A power facility fault prediction and emergency repair scheduling method and system

Abstract

The present application relates to a kind of electric power facility failure prediction and emergency repair scheduling method, comprising the following steps: collecting various data in dispatching area, obtain multi-source time series data, with the multi-source time series data of each type of electric power facility as characteristic variable, with the failure probability of corresponding electric power facility as target variable, train random forest model, predict the failure probability of each electric power facility at current time, and based on the risk level of the preset failure probability interval division, carry out dispatching optimization to high-risk facility. The matching framework of repair team capability and fault point demand is constructed, and the repair team capability and fault point demand are quantitatively matched under the same dimension. A repair team scheduling optimization model is constructed, a variety of driving strategies are embedded in the scheduling optimization model, iterative search is carried out using NSGA-II algorithm, a Pareto optimal solution set is obtained and partial order comparison and sorting are carried out, and a single optimal repair scheduling decision scheme is selected using fuzzy comprehensive evaluation method.

Description

Technical Field

[0001] This application relates to the field of power facility maintenance, specifically to a method and system for power facility fault prediction and emergency repair dispatch. Background Technology

[0002] Frequent rainstorms and flash floods pose a serious threat to power infrastructure, easily causing power outages and disrupting normal social operations. Therefore, improving the emergency response capabilities of the power system under disasters is of paramount importance.

[0003] Existing technologies mainly fall into two categories: one utilizes meteorological, geographical, and historical data to predict the probability of power facility failures and identify high-risk areas through statistical analysis or machine learning methods; the other builds optimization models after a disaster to dispatch repair resources to minimize recovery time or cost. However, existing technologies have significant shortcomings: First, failure prediction and repair dispatch are disconnected, and the prediction results fail to effectively guide dispatch decisions, resulting in a lack of foresight in resource allocation. Second, dispatch models often treat repair teams as homogeneous resources, ignoring their capability differences and the specific needs of different failures, easily leading to resource misallocation. Finally, multi-objective optimization often relies on subjective weighting simplification, making it difficult to balance cost, time, and social benefits, resulting in low decision-making transparency.

[0004] In summary, existing technologies are insufficient to meet the needs of complex disaster scenarios. There is an urgent need for an integrated solution that combines fault prediction and intelligent scheduling, considers capacity matching, and can scientifically handle multi-objective optimization, in order to improve the efficiency and intelligence level of power emergency repair. Summary of the Invention

[0005] The present invention aims to provide a method for predicting power facility failures and dispatching emergency repairs to address the shortcomings of existing technologies. The technical problem to be solved by the present invention is achieved through the following technical solution.

[0006] A method for predicting power facility failures and dispatching emergency repairs includes the following steps: Collect distribution network data, historical data of distribution network faults, meteorological data, geographical data and population distribution data within the dispatch area, and preprocess the above data to obtain multi-source time series data; Using multi-source time-series data of each type of power facility as feature variables and the failure probability of the corresponding power facility as the target variable, a random forest model is trained to predict the failure probability of each power facility at the current moment. Based on the preset failure probability interval, risk levels are divided, and scheduling optimization is carried out for high-risk facilities. Based on group preference analysis and recommendation mechanism, a matching framework for the capabilities of emergency repair teams and the needs of fault points is constructed. The capabilities of emergency repair teams and the needs of fault points are placed under the same dimension for quantitative matching. The degree of matching is quantified by Euclidean distance to obtain the allocation utility value. An optimization model for emergency repair team scheduling is constructed with the dual objectives of minimizing the total distance of emergency repair routes and maximizing the overall allocation utility considering time-delay. Constraints are set on route continuity, working time limit, and road accessibility. Three elastic scheduling strategies—fault level-driven strategy, proximity response-driven strategy, and comprehensive performance-driven strategy—are embedded into the scheduling optimization model. The NSGA-II algorithm is used for iterative search to obtain the Pareto optimal solution set. Based on preset group preferences, the solutions in the Pareto optimal solution set are compared and sorted in a partial order, and a fuzzy comprehensive evaluation method is used to select a single optimal emergency repair scheduling decision scheme.

[0007] Preferably, the step of using multi-source time-series data of each type of power facility as feature variables and the failure probability of the corresponding power facility as the target variable to train a random forest model to predict the failure probability of each power facility at the current moment includes: for multi-source time-series data of each type of power facility, randomly and with replacement, multiple multi-source time-series data are extracted to form multiple sample sets, each sample set is used to train a decision tree, for each decision tree, a portion of features in the feature vectors are randomly selected, the Gini index of each feature vector is calculated at each node, the feature vector with the smallest Gini index is selected as the splitting feature, and the decision tree node is split as the basis for splitting, each decision tree node is split until the decision tree stops growing; the target variable values ​​corresponding to all feature vectors that have reached the leaf nodes of the decision tree are used as the predicted failure probability values ​​of individual power facilities, wherein the decision tree stops growing includes the number of decision tree nodes being less than a specified value, the Gini index being less than a preset threshold, the decision tree depth reaching a specified value, or all multi-source time-series data samples being used up.

[0008] Preferably, minimizing the total repair path distance and maximizing the overall allocation utility considering time-related degradation are achieved through the following objective function.

[0009]

[0010] in, RRC The total distance of the repair route. i and j The fault point k For the repair team, m To determine the number of vehicles requiring emergency repairs. n The number of fault points. S For the set of fault points, T To facilitate the assembly of readily available emergency repair teams, l ijk As a path decision variable, if the repair team k From the point of failure i Drive to the fault locationj If the value is 1, then the value is 0; otherwise, the value is 0. x ij Fault point i To the fault point j The actual distance; DU For the overall coordination effect, z ik Assign decision variables to the task, if the fault point i By the emergency repair team k If responsible, the value is 1; otherwise, it is 0. U ik For matching utility, e -λTi This is the time-related decay coefficient. l This is a decay parameter used to reflect the sensitivity to the timeliness of emergency repairs under disaster scenarios. T i For the repair team to reach the fault point i The time.

[0011] Preferably, decisions are also made by comparing the number of fault points with the number of repair vehicles. Specifically,

[0012] in, i The fault point k For the repair team, m To determine the number of vehicles requiring emergency repairs. S For the set of fault points, T To facilitate the assembly of readily available emergency repair teams, z ik Assign decision variables to tasks. When the ratio of the number of dispatchable repair teams to a fault point is less than 1, each fault point is handled by only one repair team; when the number of dispatchable repair teams to a fault point is greater than 1, each fault point can be handled by multiple repair teams.

[0013] Preferably, the path continuity constraints are as follows:

[0014] The upper limit constraints on working hours are as follows:

[0015] in, Q k For the repair team k Maximum working hours limit H The man-hours required to repair the fault point; The road accessibility constraints are as follows:

[0016] in, d ijLet be the decision variable, representing the fault point. i To the fault point j Whether the road section is damaged. w k Indicates the emergency repair team k Whether or not to equip with inflatable boats depends on the condition of the road section. d ij =0, any repair team can pass through this section of the road; when the road section is damaged, that is... d ij When the value is 1, only emergency repair teams equipped with inflatable boats can use this section of the road.

[0017] Preferably, the iterative search using the NSGA-II algorithm to obtain the Pareto optimal solution set includes: The allocation relationship between each emergency repair task and at least one emergency repair team is represented as a chromosome sequence using an encoding rule. This chromosome contains information on the task execution order and resource usage. An initial population of size N is randomly generated based on the demand quantity and the number of emergency repair teams. Feasibility is verified for each individual in the initial population. Individuals that violate the constraints are corrected or regenerated. The multi-objective fitness of each individual in the population is calculated, including the total path distance and allocation utility corresponding to each chromosome. A fitness vector is formed as the input for non-dominated sorting. The number of individuals dominated by each individual and the set of target individuals dominated by each individual are calculated. The population is divided into a set of Pareto front solutions of K layers according to the dominance relationship. Individuals in the same layer are sorted in ascending order according to each objective value. The objective difference between adjacent individuals is calculated for each objective dimension and normalized. The normalized differences are accumulated as the crowding distance of individuals to measure the sparsity of the solution distribution. Parents are selected based on non-dominated levels and crowding distance. A crossover operator is used to cross over parent individuals to generate some offspring. A mutation operator is used to perturb the chromosomes of some offspring. The parent and offspring populations are merged and a population update is performed to form an intermediate population of size 2N. Fast non-dominated sorting is performed on this intermediate population, and the top several Pareto solutions are selected sequentially according to the non-dominated level to fill the new population. The iteration terminates when the preset maximum number of iterations is reached or the changes in the non-dominated frontier are all below a threshold for several consecutive generations. The final Pareto optimal solution set is output as the set of emergency repair scheduling schemes.

[0018] This application also provides a power facility fault prediction and emergency repair dispatch system, characterized in that it includes: The data acquisition and preprocessing module collects distribution network data, historical data of distribution network faults, meteorological data, geographical data and population distribution data within the dispatch area, and preprocesses the above data to obtain multi-source time series data. The fault probability prediction and risk classification module uses multi-source time series data of each type of power facility as feature variables and the fault probability of the corresponding power facility as the target variable to train a random forest model, predict the fault probability of each power facility at the current time, and classify the risk level based on the preset fault probability interval, and carry out scheduling optimization for high-risk facilities. The matching module, based on group preference analysis and recommendation mechanism, constructs a matching framework between the capabilities of the emergency repair team and the needs of the fault point. It places the capabilities of the emergency repair team and the needs of the fault point under the same dimension for quantitative matching, and uses Euclidean distance to quantify the degree of matching to obtain the allocation utility value. The model building module constructs an optimization model for emergency repair team scheduling with the dual objectives of minimizing the total distance of emergency repair paths and maximizing the overall allocation utility considering time-delay, and sets constraints on path continuity, working time limits, and road accessibility. The scheduling strategy and solution module embeds three elastic scheduling strategies—fault level driven strategy, nearest response driven strategy, and comprehensive performance driven strategy—into the scheduling optimization model, and uses the NSGA-II algorithm for iterative search to obtain the Pareto optimal solution set. The output module, based on preset group preferences, performs partial order comparison and sorting of the solutions in the Pareto optimal solution set, and uses a fuzzy comprehensive evaluation method to select a single optimal emergency repair scheduling decision scheme.

[0019] Compared with existing technologies, the power facility fault prediction and emergency repair dispatch method of the present invention has the following advantages: 1. By directly using fault probability prediction results based on multi-source data as the core input for emergency repair scheduling, and through risk level classification, scheduling resources can be precisely tilted towards high-risk areas. This changes the situation where prediction and scheduling are disconnected in existing technologies, realizes the transformation from passive response to proactive prevention, and improves the ability to perceive disasters in advance.

[0020] 2. By fully considering the differences in the capabilities of emergency repair teams and the specific needs of the fault locations, a scientific matching framework was constructed to ensure that personnel and resources are used to their fullest potential. This effectively avoids the inefficiency caused by resource mismatch in existing technologies, improves the overall effectiveness of limited emergency repair resources, and shortens the average repair time.

[0021] 3. The system adopts minimizing path cost and maximizing dispatch utility considering time-delay as dual objectives, and incorporates three elastic scheduling strategies: fault level, nearest response, and overall efficiency. The Pareto optimal solution set is obtained through the NSGA-II algorithm, and decision-making is made in conjunction with preference analysis, enabling dynamic balancing of multiple objectives under different disaster scenarios.

[0022] 4. Through synergy, it is possible to accurately predict power facility failures, intelligently allocate emergency repair resources, and globally optimize emergency repair routes, thereby minimizing power outage time, reducing economic losses and social impact caused by power interruptions, and comprehensively enhancing the resilience and risk resistance of the power system under extreme disaster conditions. Detailed Implementation

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the embodiments.

[0024] The power facility fault prediction and emergency repair dispatching method of the present invention includes the following steps: S1. Collect distribution network data, historical distribution network fault data, meteorological data, geographical data, and population distribution data within the dispatch area, specifically as follows: (1) Distribution network data, including: Facility data: including facility ledgers and facility technical parameters; Line data: number of lines, total length, transmission capacity, transmission distance, insulation status, and conductor condition; Geographic coordinates of the facility.

[0025] (2) Distribution network fault data, including: Fault types: Various types of faults that occur in the power distribution network, such as facility failures, line power outages, switch failures, overloads, etc. Time of failure: The specific moment when the failure occurred; Fault duration: The time from the occurrence of a fault to its repair; Fault location: The specific location where the fault occurred (transformer, switchgear, line, etc.); Fault analysis: The specific circumstances and manifestations of the fault; Fault repair: Dispatch status of fault repair teams; Failure rate: In the dispatching area, it represents the probability or frequency of a failure occurring per unit of time. It reflects the ratio of the number of failures occurring in the distribution network within a specific time period to the total operating time; the failure rate can be calculated based on the statistical number of failures and failure times.

[0026] (3) Meteorological data, including: Rainfall: Total rainfall, hourly rainfall, effective rainfall; Rainfall duration: The length of time from the onset of rainfall to its end; Rainfall type: The characteristics of rainfall.

[0027] (4) Geographic data, including; Topographic data: slope, aspect, and elevation information; Environmental data: soil type, vegetation cover, land use type.

[0028] (5) Census data, including; Total population: The population distribution within the dispatch area.

[0029] The above data undergoes preprocessing, including deduplication, data correction, continuity testing, outlier handling, missing value imputation, and normalization. Specifically, during missing value imputation and normalization, missing values ​​in the scheduling region's basic data are filled with zero values. Continuous features are processed using extreme value normalization, mapping each feature to the [0,1] interval. Discrete features are numerically represented using one-hot encoding, resulting in preprocessed multi-source time-series data.

[0030] S2. Using multi-source time-series data of each type of power facility as feature variables and the failure probability of the corresponding power facility as the target variable, train a random forest model to predict the failure probability of each power facility at the current time, and divide the risk level based on the preset failure probability interval, and carry out scheduling optimization for high-risk facilities.

[0031] For each type of power facility, multiple multi-source time-series data are randomly and with replacement to form multiple subsets. Each subset is used to train a decision tree. For each decision tree, a subset of features from the feature vectors is randomly selected. The Gini index of each feature vector is calculated at each node. The feature vector with the smallest Gini index is selected as the splitting feature and serves as the basis for splitting the decision tree node. Each decision tree splits nodes until the decision tree stops growing. The target variable values ​​corresponding to all feature vectors that reach the leaf nodes of the decision tree are used as the predicted failure probability values ​​for individual power facilities. The conditions for the decision tree to stop growing include the number of decision tree nodes being less than a specified value, the Gini index being less than a preset threshold, the decision tree depth reaching a specified value, or all multi-source time-series data samples being used up.

[0032] Specifically, the preprocessed multi-source time-series data is divided into training and testing sets according to a preset ratio, with the training set preferably accounting for 80% and the testing set for 20%. A random forest classification model is constructed, and initial parameters are set, including but not limited to the number of decision trees, maximum tree depth, minimum number of leaf node samples, and the random selection ratio of features. The number of decision trees is gradually adjusted, and the classification accuracy of the model on the validation set is calculated for each parameter combination. The parameter combination that optimizes the prediction performance is selected as the final model parameters. The random forest model is trained based on the training set, and predictions are made for the test set samples, outputting the probability of failure for each power facility. Based on the predicted failure probability, the power facilities are divided into multiple risk levels, preferably five levels. When the predicted failure probability is greater than or equal to 60%, the facility is determined to be of high risk or above. Facilities of high risk or above are selected as the target set for dispatch facilities, and their spatial location, facility type, and predicted risk level information are extracted as inputs for the subsequent emergency repair and dispatch model.

[0033] S3. Based on group preference analysis and recommendation mechanism, construct a matching framework between the capabilities of emergency repair teams and the needs of fault points. Place the capabilities of emergency repair teams and the needs of fault points under the same dimension for quantitative matching. Use Euclidean distance to quantify the degree of matching and obtain the allocation utility value.

[0034] The dispatch facility target information includes fault type, risk level, facility spatial coordinates, and estimated repair time; the team information includes professional maintenance capabilities, team size and maximum available working hours, transportation capacity, and special equipment configuration. After normalization, each piece of information forms a fault point demand feature vector and a repair team capability feature vector with unified dimensions. These two vectors are projected onto the same latent feature space. Within this space, Euclidean distance is used to calculate the similarity between the repair team and the fault point. The distance results are then normalized, and the normalized distance is converted into a matching utility value, constructing a matching utility matrix between the repair team and the fault point.

[0035] S4. Construct an emergency repair team scheduling optimization model with the dual objectives of minimizing the total distance of emergency repair paths and maximizing the overall allocation utility considering time-delay, and set constraints on path continuity, working time limit and road accessibility.

[0036] First, the target information of the dispatch facilities is collected and preprocessed to calculate the emergency repair priority value. The target information of the dispatch facilities includes the risk level, the degree of social impact, and the repairability. The degree of social impact includes the population affected by the facility and the important buildings affected by the facility, and is calculated in the following way:

[0037] Comprehensive evaluation value of the degree of social impactE i It is calculated by the linear weighted sum of two sub-indices. oh 1. oh 2 are the weighting coefficients, satisfying oh 1 +oh 2 = 1. I i This indicator measures the importance of buildings affected by the power outage. It is obtained by weighting and summing the importance of various buildings within the power supply range of the fault point, assigning differentiated weight coefficients to buildings with different functions. For example, lifeline facilities (such as hospitals and fire stations) have a weight of 5, critical public service facilities (such as banks and train stations) have a weight of 4, and important industrial and commercial facilities have a weight of 3. R i It refers to the population size covered by the facility, which is estimated based on the census data and power load distribution to cover the population size (in ten thousand people) under normal power supply.

[0038] Fault point i Comprehensive emergency repair priority value P i The risk level is determined by a weighted integration of the indicators from the three dimensions mentioned above. f i Its value is equivalent to the risk level classification result of the random forest algorithm, while the repairability is determined by the estimated repair time. H i The reciprocal of the value is used to represent the physical meaning that the shorter the repair time, the faster the point can return to normal operation, the better the repairability, and therefore it should be given a higher priority for emergency repair.

[0039]

[0040] The goal of minimizing the total repair path distance and maximizing the overall allocation utility considering time-related degradation is achieved through the following objective function.

[0041]

[0042] in, RRC The total distance of the repair route. l ijk As a path decision variable, if the repair team k From the point of failure i Drive to the fault location j If the value is 1, then the value is 0; otherwise, the value is 0. x ij Fault point i To the fault point j The actual distance; DU For the overall coordination effect, zik Assign decision variables to the task, if the fault point i By the emergency repair team k If responsible, the value is 1; otherwise, it is 0. U ik For matching utility, e -λTi This is the time-related decay coefficient. l This is a decay parameter used to reflect the sensitivity to the timeliness of emergency repairs under disaster scenarios. T i For the repair team to reach the fault point i The time.

[0043] They also made corresponding decisions by comparing the number of fault points with the number of repair vehicles. Specifically,

[0044] in, i The fault point k For the repair team, m To determine the number of vehicles requiring emergency repairs. S For the set of fault points, T To facilitate the assembly of readily available emergency repair teams, z ik Assign decision variables to tasks. When the ratio of the number of dispatchable repair teams to a fault point is less than 1, each fault point is handled by only one repair team; when the number of dispatchable repair teams to a fault point is greater than 1, each fault point can be handled by multiple repair teams.

[0045] The path continuity constraints are as follows:

[0046] The upper limit constraint on working hours is as follows:

[0047] in, Q k For the repair team k Maximum working hours limit H The man-hours required to repair the fault point; The road accessibility constraints are as follows:

[0048] in, d ij Let be the decision variable, representing the fault point. i To the fault point j Whether the road section is damaged. w k Indicates the emergency repair team k Whether or not to equip with inflatable boats depends on the condition of the road section. dij When the value is 0, any repair team can pass through this section of the road; when the road section is damaged, i.e., when... d ij When the value is 1, only emergency repair teams equipped with inflatable boats can use this section of the road.

[0049] S5. Embed the three elastic scheduling strategies—fault level-driven strategy, proximity response-driven strategy, and comprehensive performance-driven strategy—into the scheduling optimization model, and use the NSGA-II algorithm for iterative search to obtain the Pareto optimal solution set.

[0050] The allocation relationship between each emergency repair task and at least one emergency repair team is represented as a chromosome sequence using an encoding rule. This chromosome contains information on the task execution order and resource usage. An initial population of size N is randomly generated based on the demand quantity and the number of emergency repair teams. Feasibility is verified for each individual in the initial population. Individuals that violate the constraints are corrected or regenerated. The multi-objective fitness of each individual in the population is calculated, including the total path distance and allocation utility corresponding to each chromosome. A fitness vector is formed as the input for non-dominated sorting. The number of individuals dominated by each individual and the set of target individuals dominated by each individual are calculated. The population is divided into a set of Pareto front solutions of K layers according to the dominance relationship. Individuals in the same layer are sorted in ascending order according to each objective value. The objective difference between adjacent individuals is calculated for each objective dimension and normalized. The normalized differences are accumulated as the crowding distance of individuals to measure the sparsity of the solution distribution.

[0051] Parents are selected based on non-dominated levels and crowding distance. A crossover operator is used to perform crossover operations on parent individuals to generate some offspring. A mutation operator is used to perturb the chromosomes of some offspring. The parent and offspring populations are merged and a population update is performed to form an intermediate population of size 2N. Fast non-dominated sorting is performed on this intermediate population. The Pareto solutions of the first few layers are selected in sequence according to the non-dominated level to fill the new population.

[0052] When the preset maximum number of iterations is reached or the changes in the non-dominated frontier are all below the threshold for several consecutive generations, the iteration is terminated, and the final Pareto optimal solution set is output as the set of emergency repair scheduling schemes.

[0053] S6. Perform partial order comparison and sorting on the solutions in the Pareto optimal solution set, select the recommended scheduling scheme that satisfies the preference based on the sorting result, and select a single scheduling scheme from the Pareto optimal solution set using the fuzzy comprehensive evaluation method.

[0054] It should be noted that the above detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0055] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments described in this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0056] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that the embodiments of this application described herein can be implemented in orders other than those described herein.

[0057] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or apparatus.

[0058] Other implementation schemes may be used and other modifications may be made without departing from the spirit or scope of the subject matter presented herein.

[0059] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting power facility faults and dispatching emergency repairs, characterized in that, Includes the following steps: Data on distribution network distribution, historical data on distribution network faults, meteorological data, geographical data, and population distribution within the dispatch area are collected and preprocessed to obtain multi-source time-series data. Using multi-source time-series data of each type of power facility as feature variables and the failure probability of the corresponding power facility as the target variable, a random forest model is trained to predict the failure probability of each power facility at the current moment. Based on the preset failure probability interval, risk levels are divided, and scheduling optimization is carried out for high-risk facilities. Based on group preference analysis and recommendation mechanism, a matching framework for the capabilities of emergency repair teams and the needs of fault points is constructed. The capabilities of emergency repair teams and the needs of fault points are quantitatively matched under the same dimension. The degree of matching is quantified by Euclidean distance to obtain the allocation utility value. An optimization model for emergency repair team scheduling is constructed with the dual objectives of minimizing the total distance of emergency repair routes and maximizing the overall allocation utility considering time-delay. Constraints are set on route continuity, working time limit, and road accessibility. Three elastic scheduling strategies—fault level-driven strategy, proximity response-driven strategy, and comprehensive performance-driven strategy—are embedded into the scheduling optimization model. The NSGA-II algorithm is used for iterative search to obtain the Pareto optimal solution set. Based on preset group preferences, the solutions in the Pareto optimal solution set are compared and sorted in a partial order, and a fuzzy comprehensive evaluation method is used to select a single optimal emergency repair scheduling decision scheme.

2. The method for predicting power facility faults and dispatching emergency repairs according to claim 1, characterized in that, The method of training a random forest model using multi-source time-series data of each type of power facility as feature variables and the failure probability of the corresponding power facility as the target variable to predict the failure probability of each power facility at the current moment includes: For each type of power facility, multiple multi-source time series data are randomly and with replacement to form multiple sample sets. Each sample set is used to train a decision tree. For each decision tree, a subset of features from the feature vectors are randomly selected. The Gini index of each feature vector is calculated at each node. The feature vector with the smallest Gini index is selected as the splitting feature and serves as the basis for splitting the decision tree node. Each decision tree splits a node until the decision tree stops growing. The target variable values ​​corresponding to all feature vectors reaching the leaf nodes of the decision tree are used as the fault probability prediction values ​​of individual power facilities. The conditions for the decision tree to stop growing include the number of decision tree nodes being less than a specified value, the Gini index being less than a preset threshold, the decision tree depth reaching a specified value, or all multi-source time series data samples being used up.

3. The method for predicting power facility faults and dispatching emergency repairs according to claim 1, characterized in that, The goal of minimizing the total repair path distance and maximizing the overall allocation utility considering time-related degradation is achieved through the following objective function. in, RRC The total distance of the repair route. i and j The fault point k For the repair team, m To determine the number of vehicles requiring emergency repairs. n The number of fault points. S For the set of fault points, T To facilitate the assembly of readily available emergency repair teams, l ijk As a path decision variable, if the repair team k From the point of failure i Drive to the fault location j If the value is 1, then the value is 0; otherwise, the value is 0. x ij Fault point i To the fault point j The actual distance; DU For the overall coordination effect, z ik Assign decision variables to the task, if the fault point i By the emergency repair team k If responsible, the value is 1; otherwise, it is 0. U ik For matching utility, e -λTi This is the time-related decay coefficient. λ This is a decay parameter used to reflect the sensitivity to the timeliness of emergency repairs under disaster scenarios. T i For the repair team to reach the fault point i The time.

4. The method for predicting power facility faults and dispatching emergency repairs according to claim 1, characterized in that, They also made corresponding decisions by comparing the number of fault points with the number of repair vehicles. Specifically, in, i The fault point k For the repair team, m To determine the number of vehicles requiring emergency repairs. S For the set of fault points, T To facilitate the assembly of readily available emergency repair teams, z ik Assign decision variables to tasks. When the ratio of the number of dispatchable repair teams to a fault point is less than 1, each fault point is handled by only one repair team; when the number of dispatchable repair teams to a fault point is greater than 1, each fault point can be handled by multiple repair teams.

5. The method for predicting power facility faults and dispatching emergency repairs according to claim 3, characterized in that, The path continuity constraints are as follows: The upper limit constraints on working hours are as follows: in, Q k For the repair team k Maximum working hours limit H The man-hours required to repair the fault point; The road accessibility constraints are as follows: in, d ij Let be the decision variable, representing the fault point. i To the fault point j Whether the road section is damaged. w k Indicates the emergency repair team k Whether or not to equip with inflatable boats depends on the condition of the road section. d ij =0, any repair team can pass through this section of the road; when the road section is damaged, that is... d ij When the value is 1, only emergency repair teams equipped with inflatable boats can use this section of the road.

6. The method for predicting power facility faults and dispatching emergency repairs according to claim 1, characterized in that, The iterative search using the NSGA-II algorithm to obtain the Pareto optimal solution set includes: The allocation relationship between each emergency repair task and at least one emergency repair team is represented as a chromosome sequence using an encoding rule. The chromosome contains information on the task execution order and resource usage. An initial population of size N is randomly generated based on the demand quantity and the number of emergency repair teams. Feasibility is checked for each individual in the initial population. Individuals that violate the constraints are corrected or regenerated. The multi-objective fitness of each individual in the population is calculated, including the total path distance and allocation utility corresponding to each chromosome. The fitness vector is formed as the input for non-dominated sorting. The number of individuals dominated by each individual and the set of individuals dominated by them are calculated. The population is divided into a set of Pareto front solutions of K layers according to the dominance relationship. Individuals in the same layer are sorted in ascending order according to each objective value. The objective difference between adjacent individuals is calculated and normalized for each objective dimension. The normalized difference is accumulated as the crowding distance of the individual to measure the sparsity of the solution distribution. Parents are selected based on non-dominated level and crowding distance. Crossover operators are used to perform crossover operations on parent individuals to generate some offspring. Mutation operators are used to perturb the chromosomes of some offspring. The parent population and the offspring population are merged and population update is performed to form an intermediate population of size 2N. Fast non-dominated sorting is performed on the intermediate population. The Pareto solutions of the first few layers are selected in sequence according to the non-dominated level to fill the new population. When the preset maximum number of iterations is reached or the changes in the non-dominated frontier are all below the threshold for several consecutive generations, the iteration is terminated, and the final Pareto optimal solution set is output as the set of emergency repair scheduling schemes.

7. A power facility fault prediction and emergency repair dispatch system, characterized in that, include: The data acquisition and preprocessing module collects distribution network data, historical data of distribution network faults, meteorological data, geographical data and population distribution data within the dispatch area, and preprocesses the above data to obtain multi-source time series data. The fault probability prediction and risk classification module uses multi-source time series data of each type of power facility as feature variables and the fault probability of the corresponding power facility as the target variable to train a random forest model, predict the fault probability of each power facility at the current time, and classify the risk level based on the preset fault probability interval, and carry out scheduling optimization for high-risk facilities. The matching module, based on group preference analysis and recommendation mechanism, constructs a matching framework between the capabilities of the emergency repair team and the needs of the fault point. It places the capabilities of the emergency repair team and the needs of the fault point under the same dimension for quantitative matching, and uses Euclidean distance to quantify the degree of matching to obtain the allocation utility value. The model building module constructs an optimization model for emergency repair team scheduling with the dual objectives of minimizing the total distance of emergency repair paths and maximizing the overall allocation utility considering time-delay, and sets constraints on path continuity, working time limits, and road accessibility. The scheduling strategy and solution module embeds three elastic scheduling strategies—fault level driven strategy, nearest response driven strategy, and comprehensive performance driven strategy—into the scheduling optimization model, and uses the NSGA-II algorithm for iterative search to obtain the Pareto optimal solution set. The output module, based on preset group preferences, performs partial order comparison and sorting of the solutions in the Pareto optimal solution set, and uses a fuzzy comprehensive evaluation method to select a single optimal emergency repair scheduling decision scheme.

Images