighted calculation, and quantifying the result into a value between 0 and 1, the system ultimately derives the "material demand weight" for that sub-zone. This reflects both the arduousness of the restoration task and its priority level, providing a scientific quantitative basis for the precise allocation of emergency resources during a disaster.

[0057] In some embodiments, step S40, which involves constructing a pre-deployment optimization model based on the disaster risk weights and material demand weights of each emergency service sub-zone to minimize response time, power outage losses, and operating costs, includes: The pre-placement optimization model is as follows: Min(Σf(x1)+ Σ(λ1×f(x2)+ λ2×f(x3))); Where f(x1) represents the construction and operation costs corresponding to the storage point and assembly point setup scheme; f(x2) represents the weighted average response time from the emergency service sub-zone to the nearest storage point and assembly point corresponding to the storage point and assembly point setup scheme; f(x3) represents the expected power outage loss of the emergency service sub-zone corresponding to the storage point and assembly point setup scheme; λ1 represents the disaster risk weight of the emergency service sub-zone; and λ2 represents the material demand weight of the emergency service sub-zone.

[0058] In some embodiments, in S50, the real-time update refers to updating the disaster prediction results and expected power outage impacts on a rolling basis at time intervals of 15 to 30 minutes, and resolving the resource collaborative scheduling model; the time interval is dynamically adjusted based on the disaster prediction results and expected power outage impacts.

[0059] In some embodiments, in S50, the decision variables of the resource collaborative scheduling model include: the number and capacity of mobile power vehicles dispatched to each area to be rescued, the type and quantity of materials sent from each storage point to the repair site, and the specific driving routes and operation sequences of the corresponding vehicles.

[0060] The "number of mobile power supply vehicles dispatched to each area awaiting rescue" refers to the number of mobile emergency power supply vehicles allocated to the set of users awaiting rescue (i.e., key transformer substations / feeders / important user nodes where power outages are caused by disaster prediction or actual failures and the power shortage is not covered) determined through rolling updates in S50. This number is a non-negative integer variable, denoted as . ,in Indicates the power supply company or assembly point number to which the mobile power supply vehicle belongs. This indicates the area number awaiting rescue (corresponding to a specific repair point or high-risk power outage node within the emergency service sub-zone defined in S30). This variable is subject to the maximum number of mobile power vehicles currently available to each power supply unit or assembly point, and is also dynamically constrained by road conditions (such as bridge closures or road closures due to flooding). Its function is to achieve spatial allocation of emergency power supply resources, ensuring that critical loads receive priority power support.

[0061] The "mobile power vehicle capacity" refers to the rated continuous output active power of each dispatched vehicle, measured in megawatts (MW). Its value is a discrete continuous variable or an element from a pre-defined finite set (such as typical specifications like 0.3 MW, 0.5 MW, 1.0 MW, 2.0 MW, and 5.0 MW). In the model, it is represented as a product term coupled with the number of vehicles. ,in For the first The nominal capacity of the mobile power supply vehicle. This capacity parameter directly determines whether it can meet the power shortage coverage requirements of the area to be rescued. Its value is calibrated based on the equipment's tolerance parameters (such as short-time overload capacity and environmental temperature adaptability), on-site access conditions (such as low-voltage side interface matching and grounding resistance limits), and historical operating data (such as the actual usable capacity of a certain model decreasing to 92% of the nominal value in a humid environment).

[0062] The "material types" include, but are not limited to, emergency cables (such as YJV22-3×240+1×120 mm² cross-linked polyethylene insulated steel tape armored cables), pole-mounted switches (such as ZW32-12 / T630A vacuum circuit breakers), poles (such as 12 m prestressed concrete poles), fittings (such as NLL-2 tension clamps), flood control sandbags, portable lighting sets, and insulation tool kits. Each type of material is identified in the model with an independent category code (such as CBL-240, SWT-ZW32, POL-12M) and is strongly correlated with the equipment type and failure mode in the power distribution network topology.

[0063] Among them, "quantity of materials" is an integer decision variable for the corresponding category, denoted as... ,in Indicates the candidate storage point number. This indicates the emergency repair site number (mapped one-to-one with the areas awaiting rescue in S50). This variable is subject to two constraints: firstly, the upper limit of inventory at storage points. The upper limit is determined by the S40 pre-deployment optimization results and is deducted in real time during the disaster based on the issued documents; secondly, the demand for emergency repair tasks. The demand is calculated by combining the expected fault equipment list output by S20, the historical average emergency repair material consumption rate in the S30 node attributes (such as the average replacement of a certain type of switch requires 6 bolts and 2 sealing rings), and the degree of damage fed back by field sensor data (such as the need to add 4 sets of reinforcement clamps when the tower tilt angle is greater than 5°).

[0064] The "specific travel route" refers to the optimal spatial trajectory from the starting point (warehouse or mobile power supply assembly point) to the destination (repair site). It is modeled based on the road network structure, and each feasible path is represented as an edge sequence. Each edge For the corresponding actual road segment, its attributes include length, average travel time, real-time congestion index, and weight / height / flood status (determined by traffic management department API or vehicle GPS + visual recognition fusion). The route decision variable is a binary variable. This indicates whether the supply vehicle or power vehicle has selected the first option. Candidate paths from node To the node This variable is coupled with the time dimension to form a spatiotemporal path network, supporting multi-vehicle conflict avoidance (such as prohibiting two heavy power vehicles from entering a narrow alley with a width of less than 4.5 m at the same time).

[0065] Among them, "operation sequence" refers to the spatiotemporal execution sequence of the same vehicle serving multiple emergency repair sites within a single dispatch cycle, which is modeled as a permutation variable. , indicating the first The vehicles served in sequence Each site is numbered. This order not only affects the total mileage but also directly impacts the cumulative power outage losses—for example, supplying power to hospitals first reduces the weighted outage time under the risk to life, while supplying power to commercial areas later reduces the weighted loss under the risk to economic losses. The order decision is constrained by the latest allowed arrival time (Time Window Constraint) for each site, which is determined by the user type: hospital ICUs require mobile power vehicles to connect to the grid within 30 minutes of an alarm, while ordinary residential areas can have this extended to 120 minutes; it is also constrained by vehicle operation time (e.g., grid connection and commissioning of the power vehicle takes an average of 18 minutes, and cable laying takes 45 minutes per kilometer).

[0066] The objective function of the scheduling model includes: Minimize the total cost of power outage losses: Consider the power outage losses from the time the rescue request is received until the mobile power supply is in place and the supplies are delivered; Minimize travel time and dispatch response time: consider preparation time, travel time, on-site setup and dismantling time; Minimize the waste of emergency power capacity redundancy: Penalize the difference between "deployment capacity - actual shortfall" to make capacity configuration more closely match demand.

[0067] In some embodiments, the constraints in S50 include at least: the upper limit of the capacity of each mobile power vehicle, that a single vehicle can only serve one repair point at a time, that the power shortage of each user awaiting rescue must be fully covered, that the inventory at the storage point does not exceed the upper limit, and that the driving route meets the road traffic conditions.

[0068] The upper limit of the capacity of each mobile power vehicle refers to the maximum active power or apparent capacity that each mobile power vehicle can output in a single dispatch task. Its value is determined by the vehicle model and the manufacturer's technical parameters, and the typical range is 0.3 MW to 2.0 MW. This upper limit is encoded as a linear inequality as a hard physical constraint. ,in This represents the power output allocated by the i-th mobile power supply vehicle to the k-th node awaiting rescue. Its rated capacity; this constraint prevents equipment overheating, protection tripping, or permanent damage due to overload operation.

[0069] The constraint that a single mobile power vehicle can only serve one repair point at a time means that within any given dispatch decision cycle, the same mobile power vehicle cannot simultaneously provide power to two or more geographically separated power outage areas. This constraint is implemented through logical variables and integer programming modeling: introducing binary variables... , indicating whether the i-th vehicle serves the j-th repair point. This ensures that the vehicle operation process is operable and controllable on-site, avoiding voltage fluctuations, relay protection malfunctions, and on-site coordination chaos caused by multiple parallel power supplies.

[0070] The complete coverage of the power shortage for each user awaiting rescue refers to the predicted power shortage for each identified node awaiting rescue (such as a transformer in a certain area or a distribution room of an important user). (Unit: kWh) is supplied jointly by one or more mobile power vehicles, and ,in The minimum expected power supply duration for this node is determined by the outage duration output from the Monte Carlo simulation in S20, with a typical value of 1 to 6 hours. This full coverage constraint is a strong feasibility guarantee mechanism to ensure that no user experiences continuous power outages due to insufficient power supply capacity.

[0071] Among them, the "inventory limit at storage points" means that the quantity of various materials (such as cables, switchgear, waterproof kits, and tools) stored at each activated emergency material storage point must not exceed the dual restrictions of its physical storage capacity and management quota, modeled as follows: ,in Let n be the quantity of type n materials transferred from the m-th storage point to the n-th emergency repair site. This sets the upper limit for the corresponding material category's inventory; this constraint prevents subsequent batches of emergency repairs from running out of materials due to excessive allocation.

[0072] The requirement that the driving route meets road accessibility conditions means that the planned driving routes of all material transport vehicles and mobile power supply vehicles avoid real-time inaccessible road sections, including but not limited to: road sections with water depth > 0.3 m as determined by weather warnings, slope road sections with displacement rates > 5 mm / d as shown by geological disaster monitoring, areas explicitly prohibited by traffic control orders, and submerged road grids identified by remote sensing images; this constraint is achieved by modeling the road network as a weighted directed graph G=(V,E), where the weight of edge e∈E is... (When the road segment corresponding to e is currently impassable) or (When path e is passable), and forcibly remove it during path optimization. It is achieved through the edges.

[0073] In some embodiments, the model also includes post-disaster assessment and incremental model updates, collecting data on the actual disaster situation, power outage range, repair time and power restoration effect in each region during the disaster, and assessing the model prediction error; incorporating the new data into the incremental learning framework to update the disaster prediction model in step S2 and the risk weight model in step S3 online without having to retrain from scratch.

[0074] Another embodiment of this application provides a power emergency resource collaborative scheduling device, including a module for performing the method described in the above embodiments of this application.

[0075] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used in these embodiments is chosen to best explain the principles, practical applications, or technological improvements to the embodiments in the market, or to enable other those skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for collaborative scheduling of power emergency resources, characterized in that, Includes the following steps: S10 acquires and integrates meteorological data, hydrological data, power operation data, on-site sensor data, remote sensing image data, and historical disaster event database, and performs time alignment, spatial registration, missing value completion, and outlier removal to form a unified data warehouse; S20, based on the data warehouse, predict the evolution of disasters within a preset time period in the future, and map the disaster prediction results to the failure probability and power outage probability of power equipment; Based on the fault probability and power outage probability combined with the power distribution network topology, the expected number of users experiencing power outages, the amount of power outage, and the duration of the power outage are determined. S30 constructs an integrated urban network with substations, switching stations, transformer substations, and pre-set important users as nodes, and power lines and roads as edges, and assigns attributes such as population, load level, user type, and historical power outage losses to each node; The importance of each node is determined based on its attributes, and clustering is performed based on the importance of each node to divide the urban integrated network into several emergency service sub-regions. S40. Based on the expected power outage impact and the node attributes of each emergency service sub-area, determine the disaster risk weight and material demand weight of each emergency service sub-area. Based on the disaster risk weight and material demand weight of each emergency service sub-area, construct a pre-deployment optimization model with the goal of minimizing response time, power outage losses, and construction and operation costs. Solve the pre-deployment optimization model to obtain the location, quantity, and inventory ratio of the optimal storage point and mobile power supply aggregation point. S50, after a disaster occurs or is warned, the disaster prediction results and expected power outage impact are updated in real time, and the optimal scheduling scheme is obtained by solving the resource collaborative scheduling model; wherein, the resource collaborative scheduling model aims to minimize the total cost of power outage loss, travel time and scheduling response time, and emergency power capacity redundancy, and considers the constraints of emergency power capacity, emergency material inventory and road access.

2. The method according to claim 1, characterized in that, In step S20, mapping the disaster prediction results to the failure probability and power outage probability of power equipment includes: Based on the tolerance parameters of power equipment, logistic regression or tree models are used to map the rainfall, wind speed and water level included in the predicted disaster evolution into the failure probability and power outage probability of power equipment.

3. The method according to claim 1, characterized in that, In step S30, the node importance I_i is calculated using a weighted sum formula: I_i = α × L_i + β × H_i + γ × D_i + δ × C_i Where L_i is the load level weight, H_i is the user type risk coefficient, D_i is the historical average power outage loss, C_i is the number of people associated with the node, and α, β, γ, and δ are adjustable empirical coefficients.

4. The method according to claim 1, characterized in that, In S30, the clustering adopts the K-medoids clustering algorithm, which uses node importance and the shortest path distance between nodes as joint similarity measures to divide the urban integrated network into several emergency service sub-regions. After iterative convergence, it ensures that the average power path between nodes in each sub-region is ≤ the preset distance and the average road travel time is ≤ the preset time, so that the nodes in each sub-region are closely coupled in terms of road travel and power supply structure. The shortest path distance between nodes is the weighted sum of the power path length and the road travel time.

5. The method according to claim 1, characterized in that, In step S40, determining the disaster risk weight and material demand weight of each emergency service sub-region based on the expected power outage impact and the node attributes of each emergency service sub-region includes: Taking the emergency service sub-zone as the evaluation object, historical disaster data, disaster prediction results and important user density are selected as indicators. By calculating the similarity coefficient between each indicator and the preset ideal plan, the disaster risk weight of the emergency service sub-zone is generated based on the similarity coefficient. The material demand weights for each emergency service sub-region are obtained based on the expected power outage volume and importance of each node in each emergency service sub-region.

6. The method according to claim 1, characterized in that, In step S40, the construction of a pre-deployment optimization model based on the disaster risk weight and material demand weight of each emergency service sub-zone, with the objective of minimizing response time, power outage losses, and operating costs, includes: The pre-placement optimization model is as follows: Min(Σf(x1)+ Σ(λ1×f(x2)+ λ2×f(x3))); Where f(x1) represents the construction and operation costs corresponding to the storage point and assembly point setup scheme; f(x2) represents the weighted average response time from the emergency service sub-zone to the nearest storage point and assembly point corresponding to the storage point and assembly point setup scheme; f(x3) represents the expected power outage loss of the emergency service sub-zone corresponding to the storage point and assembly point setup scheme; λ1 represents the disaster risk weight of the emergency service sub-zone; and λ2 represents the material demand weight of the emergency service sub-zone.

7. The method according to claim 1, characterized in that, In S50, the real-time update refers to updating the disaster prediction results and expected power outage impacts on a rolling basis at time intervals of 15 to 30 minutes, and resolving the resource collaborative scheduling model.

8. The method according to claim 1, characterized in that, In S50, the decision variables of the resource collaborative scheduling model include: the number and capacity of mobile power vehicles dispatched to each area awaiting rescue, the type and quantity of materials sent from each storage point to the repair site, and the specific driving routes and operation sequences of the corresponding vehicles.

9. The method according to claim 1, characterized in that, In S50, the constraints include at least the following: the upper limit of the capacity of each mobile power vehicle, each vehicle can only serve one emergency repair point at a time, the power shortage of each user waiting for rescue must be fully covered, the inventory of the storage point does not exceed the upper limit, and the driving route meets the road traffic conditions.

10. A power emergency resource collaborative dispatching device, characterized in that, Includes a module for performing the method according to any one of claims 1 to 9.

Images