Power distribution network emergency guarantee power supply resource intelligent matching method

Abstract

The invention provides a power distribution network emergency guaranteed power supply resource intelligent matching method, relates to the technical field of power distribution network emergency guaranteed power supply, and aims to solve the problem that the existing power distribution network emergency guaranteed power supply resource allocation efficiency is low. According to the method, through multi-source data acquisition and processing, an enhanced random forest algorithm is adopted to evaluate emergency event risks and perform graded early warning. A multi-layer perceptron model is used for predicting emergency resource requirements, and rapid matching of emergency resources and fault elements is realized based on a point supplementing method and a Kuhn-Mangels algorithm. And meanwhile, a transport path is planned by applying a genetic algorithm, and a double-layer optimization model is constructed to realize multi-voltage-level power grid collaborative recovery. Tests show that the method can significantly improve the emergency resource allocation efficiency and accuracy, reduce the load loss, shorten the emergency response time, effectively guarantee the efficient implementation of the emergency power supply of the power distribution network, and improve the ability of the power distribution network to deal with emergencies.

Description

Technical Field

[0001] This invention relates to the field of emergency power supply technology for power distribution networks, and more specifically, to a method for intelligent matching of emergency power supply resources for power distribution networks. Background Technology

[0002] In modern society, the stability and reliability of power supply play a vital role in the normal operation of the social economy and people's daily lives. As the link in the power system directly connected to users, the operation of the distribution network directly affects users' electricity experience. Due to the influence of various factors such as natural disasters, equipment failures, and external damage, power outages are inevitable in the distribution network. This necessitates efficient emergency power supply resource matching methods to ensure the power supply to important users and reduce the losses caused by power outages.

[0003] Currently, data acquisition in some power distribution networks relies primarily on traditional sensors and monitoring equipment, resulting in a relatively singular data source. This approach struggles to comprehensively and promptly obtain information on the network's operational status and fault conditions. In some remote areas, sensor coverage is limited, making it impossible to accurately monitor line faults. Furthermore, insufficient collection of user feedback hinders timely understanding of users' actual needs and the impact of power outages, leading to a lack of comprehensive information support during emergency response.

[0004] In the data preprocessing stage, traditional data cleaning methods often employ simple threshold judgments or statistical analyses. However, these methods cannot accurately identify and handle outliers for complex distribution network data. The traditional Raida criterion, when processing data with complex distributions, may misclassify normal data as outliers or miss genuine outliers, thus affecting the accuracy of subsequent analysis.

[0005] Most existing emergency event risk assessment models employ a single machine learning algorithm, such as decision trees or support vector machines. When dealing with complex distribution network data, these models fail to fully capture the inherent patterns and characteristics within the data, resulting in low accuracy in risk assessment. When faced with multiple fault types and complex operating conditions, a single algorithm cannot accurately assess the risk level of an emergency event, thus failing to provide a reliable basis for subsequent emergency decision-making.

[0006] Current risk assessment methods are mostly based on static data and lack dynamic assessment of the real-time operating status of the distribution network. During an emergency, the operating status of the distribution network changes rapidly, such as the expansion of the fault area and load fluctuations. Existing assessment methods cannot reflect these changes in a timely manner, resulting in delayed risk assessment results that fail to meet the timeliness requirements of emergency response.

[0007] Existing emergency resource forecasting methods primarily rely on historical data and empirical models, lacking consideration of real-time data and dynamic factors. When forecasting emergency resource demand, they fail to adequately account for the real-time development of events and dynamic changes in user needs, leading to significant discrepancies between forecasts and actual demand. This can result in insufficient or wasted emergency resource reserves, impacting the effectiveness of emergency power supply.

[0008] In emergency resource allocation, traditional methods often employ simple proximity-based allocation principles or fixed matching strategies, failing to comprehensively consider factors such as resource capacity, transportation costs, and repair time. This approach cannot achieve optimal resource allocation, leading to low resource utilization efficiency. In some cases, although nearby emergency resources may be selected, their capacity may be insufficient to meet actual needs, or transportation costs may be too high, thus impacting the efficiency and economy of emergency power supply.

[0009] Most existing transportation route planning methods use static planning, which does not take into account dynamic factors in the actual transportation process, such as traffic congestion and road construction. In the event of an emergency, traffic conditions can change drastically. If a statically planned route is used, it will lead to longer transportation times, or even failure to deliver emergency resources to the fault location on time.

[0010] Traditional transportation route planning methods primarily consider route length while neglecting other crucial factors such as transportation costs, transit time, and resource depletion. This can lead to situations where, despite selecting the shortest route, transportation costs are excessively high or transit times are too long, failing to meet emergency power supply requirements.

[0011] Currently, multi-voltage-level power grids lack effective coordination mechanisms during emergency restoration. High-voltage and medium-voltage distribution networks often operate independently, without adequately considering their mutual influence and coordination. This can lead to conflicts in the restoration processes between different voltage levels, affecting the overall restoration efficiency of the distribution network.

[0012] During power grid restoration, existing methods lack a real-time feedback mechanism, making it impossible to adjust restoration strategies promptly based on the actual restoration situation. When the load restoration limit provided by the high-voltage distribution network to the medium-voltage distribution network does not match the actual demand of the medium-voltage distribution network, adjustments cannot be made in a timely manner, leading to some users not receiving timely power restoration or wasting resources. Therefore, this paper proposes an intelligent matching method for emergency power supply resources in distribution networks. Summary of the Invention

[0013] The purpose of this invention is to address the problems identified in the existing background technology. To achieve the above-mentioned objective, this invention provides the following technical solution: a method for intelligent matching of emergency power supply resources in a power distribution network, comprising the following steps: Step 1: Data Acquisition and Preprocessing: Real-time acquisition of distribution network topology information, fault location and type, emergency resource inventory status, load node power demand, and operational status data is conducted through multi-source heterogeneous data channels including sensors, IoT devices, and social media. The acquired datasets are then processed... Cleaning and pretreatment were performed, and outliers were identified and removed using the modified Raida criterion. Calculate the sample mean: Calculate the standard deviation: Set the anomaly detection interval as ,in This is the initial threshold coefficient; if a certain data point satisfy: If an outlier is identified, it is removed or corrected by interpolation. Repeat the above process until no new outliers appear, and output the cleaned and valid dataset for subsequent analysis. Step 2: Emergency Event Risk Assessment An enhanced stochastic ensemble classification model is constructed to quantify the risk level of sudden events in the power distribution network; Suppose there are B random decision trees, and the prediction result of the b-th tree is f_b(x). The overall prediction function is: in The initial phase is defined by the weights of each tree. Further introduction The algorithm performs adaptive weighted optimization; let the training sample set be... ,Label Indicate low / high risk categories; initialize weights for each sample: In the m-th iteration, train the weak classifier. (i.e., a single decision tree), calculate its classification error rate: like If the training terminates, then the classifier weights are calculated. Update sample weight distribution: Where the normalization factor The final output of the strong classifier is: Based on the probability output mechanism, the risk score is defined as: Based on R(x), the risk is divided into three levels: High risk: Medium risk: Low risk: ; Step 3: Emergency Resource Forecasting Based on historical emergency event databases and current event characteristics, a multilayer perceptron neural network (MLP) is used to predict the type and quantity of emergency resources required. Input vector Includes: event types Coding; Risk Score ; Scope of the fault Total power outage load ; Number of casualties (persons); Weather and environmental index; target output This indicates the demand for various emergency resources, including the number of generators, the number of repair team members, and the length of cables. A three-layer feedforward neural network is constructed, with 100, 80, and 60 neurons in the hidden layers, and the activation function is selected. : The output layer uses a linear activation function; The dataset is split into training, validation, and test sets in a 7:2:1 ratio; the minimum value is minimized using the backpropagation algorithm. Adjust the weight matrix and bias Training stops when the MSE converges on the validation set and falls below a preset threshold; the current event features are input into the trained MLP model to obtain the emergency resource demand prediction results; Step 4: Resource Matching Algorithm The Kuhn-Mankelsen algorithm, based on the complement method, is used to achieve optimal matching between emergency resources and faulty components; let the set of emergency resources be... The set of faulty components is Establish a weighted bipartite graph , border rights Representing resources For the fault The matching benefit is defined as: c in: Geographical distance from the resource to the fault location; Resource compatibility score ; After constructing the complete bipartite graph, execute the following algorithm: Initialize top index function ,satisfy Constructing a phase diagram ;exist Search for a perfect match; if one exists, end the process; otherwise, modify the vertex index, expand the augmenting path, and repeat the iteration until the maximum weight perfect match is found. Output the set of optimal matching pairs Maximize the overall matching benefit: Step 5: Transportation route planning: Use a genetic algorithm to find the optimal transportation route from the resource storage point to multiple fault sites, minimizing the total transportation time and cost; Encoding method: Integer permutation encoding is used to represent the order in which fault points are accessed; the fitness function is defined as: in: T: Total travel time; C: Fuel and labor costs; U: Route congestion penalties; Adjustment coefficient; Step 6: Multi-voltage level grid coordinated restoration Construct a collaborative optimization model between upper and lower layers to achieve joint restoration of high-voltage (HV) and medium-voltage (MV) distribution networks; Upper-level model (HV side): The substation serves as the hub, optimizing main grid switch operation and power flow distribution to maximize the total recoverable load. Where S represents the set of substations. Its maximum power supply capacity; All substations This is passed to the lower-level MV network as a boundary condition; Lower-level model (MV side): For each medium-voltage feeder group connected to HV substation i, a "recovery group" G_i is formed, and a local optimization model is established: in Let be the load shedding amount at node j. Its importance weight; after the post-disaster emergency repairs are completed, the actual restored load will be... Feedback is sent to the upper layer to update the HV recovery boundary, thereby achieving closed-loop collaborative control.

[0014] As a preferred technical solution of the present invention, during the data acquisition process, the sensor sampling frequency is once every 5 minutes to ensure that real-time data of the power distribution network can be obtained in a timely manner.

[0015] As a preferred technical solution of the present invention, the enhanced random forest algorithm is trained with a preset number of iterations of 500 and an error rate threshold of 0.05.

[0016] As a preferred technical solution of the present invention, the hidden layer of the multilayer perceptron (MLP) model is set to 3 layers, with 100, 80 and 60 neurons in each layer, respectively.

[0017] As a preferred technical solution of the present invention, in the resource matching algorithm, the number of virtual vertices added is determined based on the difference between the number of emergency resources and the number of faulty components, and 1.5 times the difference is used as the upper limit of the number of virtual vertices added.

[0018] As a preferred technical solution of the present invention, when the genetic algorithm plans the transportation path, the population size is set to 200 and the number of iterations is 300.

[0019] As a preferred technical solution of the present invention, it also includes a multi-voltage-level power grid collaborative restoration step: In the upper-level model, the high-voltage distribution network determines the maximum restoration amount of the load at each substation node by optimizing the line switch status and power flow distribution according to the current network structure, and transmits it to the medium-voltage distribution network as the upper limit boundary value of its restoration process; the lower-level model regards all medium-voltage distribution networks connected to each high-voltage distribution network substation node as a group, and establishes optimization models with the goal of minimizing load loss and with emergency resource scheduling, system operation and dynamic network reconfiguration as constraints. After the distributed post-disaster fault repair and load restoration work is carried out, the actual load value that meets the boundary conditions is transmitted to the high-voltage distribution network as the upper limit boundary value of its restoration process, thereby achieving the purpose of multi-voltage-level power grid collaborative scheduling.

[0020] As a preferred technical solution of the present invention, when determining the maximum load recovery of each substation node in the high-voltage distribution network, the current carrying capacity of the line is considered and the current carrying capacity is limited to within 80% of the rated current carrying capacity.

[0021] As a preferred technical solution of the present invention, after emergency resource forecasting, emergency resources are divided into power generation equipment module, energy storage equipment module, emergency repair tool module, and protective material module. Based on the resource demand forecasting results, corresponding resources are quickly allocated from each module.

[0022] As a preferred technical solution of the present invention, in the early warning mechanism, the interval for sending early warning information is 10 minutes for high-risk emergency events, 30 minutes for medium-risk emergency events, and 60 minutes for low-risk emergency events.

[0023] Compared with existing technologies, the beneficial effects of this invention are as follows: This method utilizes multi-source data acquisition channels, including sensors, IoT devices, and social media, to comprehensively collect relevant data from the power distribution network. Sensors can acquire key parameters of the power distribution network operation in real time, IoT devices can accurately monitor the status of power supply equipment, and social media provides a convenient way to collect user feedback. This multi-source data acquisition method ensures the comprehensiveness and timeliness of the data, enabling the system to more accurately understand the actual operating status of the power distribution network and user needs, providing a solid data foundation for subsequent analysis and decision-making.

[0024] In the data preprocessing stage, a modified Laida criterion is used for data cleaning. Compared to the traditional Laida criterion, the modified Laida criterion adjusts the outlier judgment interval according to the actual data distribution, which can more accurately identify and handle outlier data. By replacing outliers with the mean, noisy data is effectively removed while preserving the overall characteristics of the data, improving the reliability and quality of the data, and providing more accurate data support for subsequent model training and analysis.

[0025] This paper employs the augmented random forest algorithm for risk assessment of power distribution network emergency events. This algorithm combines the advantages of random forest and the Adaboost algorithm. The multiple decision trees in a random forest can analyze data from different perspectives, while the Adaboost algorithm iteratively trains multiple weak classifiers and adjusts sample weights based on the classification error rate, allowing the model to focus more on misclassified samples, thus improving the model's accuracy and generalization ability. This algorithm can more accurately assess the risk level of emergency events, providing a reliable basis for formulating appropriate response strategies.

[0026] Different warning levels are set based on risk scores, and corresponding warning message sending intervals are set for each level. For high-risk events, a warning message is sent every 10 minutes to promptly remind relevant personnel to take emergency response measures and minimize losses; for medium-risk events, a warning message is sent every 30 minutes; and for low-risk events, a warning message is sent every 60 minutes. This ensures timely attention to different risk events while avoiding information overload caused by frequent warning messages, effectively improving the efficiency of emergency response.

[0027] Emergency resource prediction is achieved using a multilayer perceptron (MLP) model, which can handle complex nonlinear relationships. By analyzing data on event type, risk level, impact range, and casualties, and training and validating the model using historical emergency event data, the MLP model can accurately predict the type and quantity of emergency resources required. This enables proactive resource preparation and allocation plans to be made in advance when an emergency occurs, avoiding problems caused by insufficient or wasted resources and improving the efficiency of emergency resource utilization.

[0028] Dividing the dataset into training, validation, and test sets in a 7:2:1 ratio is a scientifically sound approach that facilitates model training and evaluation. The training set is used for parameter learning, the validation set for tuning hyperparameters and evaluating generalization ability, and the test set for final performance evaluation. This method ensures good model performance across different datasets, improving model stability and reliability.

[0029] In the resource matching process, a point-supplementation method is used to address the mismatch between the number of emergency resources and faulty components. By adding virtual vertices, the original weighted bipartite graph is transformed into a weighted complete bipartite graph, providing a more suitable processing foundation for subsequent matching algorithms. This method avoids matching anomalies caused by quantity mismatches, enabling the matching algorithm to more effectively find the optimal matching solution and improving the accuracy and feasibility of resource matching.

[0030] The Kuhn-Mankers algorithm based on the point-supplementation method can quickly and accurately find the maximum weighted matching pair between emergency resources and faulty components. By continuously expanding augmenting paths and modifying the top pole value, the algorithm gradually optimizes the matching scheme. Considering factors such as resource capacity, transportation costs, and repair time, it achieves efficient allocation of emergency resources, maximizing the satisfaction of emergency power supply needs.

[0031] A genetic algorithm is used for transportation route planning. This algorithm simulates the process of biological evolution, continuously optimizing transportation routes through selection, crossover, and mutation operations. By setting an appropriate population size (200 individuals) and number of iterations (300), the optimal solution can be found among many possible routes. The genetic algorithm considers multiple factors such as path length, transportation time, and transportation cost, ensuring that emergency resources can reach the fault location as quickly as possible and at the lowest cost, thus improving the timeliness of emergency response.

[0032] During transportation, the genetic algorithm can dynamically adjust based on real-time road conditions and other actual circumstances. If traffic congestion or other emergencies occur, the algorithm can replan the route to ensure that the transportation of emergency resources is not affected, further improving the flexibility and practicality of transportation route planning.

[0033] The multi-voltage-level grid collaborative restoration method enables coordinated dispatching of high-voltage and medium-voltage distribution networks. The high-voltage distribution network provides the upper limit boundary value for load restoration to the medium-voltage distribution network, which then restores the load according to its own situation and feeds back the actual load value, forming a closed-loop collaborative dispatching mechanism. This approach avoids conflicts and incoordination during the restoration process of different voltage-level grids, improves the emergency restoration efficiency of the entire distribution network, and reduces outage time and scope.

[0034] When calculating load restoration in high-voltage distribution networks, the line's current carrying capacity is considered and limited to within 80% of its rated current carrying capacity. This measure effectively avoids grid failures caused by overload, ensures the safe and stable operation of the grid during emergency restoration, reduces the risk of secondary faults, and improves grid reliability.

[0035] Dividing emergency resources into modules for power generation equipment, energy storage equipment, repair tools, and protective equipment allows for more efficient resource allocation. When an emergency occurs, the necessary resources can be quickly selected from the appropriate modules based on forecasts, reducing the time spent searching for and allocating resources and improving the speed of emergency response.

[0036] During emergency power supply operations, user power consumption and the operating status of power supply equipment are monitored in real time, and the feedback information is promptly updated in the matching model. The matching scheme is dynamically adjusted and optimized based on actual conditions to ensure it always adapts to real-world needs, thereby improving the effectiveness and reliability of emergency power supply and maximizing the power supply for critical users. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the method flow provided by the present invention; Figure 2 This is a schematic diagram of the method data parameters provided by the present invention; Figure 3 A schematic diagram of the parameter data for the method model provided by this invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention.

[0039] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely illustrates some embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention. It should be noted that, in the absence of conflict, the embodiments and features and technical solutions in the embodiments of the present invention can be combined with each other. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0040] Example 1: Step 1: Data Acquisition and Preprocessing: Real-time acquisition of distribution network topology information, fault location and type, emergency resource inventory status, load node power demand, and operational status data is achieved through multi-source heterogeneous data channels from sensors, IoT devices, and social media. The acquired dataset is then processed... Cleaning and pretreatment were performed, and outliers were identified and removed using the modified Raida criterion. Calculate the sample mean: Calculate the standard deviation: Set the anomaly detection interval as ,in This is the initial threshold coefficient; if a certain data point satisfy: If an outlier is identified, it is removed or corrected by interpolation. Repeat the above process until no new outliers appear, and output the cleaned and valid dataset for subsequent analysis. Step 2: Emergency Event Risk Assessment An enhanced stochastic ensemble classification model is constructed to quantify the risk level of sudden events in the power distribution network; Suppose there are B random decision trees, and the prediction result of the b-th tree is f_b(x). The overall prediction function is: in The initial phase is defined by the weights of each tree. Further introduction The algorithm performs adaptive weighted optimization; let the training sample set be... ,Label Indicate low / high risk categories; initialize weights for each sample: In the m-th iteration, train the weak classifier. (i.e., a single decision tree), calculate its classification error rate: like If the training terminates, then the classifier weights are calculated. Update sample weight distribution: Where the normalization factor The final output of the strong classifier is: Based on the probability output mechanism, the risk score is defined as: Based on R(x), the risk is divided into three levels: High risk: Medium risk: Low risk: ; Step 3: Emergency Resource Forecasting Based on historical emergency event databases and current event characteristics, a multilayer perceptron neural network (MLP) is used to predict the type and quantity of emergency resources required. Input vector Includes: event types Coding; Risk Score ; Scope of the fault Total power outage load ; Number of casualties (persons); Weather and environmental index; target output This indicates the demand for various emergency resources, including the number of generators, the number of repair team members, and the length of cables. A three-layer feedforward neural network is constructed, with 100, 80, and 60 neurons in the hidden layers, and the activation function is selected. : The output layer uses a linear activation function; The dataset is split into training, validation, and test sets in a 7:2:1 ratio; the minimum value is minimized using the backpropagation algorithm. Adjust the weight matrix and bias Training stops when the MSE converges on the validation set and falls below a preset threshold; the current event features are input into the trained MLP model to obtain the emergency resource demand prediction results; Step 4: Resource Matching Algorithm The Kuhn-Mankelsen algorithm, based on the complement method, is used to achieve optimal matching between emergency resources and faulty components; let the set of emergency resources be... The set of faulty components is Establish a weighted bipartite graph , border rights Representing resources For the fault The matching benefit is defined as: c in: Geographical distance from the resource to the fault location; Resource compatibility score ; After constructing the complete bipartite graph, execute the following algorithm: Initialize top index function ,satisfy Constructing a phase diagram ;exist Search for a perfect match; if one exists, end the process; otherwise, modify the vertex index, expand the augmenting path, and repeat the iteration until the maximum weight perfect match is found. Output the set of optimal matching pairs Maximize the overall matching benefit: Step 5: Transportation route planning: Use a genetic algorithm to find the optimal transportation route from the resource storage point to multiple fault sites, minimizing the total transportation time and cost; Encoding method: Integer permutation encoding is used to represent the order in which fault points are accessed; the fitness function is defined as: in: T: Total travel time; C: Fuel and labor costs; U: Route congestion penalties; Adjustment coefficient; Step 6: Multi-voltage level grid coordinated restoration Construct a collaborative optimization model between upper and lower layers to achieve joint restoration of high-voltage (HV) and medium-voltage (MV) distribution networks; Upper-level model (HV side): The substation serves as the hub, optimizing main grid switch operation and power flow distribution to maximize the total recoverable load. Where S represents the set of substations. Its maximum power supply capacity; All substations This is passed to the lower-level MV network as a boundary condition; Lower-level model (MV side): For each medium-voltage feeder group connected to HV substation i, a "recovery group" G_i is formed, and a local optimization model is established: in Let be the load shedding amount at node j. Its importance weight; after the post-disaster emergency repairs are completed, the actual restored load will be... Feedback is sent to the upper layer to update the HV recovery boundary, thereby achieving closed-loop collaborative control.

[0041] Example 2: A method for intelligent matching of emergency power supply resources in a distribution network. Step 1: Data acquisition and preprocessing: Collect data on the topology, faulty components, emergency resources, and load information of the distribution network using multi-source data acquisition channels such as sensors, IoT devices, and social media. Clean and preprocess the collected data using the modified Laida criterion. Specifically, the collected data set is {x1, x2, ... xn}, and the mean and standard deviation of the data are calculated. Determine the interval for anomaly judgment. For each data point xi in the data set, determine whether it is within the anomaly judgment interval. If it is not within the interval, it is considered an anomaly, and the mean is used to replace the anomaly. Step 2, Emergency Event Risk Assessment: The preprocessed data is analyzed using the Augmented Random Forest algorithm to assess the risk level of the emergency event. First, an Augmented Random Forest model is defined, where b is the number of trees, fb is the prediction result of the b-th tree, and AdaBoost is the adaptive boosting algorithm. hm(x) is the m-th weak classifier, and αm is the weight of this classifier. The Random Forest and AdaBoost are trained together, initializing the weights of all training samples to phase values. A decision tree is trained as a weak classifier, and the training of the weak classifier to normalize weights is repeated until a preset number of iterations or an error rate threshold is reached. The trained Augmented Random Forest model is used to assess the risk of new data. Combining the outputs of all weak classifiers, a threshold is set based on the risk score to classify the risk level: a risk score > 0.7 indicates high risk, 0.4 < risk score ≤ 0.7 indicates medium risk, and a risk score ≤ 0.4 indicates low risk. Based on the risk assessment results, a corresponding early warning mechanism is triggered, sending warning information via SMS and app push notifications. Step 3: Emergency Resource Prediction: Based on historical emergency event data, current event information, and risk assessment results, a multilayer perceptron (MLP) is used to predict the required emergency resource types and quantities. The model input type is determined, event types are encoded, risk levels are quantified, and the impact range and casualties are represented by actual numerical values. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio. The selected deep learning model is trained using the training set, and the model parameters are adjusted by minimizing the mean squared error. The trained model is evaluated using the validation set to check its accuracy and generalization ability. Once the model evaluation is passed, the feature data of the current event is input into the trained model, and the model outputs the predicted emergency resource types and quantities. Step 4, Resource Matching Algorithm: Based on the Kuhn-Mankers algorithm using the vertices complement method, a fast and optimal match is performed between emergency resources and faulty components. First, by adding virtual vertices, the weighted bipartite graph between emergency resources and faulty components is transformed into a weighted complete bipartite graph, solving the problem of mismatch. Based on the initial matching path, the maximum weighted matching pair between emergency resources and faulty components is obtained through alternating iterations of expanding augmenting paths and modifying the top pole values. Step 5: Transportation route planning: Use genetic algorithms to plan the optimal resource transportation route to ensure that materials arrive at the site quickly and accurately.

[0042] During the data acquisition process, the sensor samples every 5 minutes to ensure timely acquisition of real-time data from the power distribution network.

[0043] When training the augmented random forest algorithm, the preset number of iterations is 500 and the error rate threshold is 0.05.

[0044] The hidden layers of the multilayer perceptron (MLP) model are set to 3 layers, with 100, 80, and 60 neurons in each layer, respectively.

[0045] In the resource matching algorithm, the number of virtual vertices added is determined based on the difference between the number of emergency resources and the number of faulty components, with 1.5 times the difference serving as the upper limit for the number of virtual vertices added.

[0046] When using a genetic algorithm to plan transportation routes, the population size is set to 200 and the number of iterations is 300.

[0047] In the upper-level model, the high-voltage distribution network determines the maximum load recovery amount of each substation node by optimizing the line switch status and power flow distribution based on the current network structure, and transmits it to the medium-voltage distribution network as the upper limit boundary value of its recovery process. The lower-level model treats all medium-voltage distribution networks connected to each high-voltage distribution network substation node as a group, and establishes optimization models with the goal of minimizing load loss and with emergency resource scheduling, system operation and dynamic network reconfiguration as constraints. After the distributed post-disaster fault repair and load restoration work is carried out, the actual load value that meets the boundary conditions is transmitted to the high-voltage distribution network as the upper limit boundary value of its recovery process, thereby achieving the purpose of multi-voltage level grid coordinated scheduling.

[0048] When determining the maximum load restoration capacity of each substation node in a high-voltage distribution network, the current carrying capacity of the line is taken into account, and the current carrying capacity is limited to within 80% of the rated current carrying capacity.

[0049] After forecasting emergency resources, emergency resources are divided into modules for power generation equipment, energy storage equipment, emergency repair tools, and protective materials. Based on the forecast results of resource demand, the corresponding resources are quickly allocated from each module.

[0050] In the early warning mechanism, the interval between sending early warning information is 10 minutes for high-risk emergency events, 30 minutes for medium-risk emergency events, and 60 minutes for low-risk emergency events.

[0051] Experimental example: Experimental Objective The effectiveness, accuracy, and efficiency of the intelligent matching method for emergency power supply resources in distribution networks are verified in actual distribution network emergency scenarios, and its effect on reducing load loss and shortening emergency response time is evaluated.

[0052] Test environment Simulated distribution network system: Construct a simulation system similar to the actual distribution network structure, covering lines, substations, and load nodes at different voltage levels (such as high voltage and medium voltage), and simulating various fault scenarios.

[0053] Data acquisition equipment: Deploy various sensors and IoT devices to collect real-time operational data of the power distribution network, such as voltage, current, and power. Simultaneously, set up social media data acquisition interfaces to simulate and obtain relevant public opinion information.

[0054] Computing resources: Equipped with high-performance servers for running intelligent matching algorithms and processing large amounts of data.

[0055] Test Procedure Step 1: Data Acquisition and Preprocessing Test Turn on the data acquisition equipment and simulate the data acquisition process during normal operation and malfunction. Record the acquisition of data from multiple sources, including sensors, IoT devices, and social media, and check the integrity and accuracy of the data.

[0056] The collected data is preprocessed, including data cleaning and normalization. The quality of the preprocessed data is observed, and the effectiveness of the preprocessing algorithm is evaluated.

[0057] Step 2: Emergency Resources and Faulty Component Matching Test Multiple faulty components are set up in the simulated power distribution network, and a certain number of emergency resources (such as generator trucks and repair teams) are deployed.

[0058] Run a fast matching algorithm for emergency resources and faulty components based on weighted bipartite graphs. Record the algorithm's solution time and matching results, and analyze the accuracy and rationality of the matching.

[0059] Change the quantity and type of faulty components and emergency resources, repeat the above test, and observe the performance of the algorithm in different scenarios.

[0060] Step 3: Multi-voltage level power grid coordinated recovery test Simulate fault scenarios in multi-voltage power grids (high voltage, medium voltage), and set different fault locations and severity levels.

[0061] A multi-voltage-level power grid collaborative restoration method based on a two-layer optimization model is implemented. In the upper-layer model, the process of determining the maximum load restoration amount at each substation node in the high-voltage distribution network is observed; in the lower-layer model, the optimization process of the medium-voltage distribution network with the goal of minimizing load loss is recorded.

[0062] Monitor information exchange and coordinated dispatch between power grids of different voltage levels, and evaluate the effectiveness of coordinated recovery methods in reducing load loss and improving recovery efficiency.

[0063] Step 4: Comprehensive Testing Simulate complex emergency scenarios in power distribution networks, including multiple faults occurring simultaneously and the allocation of different types of emergency resources.

[0064] The entire distribution network emergency power supply resource intelligent matching system is activated to record the system's response time, load recovery status, and resource utilization efficiency indicators.

[0065] Compared with traditional emergency response methods, this study analyzes the advantages of intelligent matching methods in reducing load loss and shortening recovery time.

[0066] Test indicators and evaluation

[0067] This table shows the changes in the number of outliers, the outlier removal rate, and the improvement in data integrity before and after preprocessing for different types of data, demonstrating the effectiveness of data preprocessing in improving data quality.

[0068] This table shows the changes in the number of outliers, the outlier handling rate, and the improvement in data integrity before and after preprocessing for different types of data, demonstrating the effect of data preprocessing steps on improving data quality.

[0069] This table compares the performance of traditional methods and intelligent matching methods in terms of emergency response time, load loss rate, and resource allocation efficiency, comprehensively demonstrating the advantages of intelligent matching methods. Expected test results The data acquisition and preprocessing process can accurately and completely obtain the operating data of the distribution network, and the quality of the preprocessed data meets the requirements of subsequent algorithms.

[0070] The matching algorithm for emergency resources and faulty components can obtain accurate matching results in a short time with a high matching accuracy.

[0071] The multi-voltage-level grid collaborative restoration method can achieve effective collaborative scheduling between grids of different voltage levels, significantly reduce load loss, and improve restoration efficiency.

[0072] Comprehensive test results show that the intelligent matching system for emergency power supply resources in distribution networks can respond quickly in complex scenarios, effectively reduce load losses, and improve the emergency response capabilities of distribution networks. It has significant advantages compared to traditional methods.

[0073] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the specific embodiments described above. Therefore, any modifications or substitutions to the present invention, and all technical solutions and improvements that do not depart from the spirit and scope of the invention, are covered within the scope of the claims of the present invention.

Claims

1. An intelligent matching method for power supply resources in an emergency power distribution network, characterized in that, The method comprises the following steps: Step 1: Data collection and preprocessing: Through the multi-source heterogeneous data channels of sensors, Internet of Things devices, and social media, real-time collection of power distribution network topology information, fault component location and type, emergency resource inventory status, load node power demand, and operating state data; the collected data set is cleaned and preprocessed, and the modified Lelida criterion is used to identify and eliminate outliers: Calculate the sample mean: Calculate the standard deviation: The abnormality determination interval is set as wherein is an initial threshold coefficient; if a certain data point satisfies: If it is determined as an abnormal value, it is removed or interpolated and repaired; repeat the above process until no new abnormal value appears, and output the cleaned effective data set for subsequent analysis; Step 2: Emergency event risk assessment An enhanced random forest classification model is constructed to quantify the risk level of the power distribution network emergency; Let the random set contain B decision trees, and the prediction result of the bth tree is f_b(x), and the overall prediction function is: wherein are the weights of the trees, initial phase; Further introduced The algorithm performs adaptive weighted optimization; let the training sample set be , and the label represents the low / high risk category; initialize the weight of each sample: In the mth iteration, the weak classifier is trained (i.e., a single decision tree), and its classification error rate is computed: If , terminate training; otherwise compute the classifier weights: Update the sample weight distribution: where the normalization factor The final strong classifier output is: Combined with the probability output mechanism, the risk score is defined as: According to R(x), the risk is divided into three levels: High risk: ; medium risk: ; low risk: ; Step 3: Emergency resource prediction Based on the historical emergency event database and the current event characteristics, a multi-layer perceptron neural network (MLP) is used to predict the type and quantity of emergency resources required; Input vector including: event type encoding; risk score ; failure impact range ; total outage load ; number of casualties (people) Weather environment index; target output Indicate the demand of various emergency resources, including the number of generators, the number of repair teams, the length of cables; A three-layer feedforward neural network was constructed, with 100, 80, and 60 neurons in the hidden layers, respectively, and the activation function selected as : The output layer uses a linear activation function; The dataset is split into training, validation, and test sets in a 7:2:1 ratio; the minimum value is minimized using the backpropagation algorithm. Adjust the weight matrix and bias Training stops when the MSE converges on the validation set and falls below a preset threshold; the current event features are input into the trained MLP model to obtain the emergency resource demand prediction results; Step 4: Resource matching algorithm The Kuhn-Munkres algorithm based on the complementary point method is used to realize the optimal matching between the emergency resources and the fault elements. The set of emergency resources is denoted as , the set of fault elements is denoted as , a weighted bipartite graph is established , and the edge weight represents the matching benefit of the resource to the fault , which is defined as: c Where: : a geographic distance of the resource to the point of failure; : a resource compatibility score ; After constructing the complete bipartite graph, the following algorithm is executed: initializing the top label function , satisfying constructing the subgraph of edges ; finding a perfect matching on ; if one exists, then end; otherwise, modify the top label, extend the augmenting path, and repeat the iteration until a maximum weight perfect matching is found; output optimal matching pair set maximizing total matching benefit: Step 5: Transportation path planning: genetic algorithm is used to solve the optimal transportation path from the resource storage point to multiple fault sites, minimizing the total transportation time and cost; Coding method: integer permutation coding is used to represent the order of visiting fault points; the fitness function is defined as: Where: T: total travel time; C: fuel and manpower cost; U: path congestion penalty term; Adjustment coefficient; Step 6: Multi-voltage level power grid collaborative recovery An upper and lower layer collaborative optimization model is constructed to realize joint recovery of high-voltage (HV) and medium-voltage (MV) distribution networks; Upper model (HV side): The substation is the hub, and the switch operation and power flow distribution of the main network are optimized to maximize the total recoverable load: where S is a set of substations, is its maximum power supply capacity; The voltage of each substation is are passed to the lower MV network as boundary conditions; Lower model (MV side): For each medium-voltage feeder group connected to HV substation i, form a "recovery group" G_i, and establish a local optimization model: wherein is the loss load of node j, is its importance weight; after the post-disaster repair is completed, the actual recovery load is fed back to the upper layer, the HV recovery boundary is updated, and closed-loop collaborative control is realized.

2. The power distribution network emergency power supply resource intelligent matching method according to claim 1, characterized in that, During the data collection process, the sampling frequency of the sensor is once every 5 minutes to ensure that real-time data of the power distribution network can be obtained in a timely manner.

3. The power distribution network emergency power supply resource intelligent matching method according to claim 2, characterized in that, When training the enhanced random forest algorithm, the preset number of iterations is 500 times, and the error rate threshold is 0.

05.

4. The power distribution network emergency power supply resource intelligent matching method according to claim 3, characterized in that, The hidden layer of the multi-layer perceptron MLP model is set to 3 layers, and the number of neurons in each layer is 100, 80, and 60, respectively.

5. The power distribution network emergency power supply resource intelligent matching method according to claim 4, characterized in that, In the resource matching algorithm, the number of virtual vertices is determined according to the difference between the number of emergency resources and fault elements, and 1.5 times the difference is used as the upper limit of the number of virtual vertices.

6. The power distribution network emergency power supply resource intelligent matching method according to claim 5, characterized in that, When the genetic algorithm is used to plan the transportation path, the population size is set to 200, and the number of iterations is 300.

7. The power distribution network emergency power supply resource intelligent matching method according to claim 6, characterized in that, It also includes the multi-voltage level power grid collaborative recovery step: In the upper model, the high-voltage distribution network determines the maximum recovery amount of each substation node load by optimizing the line switch state and power flow distribution according to the current network structure, and transmits it to the medium-voltage distribution network as the upper limit boundary value of its recovery process; The lower model regards all medium-voltage distribution networks connected to each high-voltage distribution network substation node as a group, and establishes an optimization model with the minimum loss of load as the objective, and emergency resource scheduling, system operation, and network dynamic reconstruction as the constraint conditions. After the distributed post-disaster fault repair and load recovery work is completed, the actual value of the load that meets the boundary conditions is transmitted to the high-voltage distribution network as the upper limit boundary value of its recovery process.

8. The power distribution network emergency power supply resource intelligent matching method according to claim 7, characterized in that, The high-voltage distribution network determines the maximum recovery amount of each substation node load, and the carrying capacity is limited within 80% of the rated carrying capacity.

9. The power distribution network emergency power supply resource intelligent matching method according to claim 8, characterized in that, After the emergency resource prediction, the emergency resources are divided into power generation equipment modules, energy storage equipment modules, repair tool modules and protective material modules, and the corresponding resources are quickly allocated from each module according to the resource demand prediction results.

10. The power distribution network emergency power supply resource intelligent matching method according to claim 9, characterized in that, In the early warning mechanism, for high-risk emergency events, the interval time of early warning information sending is 10 minutes, for medium-risk emergency events, the interval time of early warning information sending is 30 minutes, and for low-risk emergency events, the interval time of early warning information sending is 60 minutes.

Images