Intelligent dispatching method and system for distribution network repair command based on multi-source data fusion

Abstract

This invention discloses a method and system for intelligent dispatching of distribution network emergency repair command based on multi-source data fusion, belonging to the field of distribution network operation and maintenance emergency repair technology. The method first collects multi-source business data such as SCADA, 95598, GIS, meteorology, personnel, and materials. After standardized preprocessing, it uses unique equipment identifiers as the core and combines spatiotemporal dimensions to complete data association and fusion, constructing a fault feature library. Then, it uses a rule engine combined with a random forest model to achieve accurate fault assessment, quantifying and ranking fault priorities according to indicators such as equipment importance and the number of affected users. Based on information such as personnel, materials, and road conditions, and with the goal of minimizing the total repair time, it uses an improved genetic algorithm to complete multi-constraint path optimization and generate the optimal dispatching scheme. Finally, it tracks all nodes of the repair process in real time, automatically warns of anomalies, and generates statistical reports, realizing the integration of data fusion, intelligent assessment, optimal dispatching, and closed-loop management, significantly improving the efficiency of distribution network fault handling and power supply reliability.

Description

Technical Field

[0001] This invention relates to the field of power distribution network operation and maintenance emergency repair technology, specifically to a method and system for intelligent dispatching of power distribution network emergency repair command based on multi-source data fusion. Background Technology

[0002] With the comprehensive advancement of the construction of the new power system, the distribution network has transformed from a traditional power transmission endpoint into a core hub receiving diverse loads such as distributed renewable energy, energy storage, and electric vehicle charging stations. Its operational safety and power supply reliability are directly related to national economic development and the guarantee of electricity supply for people's livelihoods. The National Energy Administration and power grid companies have imposed stringent requirements on the reliability of distribution network power supply. Industry statistics show that distribution network faults account for more than 80% of all power system faults, and regional concentrated faults are particularly prone to occur under extreme weather conditions, posing a significant challenge to distribution network emergency repair and command.

[0003] Currently, emergency repair command in China's power distribution networks is still primarily manual, with supporting information systems mostly built independently for each line of work. While this achieves digital control of individual links, it suffers from significant technical shortcomings in areas such as end-to-end collaboration, multi-source data fusion, and intelligent decision support. Existing business systems belong to different lines of work, such as dispatching, production, marketing, and materials, creating severe "data silos." So-called data fusion merely achieves simple aggregation of multi-source data, without establishing a spatiotemporal correlation fusion mechanism centered on fault events. Fault information is severely fragmented, requiring dispatchers to repeatedly verify data across systems, significantly lengthening the fault handling cycle.

[0004] In the fault diagnosis phase, existing methods heavily rely on human experience, manually investigating only scattered alarms and repair reports. This results in low fault location accuracy and large errors in fault type identification. Some automated diagnosis solutions based on a single data source have extremely low fault tolerance, with persistently high rates of misdiagnosis and missed diagnosis. Furthermore, fault prioritization often uses simple qualitative rules without establishing a quantitative, multi-dimensional evaluation system. This fails to consider core factors such as equipment importance, the scale of affected users, weather risks, and power restoration timelines, easily leading to an imbalance in response priorities.

[0005] In the emergency repair dispatching phase, the traditional manual dispatching model relies solely on experience to assign tasks based on proximity, failing to comprehensively consider multiple factors such as personnel skills, material matching, and real-time road conditions. This results in prolonged ineffective time consumption and low resource utilization. Existing route planning schemes mostly only achieve shortest distance planning for a single fault point, unable to achieve globally optimal dispatching under multiple constraints, making dispatching chaos highly likely in scenarios with multiple concurrent faults. Furthermore, existing technologies lack closed-loop management of the entire emergency repair process, cannot provide automatic early warnings for abnormal work nodes and safety risks, and have a single dimension for post-event review, making it difficult to support distribution network operation and maintenance optimization and unable to meet the core requirements of intelligent operation and maintenance of distribution networks under the new power system. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention aims to provide a method and system for intelligent dispatching of distribution network emergency repair command based on multi-source data fusion. With the core objective of optimizing the efficiency of the entire distribution network fault handling process, it constructs a standardized and unified fault view through spatiotemporal correlation fusion of multi-source data using unique equipment identifiers as the core anchor point. A dual-wheel approach combining rule engine initial screening and machine learning fine-tuning achieves accurate fault assessment and multi-dimensional quantitative priority ranking. Furthermore, with the goal of minimizing total repair time and using multiple constraint boundaries as rules, an intelligent optimization algorithm completes global matching and path optimization of repair resources. Ultimately, it achieves real-time tracking of nodes throughout the entire distribution network emergency repair process, hierarchical early warning of abnormal states, and closed-loop iterative optimization of handling data, fundamentally improving the efficiency of distribution network fault handling and power supply reliability. This system achieves unified fusion of distribution network fault data, intelligent and accurate fault assessment, optimal dispatching of repair resources, and real-time control of the repair process, thereby improving distribution network emergency repair efficiency and power supply reliability.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for intelligent dispatching of power distribution network emergency repair command based on multi-source data fusion includes the following steps:

[0009] S1: Collect raw data from multiple sources of distribution network services, perform standardized preprocessing on the collected raw data, and obtain a standardized dataset in a unified format;

[0010] S2: Using the unique device identifier as the matching condition, and combining the two dimensions of timestamp and geographic coordinates, perform spatiotemporal correlation fusion on the standardized dataset to obtain a fused dataset that corresponds one-to-one with a single fault event. Then, extract the core fault features from the fused dataset and construct a distribution network fault feature library.

[0011] S3: Based on the fused dataset and the distribution network fault feature library, the system completes intelligent fault assessment by combining a rule engine and a machine learning model, and outputs the fault type, accurate fault location, and fault impact range; then, based on four preset indicators, namely the importance of fault-related equipment, the scale of affected users, regional meteorological risk, and allowable power restoration time limit, the system completes the fault priority ranking and outputs a fault handling list with priority labels.

[0012] S4: Based on the fault handling list with priority labels, combined with the original data of multi-source distribution network services collected in S1, with the goal of minimizing the total repair time, set multiple constraints to complete the repair path optimization, generate the optimal repair scheduling scheme, and output the standardized repair scheduling task.

[0013] S5: Based on the standardized emergency repair scheduling task, the entire process nodes of task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation are tracked in real time. The abnormal status of node timeout, path deviation, and safety risk is automatically triggered with graded early warning. After the task is completed, a full-dimensional statistical report of emergency repair is generated.

[0014] Furthermore, the raw data of the multi-source distribution network business collected in step S1 includes equipment operation and fault alarm data of the distribution network SCADA system and distribution automation system, user request data of the 95598 user repair system, spatial coordinate data of the GIS geographic information system, environmental data of the meteorological monitoring system, location and status data of the emergency repair personnel terminal, and inventory data of the material management system; the standardized preprocessing includes format unification, duplicate data removal, and outlier cleaning.

[0015] Furthermore, the spatiotemporal correlation fusion described in step S2 specifically includes the following sub-steps:

[0016] 1) Using the unique device identifier as the matching condition, perform an initial screening of multi-source data in the standardized dataset, and retain only the data that matches the device identifier for subsequent matching;

[0017] 2) Extract the timestamp information of the data after initial screening, calculate the time difference between the data to be matched, and control the time difference within the preset time window to obtain the time dimension matching data;

[0018] 3) Extract the geographic coordinate information of the time dimension matching data, calculate the spatial distance between the data, and control the spatial distance within a preset range to obtain the spatial dimension matching data;

[0019] 4) The matching results of the three dimensions of device identification, time and space are weighted and calculated to obtain the comprehensive matching degree. Among them, the device identification dimension has the highest weight, the time dimension has the second highest weight, and the space dimension has the lowest weight. The sum of the weight coefficients of the three dimensions is 1.

[0020] 5) Compare the overall matching degree with the preset matching threshold. When the overall matching degree is greater than or equal to the preset matching threshold, it is determined that multiple sets of data belong to the same fault event, and the spatiotemporal correlation fusion of multi-source data is completed to generate a fusion dataset that corresponds one-to-one with the fault event.

[0021] Furthermore, the step S2, which involves extracting core fault features and constructing a distribution network fault feature library, specifically includes the following sub-steps:

[0022] 1) Extract six core fault features from the fused dataset: equipment operation exceeding limits amplitude, frequency of user repair reports in the same area, importance level of fault-related equipment, meteorological risk level of the fault area, number of users affected by the fault, and distribution network topology correlation.

[0023] 2) Normalize the six core fault features and convert them into standardized values ​​in the range of 0 to 1 to form a unified quantified fault feature vector;

[0024] 3) Associate and bind the fault feature vector with the corresponding fault type, fault location, and fault impact range to form a single fault feature sample;

[0025] 4) Perform secondary association and binding between a single fault feature sample and the corresponding historical emergency repair and handling plan, standard emergency repair duration, and final power restoration result;

[0026] 5) Store all the fault feature samples that have been associated and bound into a distributed database to build a distribution network fault feature library that can be called in real time and traced back in history.

[0027] Furthermore, the intelligent fault assessment in step S3 specifically includes the following sub-steps:

[0028] 1) Retrieve the fusion dataset corresponding to the fault event, input it into the preset rule engine, and perform preliminary screening of the data through the preset fault judgment rules to filter out the suspected fault set that meets the fault triggering conditions; the preset fault judgment rules include electrical quantity abnormality triggering rules, user repair report related triggering rules, meteorological risk related triggering rules and distribution network topology related triggering rules.

[0029] 2) Retrieve the standardized fault feature vectors corresponding to the suspected fault set and input them into the pre-trained and validated random forest machine learning classification model;

[0030] 3) The fault feature vector is classified and calculated using a random forest machine learning classification model, and the output is the fault type, precise fault location area and fault impact range corresponding to the fault event.

[0031] Furthermore, the fault priority sorting in step S3 specifically includes the following sub-steps:

[0032] 1) Set corresponding weight coefficients for the four indicators respectively, and the sum of all weight coefficients is 1, among which the importance weight of fault-related equipment is the highest;

[0033] 2) Standardize and quantify the four indicators, converting them into score values ​​in the range of 0 to 1;

[0034] 3) Multiply the scores of each indicator by their corresponding weight coefficients and sum them to obtain the total priority score for the fault event;

[0035] 4) Determine the emergency repair priority of each fault according to the total priority score of all fault events from high to low, and generate a fault handling list with priority tags.

[0036] Furthermore, the multiple constraints in step S4 include: a single fault point is only the responsibility of one emergency repair team, the material carrying capacity of the emergency repair team does not exceed its maximum carrying capacity limit, and the time for the emergency repair team to reach the fault point does not exceed the maximum allowable power restoration time limit. The emergency repair path optimization is implemented using an improved genetic algorithm, with the fitness function being the minimization of the sum of the total emergency repair travel time and the task waiting time, and the multiple constraints being the selection rules, to iteratively solve for the optimal emergency repair scheduling scheme.

[0037] Furthermore, the standardized emergency repair scheduling task output in step S4 is simultaneously pushed to the mobile terminal of the corresponding emergency repair personnel. The pushed content includes the fault type, fault location coordinates, and fault impact range.

[0038] Furthermore, step S5 specifically includes the following sub-steps:

[0039] S51: Based on the standardized emergency repair scheduling task, six control nodes are defined: task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation. The completion status and corresponding timestamp of each node are collected in real time through the mobile terminal of the emergency repair personnel and the power distribution automation system to realize real-time status tracking of the entire emergency repair process.

[0040] S52: For three types of abnormal states, namely node timeout, path deviation and security risk, pre-set quantitative graded warning triggering rules, divide the warning into three levels according to the severity of the abnormality, match the corresponding push objects, response time limits and closed-loop handling processes for different warning levels, and automatically trigger the corresponding graded warning when the abnormal state meets the warning triggering rules.

[0041] S53: Once all emergency repair tasks are completed, an automatic full-dimensional statistical report on emergency repair will be generated, including statistics on the entire fault process duration, fault type and cause analysis, resource consumption statistics, personnel performance statistics, abnormal event statistics, and power restoration completion statistics.

[0042] A distribution network emergency repair command and dispatch system based on multi-source data fusion, used to implement the method, is characterized by comprising a data acquisition unit, a data fusion module, a fault analysis module, a resource scheduling module, and a visualization and control module connected in sequence via communication. The data acquisition unit is used to collect raw data from multiple sources in the distribution network, perform standardized preprocessing on the raw data, and output a standardized dataset in a unified format to the data fusion module. The data fusion module is used to perform spatiotemporal correlation fusion of multi-source data using unique device identifiers as the core matching condition, combined with timestamps and geographic coordinates, to generate a fused dataset corresponding to a single fault event and extract core fault features. A distribution network fault feature library is constructed and output to the fault analysis module. The fault analysis module is used to complete intelligent fault analysis by combining a rule engine and a random forest model. The fault priority is ranked by weighted quantification of four core indicators, and a fault handling list with priority labels is output to the resource scheduling module. The resource scheduling module is used to optimize the emergency repair path based on multiple constraints with the goal of minimizing the total repair time. The optimal scheduling scheme is generated and standardized emergency repair scheduling tasks are output to the visualization and management module. The visualization and management module is used to realize real-time tracking of nodes in the entire emergency repair process, hierarchical early warning of abnormal status, and generate a full-dimensional statistical report of emergency repair after the task is completed.

[0043] This invention aims to optimize efficiency and achieve a closed-loop management system for the entire process of power distribution network fault handling. It constructs an endogenous collaborative operation system encompassing "data spatiotemporal fusion, dual-wheel intelligent analysis, global optimization scheduling, and full-process closed-loop management," forming a complete technical logic loop from underlying data governance to top-level decision execution. Firstly, at the data layer, using the unique identifier of each device as the core spatiotemporal anchor, a weighted association mechanism based on "device identifier priority, time dimension matching, and spatial dimension verification" accurately aligns heterogeneous and fragmented data originally scattered across multiple business systems to the same fault event. This completely breaks down data silos, providing a complete, unified, and standardized multi-dimensional view of faults for fault analysis, fundamentally solving the core pain points of information asymmetry and high verification costs inherent in traditional models. Secondly, at the analysis layer, a dual-drive mechanism of "rule engine initial screening + machine learning fine judgment" is adopted. First, invalid false alarm data is filtered out and the scope of suspected faults is locked through solidified fault rules. Then, based on the historical prior knowledge of the distribution network fault feature library, a random forest classification model is used to classify and calculate standardized fault feature vectors. This deeply combines the rule-based determinism of manual handling experience with the generalization and adaptability of machine learning, greatly improving the accuracy and efficiency of fault analysis. At the same time, a priority ranking model with multi-dimensional index weighted quantification is used to transform qualitative emergency repair experience into a quantifiable and reproducible priority scoring system, achieving precise priority matching of fault handling resources. At the scheduling layer, based on a multi-constraint global optimization mechanism, with the core objective of minimizing the total repair time, boundary conditions such as personnel skills and qualifications, real-time location, material carrying capacity, power restoration time limit, and real-time road conditions are transformed into algorithmic constraint rules. Through iterative optimization of an improved genetic algorithm, global optimal matching and path planning for multiple fault points and multiple emergency repair teams are achieved, solving the problems of local optima and resource mismatch in traditional manual dispatching. Finally, at the control layer, a closed-loop control and self-optimization mechanism for all process nodes is constructed. By tracking the real-time status of key nodes in the entire emergency repair process, abnormal states such as node timeouts, path deviations, and safety risks are transformed into quantifiable, graded early warning trigger conditions, enabling automatic identification and proactive early warning of abnormal states. At the same time, through retrospective statistics and feature accumulation of data from the entire process, the fault feature library, judgment model, and scheduling algorithm are continuously iterated and optimized, forming a self-learning and self-optimizing endogenous cycle for the entire system, continuously improving the intelligence level and handling efficiency of distribution network emergency repair command.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] Deep integration of multi-source data: Breaking down system barriers, enabling spatiotemporal correlation of data from equipment, repair requests, GIS, meteorology, personnel, and materials, forming a complete fault view, and eliminating information silos;

[0046] Accurate and efficient fault assessment: The combination of rule engine and machine learning improves the accuracy of fault location and type identification, and multi-dimensional index quantification and ranking ensures that high-priority faults are handled first.

[0047] Intelligent optimization of emergency repair scheduling: Path optimization under multiple constraints to minimize the total repair time and achieve efficient matching of personnel, materials and vehicles;

[0048] Closed-loop management throughout the entire process: real-time tracking and intelligent early warning reduce node timeout and security risks, post-event statistics support operation and maintenance optimization, and significantly improve the intelligence level of distribution network emergency repair command and power supply reliability. Attached Figure Description

[0049] Figure 1 This is a flowchart of an intelligent dispatching method for power distribution network emergency repair command based on multi-source data fusion, according to the present invention. Detailed Implementation

[0050] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0051] like Figure 1 As shown, a method for intelligent dispatching of power distribution network emergency repair command based on multi-source data fusion includes the following steps:

[0052] A method for intelligent dispatching of power distribution network emergency repair command based on multi-source data fusion includes the following steps:

[0053] S1: Collect raw data from multiple sources of distribution network services, perform standardized preprocessing on the collected raw data, and obtain a standardized dataset in a unified format;

[0054] Furthermore, the raw data of the multi-source distribution network business collected in step S1 includes equipment operation and fault alarm data of the distribution network SCADA system and distribution automation system, user request data of the 95598 user repair system, spatial coordinate data of the GIS geographic information system, environmental data of the meteorological monitoring system, location and status data of the emergency repair personnel terminal, and inventory data of the material management system; the standardized preprocessing includes format unification, duplicate data removal, and outlier cleaning.

[0055] Furthermore, in another embodiment of the invention;

[0056] By connecting to the full-service distribution network system through standardized communication interfaces, raw datasets are collected. It includes seven core data subsets: equipment operation and fault alarm data from the distribution network SCADA / distribution automation system; user request data from the 95598 user repair system; spatial coordinate data from the GIS geographic information system; environmental data from the meteorological monitoring system; location and status data of emergency repair personnel terminals; inventory data from the materials management system; and real-time traffic data from the transportation platform. All individual data entries contain a unique equipment identifier. timestamp Geographic coordinates Three core related fields.

[0057] For the original dataset Perform preprocessing, specifically including:

[0058] 1. Remove duplicate data: Delete redundant data with identical device identifiers and timestamps;

[0059] 2. Outlier cleaning: The 3σ criterion is used to remove outliers from electrical quantities, coordinates, and time data, resulting in a cleaned dataset. .

[0060] 3. Perform min-max standardization on all numerical data in the cleaned dataset to uniformly map the values ​​to... The intervals are used to obtain a standardized dataset with a uniform format. The standardized formula is as follows:

[0061] ;

[0062] In the formula, For the first The output value after standardization of numerical data; For the first Measured values ​​of raw numerical data; It is the minimum value in the data indicator sequence; This is the maximum value in the data index sequence.

[0063] For non-numerical data such as equipment type and personnel skill tags, one-hot encoding is used to achieve a unified format, ensuring that all data meets the requirements of the standardized format.

[0064] S2: Using the unique device identifier as the matching condition, and combining the two dimensions of timestamp and geographic coordinates, perform spatiotemporal correlation fusion on the standardized dataset to obtain a fused dataset that corresponds one-to-one with a single fault event. Then, extract the core fault features from the fused dataset and construct a distribution network fault feature library.

[0065] Furthermore, the spatiotemporal correlation fusion described in step S2 specifically includes the following sub-steps:

[0066] 1) Using the unique device identifier as the matching condition, perform an initial screening of multi-source data in the standardized dataset, and retain only the data that matches the device identifier for subsequent matching;

[0067] 2) Extract the timestamp information of the data after initial screening, calculate the time difference between the data to be matched, and control the time difference within the preset time window to obtain the time dimension matching data;

[0068] 3) Extract the geographic coordinate information of the time dimension matching data, calculate the spatial distance between the data, and control the spatial distance within a preset range to obtain the spatial dimension matching data;

[0069] 4) The matching results of the three dimensions of device identification, time and space are weighted and calculated to obtain the comprehensive matching degree. Among them, the device identification dimension has the highest weight, the time dimension has the second highest weight, and the space dimension has the lowest weight. The sum of the weight coefficients of the three dimensions is 1.

[0070] 5) Compare the overall matching degree with the preset matching threshold. When the overall matching degree is greater than or equal to the preset matching threshold, it is determined that multiple sets of data belong to the same fault event, and the spatiotemporal correlation fusion of multi-source data is completed to generate a fusion dataset that corresponds one-to-one with the fault event.

[0071] Furthermore, the step S2, which involves extracting core fault features and constructing a distribution network fault feature library, specifically includes the following sub-steps:

[0072] 1) Extract six core fault features from the fused dataset: equipment operation exceeding limits amplitude, frequency of user repair reports in the same area, importance level of fault-related equipment, meteorological risk level of the fault area, number of users affected by the fault, and distribution network topology correlation.

[0073] 2) Normalize the six core fault features and convert them into standardized values ​​in the range of 0 to 1 to form a unified quantified fault feature vector;

[0074] 3) Associate and bind the fault feature vector with the corresponding fault type, fault location, and fault impact range to form a single fault feature sample;

[0075] 4) Perform secondary association and binding between a single fault feature sample and the corresponding historical emergency repair and handling plan, standard emergency repair duration, and final power restoration result;

[0076] 5) Store all the fault feature samples that have been associated and bound into a distributed database to build a distribution network fault feature library that can be called in real time and traced back in history.

[0077] Furthermore, in another embodiment of the invention;

[0078] Multi-source data spatiotemporal correlation fusion specifically includes:

[0079] Unique device identifier For matching conditions, on the standardized dataset Perform an initial screening, retaining only data with consistent device identifiers to obtain the initial screening dataset. Among them, the device's unique identifier It is obtained from the equipment operation data of the distribution network SCADA system and distribution automation system, specifically the power equipment asset code uniformly implemented by the power grid enterprise;

[0080] Initial screening dataset The formula is expressed as:

[0081] ;

[0082] In the formula, For standardized datasets A single data record in the data; A unique identifier for the target faulty device; This is the initial screening dataset after matching device identifiers.

[0083] Extract the timestamp of each data point from the initial screening dataset. Calculate the time difference between the data to be matched and the baseline fault alarm data. ,Will Control within the preset time window Within, a time-dimensional matching dataset is obtained. The formula is as follows:

[0084] ;

[0085] In the formula, The timestamp of the data to be matched; Timestamp for equipment baseline fault alarm data; This represents the time difference between the data to be matched and the baseline data. The preset time window threshold (value ranges from 5 to 30 minutes, dynamically adjusted according to the distribution network scale); This is the dataset after matching along the time dimension.

[0086] Extracting geographic coordinates of time-matched data The spatial distance between the data to be matched and the reference device is calculated using the Euclidean distance formula. ,Will Controlled within a preset distance threshold Within, a spatial dimension matching dataset is obtained. The formula is as follows:

[0087] ;

[0088] In the formula, The geographic coordinates of the data to be matched; The geographical coordinates of the reference faulty device; The spatial distance between the data to be matched and the reference device; The preset spatial distance threshold is set (values ​​range from 50 to 500m, which can be adjusted according to the power distribution radius). This is the dataset after spatial dimension matching.

[0089] The matching results of the three dimensions of device identification, time, and space are weighted and calculated to obtain the comprehensive matching degree. The weighting coefficients satisfy the following condition: the device identification dimension has the highest weight, followed by the time dimension, and the spatial dimension has the lowest weight, with the sum of the weights being 1. The formula is as follows:

[0090] ;

[0091] In the formula, The overall matching degree between multi-source data and target fault events; The matching score is set to 1 for a matching result and 0 for a mismatch. For time dimension matching degree, hour Otherwise, the value is 0; For spatial dimension matching degree, hour Otherwise, the value is 0; The weight coefficients for the three dimensions (values) , , ).

[0092] Overall matching degree Matching threshold (Taking a value of 0.7) Comparison, when When multiple sets of data are determined to belong to the same fault event, the spatiotemporal correlation fusion of multi-source data is completed, generating a fused dataset that corresponds one-to-one with a single fault event. .

[0093] The extraction of core fault features and the construction of a distribution network fault feature database specifically include:

[0094] From the fused dataset Six core fault features were extracted to form a feature set. ,in For equipment operating beyond the limit amplitude, For users in the same area, the frequency of repair requests For the importance level of the fault-related equipment, For the meteorological risk level of the fault area, The number of users affected by the fault This refers to the correlation degree of the distribution network topology.

[0095] The six core fault characteristics were normalized and converted into... Standardized values ​​within the interval form a unified, quantified fault feature vector. The formula is as follows:

[0096] ;

[0097] In the formula, To standardize the fault feature vector; The values ​​are normalized values ​​for the six core fault characteristics, and their ranges are all... .

[0098] Fault feature vectors are associated and bound with their corresponding fault types, fault locations, and impact ranges to form single fault feature samples. These samples are then further associated and bound with the corresponding historical emergency repair plans, standard repair durations, and power restoration results to form complete fault feature samples. All samples are stored in a distributed database to construct a distribution network fault feature library that can be accessed in real-time and has historical traceability capabilities. .

[0099] The structured mathematical expression for a single association binding is as follows:

[0100] The essence of this binding is to construct a one-to-one mapping relationship between "feature vector - fault attribute label" using the unique ID of the fault event as the primary key. Its mathematical expression is:

[0101] ;

[0102] In the formula, For the first A single fault feature sample corresponding to a fault event; For the first A globally unique ID for each fault event; This is the standardized fault feature vector corresponding to the fault event; The set of fault attribute labels bound to the feature vector is expressed as:

[0103] ;

[0104] In the formula: Standardized labels for fault types, including transformer faults, line breaks, and switch malfunctions, are generated through hot-coded processing and adapted to the feature vector format. For fault location tags, this includes binding a unique identifier for the faulty device with its precise geographic coordinates. ; Quantify the scope of the fault's impact with standardized values ​​for the number of affected users, the number of distribution transformers, and the power supply area.

[0105] The structured mathematical expression for secondary association binding is as follows: based on primary binding, historical closed-loop data is superimposed to form a complete sample, expressed as:

[0106] ;

[0107] In the formula, This is a set of historical handling tags corresponding to the fault event, including standardized historical emergency repair and handling plans, standard emergency repair duration, and final power restoration results, completing a full-dimensional association and binding.

[0108] S3: Based on the fused dataset and the distribution network fault feature library, the system completes intelligent fault assessment by combining a rule engine and a machine learning model, and outputs the fault type, accurate fault location, and fault impact range; then, based on four preset indicators, namely the importance of fault-related equipment, the scale of affected users, regional meteorological risk, and allowable power restoration time limit, the system completes the fault priority ranking and outputs a fault handling list with priority labels.

[0109] Furthermore, the intelligent fault assessment in step S3 specifically includes the following sub-steps:

[0110] 1) Retrieve the fusion dataset corresponding to the fault event, input it into the preset rule engine, and perform preliminary screening of the data through the preset fault judgment rules to filter out the suspected fault set that meets the fault triggering conditions; the preset fault judgment rules include electrical quantity abnormality triggering rules, user repair report related triggering rules, meteorological risk related triggering rules and distribution network topology related triggering rules.

[0111] 2) Retrieve the standardized fault feature vectors corresponding to the suspected fault set and input them into the pre-trained and validated random forest machine learning classification model;

[0112] 3) The fault feature vector is classified and calculated using a random forest machine learning classification model, and the output is the fault type, precise fault location area and fault impact range corresponding to the fault event.

[0113] Furthermore, the fault priority sorting in step S3 specifically includes the following sub-steps:

[0114] 1) Set corresponding weight coefficients for the four indicators respectively, and the sum of all weight coefficients is 1, among which the importance weight of fault-related equipment is the highest;

[0115] 2) Standardize and quantify the four indicators, converting them into score values ​​in the range of 0 to 1;

[0116] 3) Multiply the scores of each indicator by their corresponding weight coefficients and sum them to obtain the total priority score for the fault event;

[0117] 4) Determine the emergency repair priority of each fault according to the total priority score of all fault events from high to low, and generate a fault handling list with priority tags.

[0118] Furthermore, in another embodiment of the invention;

[0119] The intelligent fault assessment combining rule engines and machine learning specifically includes:

[0120] Retrieve the fused dataset corresponding to the fault event Input the preset rule engine, and complete the initial screening through fault triggering rules. This engine is built based on industry standards for distribution network emergency repair, historical fault data statistics, and expert experience. The preset rule set is as follows:

[0121] ;

[0122] In the formula, This is the complete set of rules for the preset rule engine; Rules for triggering electrical quantity anomalies; Associating user repair requests with trigger rules; Meteorological risk-related triggering rules; For distribution network topology association trigger rules;

[0123] Electrical quantity anomaly triggering rules:

[0124] Data source: Real-time operating data of equipment collected by the power distribution network SCADA system and power distribution automation system.

[0125] Hierarchical triggering logic and quantification rules:

[0126] Over-limit hard triggers: corresponding The core formula is that when the amplitude of the electrical quantity collected by the device in real time exceeds the preset rated threshold and the duration exceeds the anti-shake time (to filter out false alarms caused by instantaneous power grid disturbances), the rule is directly triggered.

[0127] Switch position change trigger: When a circuit breaker or load switch spontaneously trips without a remote control command, and is accompanied by a sudden drop in line current to 0 and a sudden drop in voltage, the rule is directly triggered and it is judged as a suspected fault trip.

[0128] Abnormal trend warning trigger: Electrical quantities do not directly exceed limits, but there are drastic fluctuations in a short period of time (such as voltage fluctuation rate exceeding ±20% / 10s), and equipment status quantities continue to climb and approach the threshold, triggering a warning-level anomaly and being included in the suspected monitoring range.

[0129] Key parameters: Over-limit threshold is set based on the "3kV~10kV Distribution Network Relay Protection Setting Regulations"; Anti-jitter duration is 20ms~500ms (adjusted according to equipment type); rule weight is 0.4.

[0130] User repair request associated trigger rules:

[0131] Data sources: User repair requests from the 95598 customer service system, the State Grid online APP, and government hotlines, along with supporting GIS power supply topology data.

[0132] Hierarchical triggering logic and quantization rules: Corresponding The core formula and core logic are the aggregation of demands from the same region, the same time period, and the same phenomenon:

[0133] Centralized repair triggering: Within a preset time window (patent recommendation 5~30min, adjusted according to urban / rural power supply density), if ≥3 users in the same transformer area report power outages / voltage abnormalities, or if ≥10 users on the same 10kV line report repairs in a concentrated manner, the rule will be triggered directly.

[0134] Strong triggering for critical users: For first- or second-level critical power supply users such as hospitals, government agencies, rail transit, and chemical enterprises, a single user reporting a power outage will trigger the rule directly, regardless of the frequency of the report, to meet the high priority requirements of power supply protection scenarios.

[0135] Enhanced triggering based on phenomenon matching: When the fault phenomena reported by users (such as power outages, frequent tripping, and unstable voltage) are highly matched with the abnormal electrical characteristics of the corresponding power supply equipment, the triggering weight of the rule is directly increased, and the determination of the authenticity of the fault is strengthened.

[0136] Key parameter: Repair frequency threshold (3 households at the district level, 10 households at the line level); time window 5~30min; rule weight 0.25.

[0137] Meteorological risk associated triggering rules:

[0138] Data sources: Real-time / forecast meteorological data from the national meteorological monitoring system and geological disaster early warning system, and GIS geographic coordinate data from supporting equipment.

[0139] Hierarchical triggering logic and quantization rules: Corresponding The core formula and logic are based on the spatial matching of meteorological risk levels and fault areas:

[0140] Meteorological Factor Classification: Sensitive meteorological factors for the power distribution network are divided into four risk levels, with high risk and above directly triggering the rules:

[0141] Highest risks: typhoons, blizzards, power line icing, wildfires, landslides;

[0142] High risk: Thunderstorms, gale-force winds of level 8 or above, hail, and short-term heavy rainfall;

[0143] Medium risk: High temperatures above 38℃, cold waves with a temperature drop of ≥10℃ in 24 hours;

[0144] Low risk: Regular rainfall, heavy fog, high humidity.

[0145] Spatial matching verification: If the geographical coordinates of the meteorological warning area and the power supply area of ​​the distribution network equipment overlap by ≥80%, the equipment in that area is determined to be within the scope of meteorological risk.

[0146] Dual-mode triggering:

[0147] Direct trigger: When the regional meteorological risk reaches high risk or above, all equipment in the region is included in the suspected fault monitoring scope, and the rule takes effect;

[0148] Enhanced triggering: Existing electrical anomalies / fault events reported by users, combined with the corresponding regional meteorological risk, directly increase the overall score and determine it as a high-probability real fault.

[0149] Key parameter: meteorological risk threshold (Medium risk and above); spatial overlap threshold ≥80%; temporal matching window is 30 minutes before and after the meteorological event; rule weight 0.2.

[0150] Distribution network topology association triggering rules:

[0151] Data sources: Electrical topology connection data from GIS geographic information system and distribution network CIM model, along with corresponding switch position changes and relay protection action data.

[0152] Fault source tracing trigger: When a device triggers an electrical quantity abnormality rule, trace the source up two levels along the distribution network topology (find the upstream switch, bus, and power supply point) and down all levels (find all downstream branches, transformer areas, and users) to pinpoint the source of the fault, predict the complete scope of impact, and include all related devices in the suspected fault set.

[0153] Chain fault triggering: When the upstream main equipment (such as 10kV busbar, outgoing switch) trips due to a fault, all downstream branch lines and transformer equipment, regardless of whether they trigger electrical quantity rules individually, are automatically included in the suspected fault set, and the rules take effect to avoid missing chain faults.

[0154] Ring network association trigger: For ring network power supply lines, when a fault occurs on one side of the line, the status of the ring network switch and the load status of the other side of the line are automatically associated to predict the risk of overload after power transfer and trigger the early warning rule.

[0155] Key parameters: Topology tracing level (2 levels upwards, all levels downwards); only applies to devices with direct electrical connections; rule weight 0.15.

[0156] Four rules are judged in parallel and weighted for scoring. The comprehensive score of a single fault event is calculated using binary logic with trigger = 1 and no trigger = 0. The score is calculated as follows: Electrical quantity rule score × 0.4 + User repair rule score × 0.25 + Meteorological risk rule score × 0.2 + Topology association rule score × 0.15.

[0157] A comprehensive score ≥ 0.6 indicates a valid suspected fault, which is then added to the suspected fault set and fed into the random forest model for accurate analysis.

[0158] A composite score < 0.6 indicates an invalid false alarm or a low-probability fault, which is then filtered out without proceeding to the subsequent machine learning stage.

[0159] First, the validity of the data is determined in parallel by each sub-rule, and then the overall rule score is calculated (each rule has a preset weight). , , , ), filter out Valid data, combined with the fault triggering rule quantification formula, is used to complete the initial screening and obtain a set of suspected faults. ;

[0160] After initial screening, a set of suspected faults was obtained. Meanwhile, it was clarified that the connection logic between the initial screening results and the subsequent random forest model is that low-confidence data are directly eliminated, while high-confidence data and data with confirmed faults are respectively input into the model for fine-tuning or verification, to ensure the efficiency and accuracy of the analysis.

[0161] Retrieve the standardized fault feature vectors corresponding to the suspected fault set Input a pre-trained and validated random forest classification model, and output the final judgment result through voting by multiple decision trees. The classification formula is as follows:

[0162] ;

[0163] In the formula, The final classification output of the model includes the fault type, precise fault location area, and fault impact range. The input is a standardized fault feature vector. ; This represents the number of decision trees in the random forest model. For the first The classification output of each decision tree; This is a set of fault classification tags, including tags for fault type, location, and scope of impact. This is an indicator function; it takes the value 1 if the condition inside the parentheses is true, and 0 otherwise. To obtain the category label with the largest value within the parentheses As the final output.

[0164] The complete training and validation implementation details of the random forest model are as follows:

[0165] ① Training dataset construction: The data source is the real distribution network fault historical samples of the target power supply area in the past 3 years verified on-site from the distribution network fault feature database. After removing invalid samples, the total number of valid samples is 12,800. The input of a single sample is a standardized feature vector composed of 6 core fault features, and the output label is fault type, fault location area, and fault impact range. The SMOTE oversampling algorithm is used to deal with the sample imbalance problem. The dataset is divided into training set (8,960 samples), validation set (1,920 samples), and test set (1,920 samples) in a ratio of 7:1.5:1.5. The fixed random seed is 42 to ensure the reproducibility of the results.

[0166] ② Core hyperparameter settings: Implemented using the Python language and the Scikit-learn framework, the core hyperparameter is fixed as: number of decision trees. Maximum depth of decision tree Minimum number of split samples for a node Minimum number of samples for leaf nodes Out-of-bag data ratio Random seed Number of parallel jobs .

[0167] ③ Model training process: Input the training set data into the initialized model, use sampling with replacement to construct a training subset of a single decision tree, randomly select 3 features from 6 features when splitting a node, select the feature with the smallest Gini coefficient to complete the split, until the stopping condition is reached; after all decision trees are trained, calculate the OOB error using out-of-bag data, fine-tune the hyperparameters using the validation set, and finally verify the generalization ability using the test set.

[0168] ④ Model qualification threshold: The model must simultaneously meet the following requirements: overall fault type classification accuracy ≥ 95%, single-type fault identification recall ≥ 90%, fault location area identification accuracy ≥ 92%, fault impact range identification error ≤ 10%, OOB error ≤ 5%, and single sample inference time ≤ 100ms before it can be put into field use.

[0169] ⑤ Model iteration mechanism: Adopt an iterative mode of quarterly incremental training + annual full training to continuously optimize the model's judgment accuracy.

[0170] Perform multi-dimensional fault priority ranking, specifically including:

[0171] Four core evaluation indicators were selected to form a priority evaluation set. ,in Importance of fault-related equipment, To affect user scale, For regional meteorological risks, To allow for a power restoration time limit, weight coefficients are assigned to four indicators, ensuring that the sum of the weights equals 1 and the equipment with the highest importance weight is selected. The formula is as follows:

[0172] ;

[0173] In the formula, This is the weight vector for priority evaluation indicators; The weighting coefficients for the four indicators (values) , , , ).

[0174] Standardization was applied to the four indicators to obtain... Standardized score within the interval The priority score of the fault events is calculated by weighted summation. The formula is as follows:

[0175] ;

[0176] In the formula, The overall priority score for fault events, with a value range of [value missing]. The higher the score, the higher the priority of emergency repair; The scores are standardized scores for the four core indicators.

[0177] Based on the total priority score of all fault events, sort them from highest to lowest to determine the emergency repair priority of each fault, and generate a fault handling list with priority tags. .

[0178] S4: Based on the fault handling list with priority labels, combined with the original data of multi-source distribution network services collected in S1, with the goal of minimizing the total repair time, set multiple constraints to complete the repair path optimization, generate the optimal repair scheduling scheme, and output the standardized repair scheduling task.

[0179] Furthermore, the multiple constraints in step S4 include: a single fault point is only the responsibility of one emergency repair team, the material carrying capacity of the emergency repair team does not exceed its maximum carrying capacity limit, and the time for the emergency repair team to reach the fault point does not exceed the maximum allowable power restoration time limit. The emergency repair path optimization is implemented using an improved genetic algorithm, with the fitness function being the minimization of the sum of the total emergency repair travel time and the task waiting time, and the multiple constraints being the selection rules, to iteratively solve for the optimal emergency repair scheduling scheme.

[0180] Furthermore, the standardized emergency repair scheduling task output in step S4 is simultaneously pushed to the mobile terminal of the corresponding emergency repair personnel. The pushed content includes the fault type, fault location coordinates, and fault impact range.

[0181] Furthermore, the algorithm used for path planning can be replaced with Dijkstra's algorithm or other algorithms depending on the on-site computational efficiency requirements. algorithm.

[0182] Furthermore, in another embodiment of the invention;

[0183] Perform mathematical modeling with multiple constraints:

[0184] With minimizing the total repair time as the core objective, three rigid constraints are set, as expressed mathematically below:

[0185] 1. The single fault point unique responsibility constraint ensures that each fault point is handled by exactly one emergency repair team, avoiding duplicate dispatching.

[0186] ;

[0187] In the formula, This represents the total number of fault points awaiting resolution. This represents the total number of available emergency repair teams. For decision variables, the first Each work group is responsible for handling the first... The value is 1 when there is a fault point, and 0 otherwise.

[0188] 2. The material carrying capacity constraint is that the total material demand undertaken by a single work team shall not exceed its maximum carrying capacity limit.

[0189] ;

[0190] In the formula, In order to handle the first The total amount of emergency repair materials required for each fault point; For the first The maximum material carrying capacity of each repair team's vehicles.

[0191] 3. The power restoration time limit constraint is that the time for the emergency repair team to arrive at the fault point must not exceed the maximum allowable power restoration time limit for that fault.

[0192] ;

[0193] In the formula, For the first The first repair team arrived at the... The time for each fault point; For the first The maximum allowable power restoration time for each fault point.

[0194] The path optimization based on the improved genetic algorithm is as follows:

[0195] An improved genetic algorithm is used to perform global optimization. A fitness function is constructed with the goal of minimizing the total repair time. The optimal repair scheduling scheme is iteratively solved using the three constraints mentioned above as selection rules. The fitness function formula is as follows:

[0196] ;

[0197] In the formula, The fitness function value, i.e., the total repair time, is the optimization objective to find the minimum value of this value. For the first The first repair team went to the... The travel time to each fault point is calculated based on real-time traffic conditions. For the first The waiting time for each emergency repair team is the interval between completing the previous task and starting the next task.

[0198] The specific improvements and implementation process of the improved genetic algorithm for distribution network emergency repair scenarios are as follows:

[0199] Improvement 1: Greedy initialization strategy based on fault priority

[0200] Faults are sorted from highest to lowest based on their total fault priority score. High-priority faults are prioritized for allocation to the nearest available work team that matches their skills, has sufficient resources, and complies with the power restoration time limit, thus generating an initial feasible solution. The initial feasible solution is then subjected to a random perturbation of the same priority to generate an initial population of 100, ensuring that 100% of the initial population are feasible solutions and accelerating the convergence speed.

[0201] Improvement 2: Constraint-oriented mutation operators and elite preservation strategies

[0202] The basic mutation probability is set to 0.05, and the mutation probability of gene loci that violate the constraints is increased to 0.2 to correct infeasible solutions in a targeted manner. In each iteration, the top 10% of elite individuals in terms of fitness within the population are retained, and crossover mutation is not performed to avoid the degradation of the optimal solution.

[0203] Improvement 3: Dual-threshold iteration termination strategy

[0204] Set the maximum number of iterations to 200, and also set the fitness change threshold. The iteration is terminated early when the change in optimal fitness is less than the threshold after 20 consecutive iterations, balancing solution accuracy and response speed. In emergency fault scenarios, the maximum number of iterations can be reduced to 50, and a feasible solution can be output within 10 seconds.

[0205] Based on the on-site computational efficiency requirements, the path planning algorithm can be replaced by Dijkstra's algorithm (single-source shortest path solution) or... The algorithm (heuristic path search) is adapted to the computing power requirements of different scenarios.

[0206] Replaced with Dijkstra's algorithm: Suitable for scenarios where a single fault point is handled independently and the terminal's computing power is limited. The three constraints are transformed into path weight penalty terms. The path weight for violating the constraints is set to infinity. The path weight is based on the travel time corresponding to real-time traffic conditions. It can be calculated locally on the mobile terminal with a response time of ≤1 second.

[0207] Replace with Algorithm: Suitable for scenarios with 2-5 concurrent fault points and edge computing capabilities; design a heuristic function. ( For fault priority weights, To estimate the shortest travel time, the constraints are set as path pruning rules. Paths are planned from high to low priority, with a response time of ≤5 seconds and a solution accuracy close to the global optimum.

[0208] After iterative convergence, the optimal emergency repair scheduling plan is output, a standardized emergency repair scheduling task is generated, and it is simultaneously pushed to the mobile terminals of the corresponding emergency repair personnel. The pushed content includes the fault type, fault location coordinates, fault impact range, on-site operation requirements, safety precautions, list of required emergency repair materials, and planned emergency repair route.

[0209] S5: Based on the standardized emergency repair scheduling task, the entire process nodes of task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation are tracked in real time. The abnormal status of node timeout, path deviation, and safety risk is automatically triggered with graded early warning. After the task is completed, a full-dimensional statistical report of emergency repair is generated.

[0210] Furthermore, step S5 specifically includes the following sub-steps:

[0211] S51: Based on the standardized emergency repair scheduling task, six control nodes are defined: task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation. The completion status and corresponding timestamp of each node are collected in real time through the mobile terminal of the emergency repair personnel and the power distribution automation system to realize real-time status tracking of the entire emergency repair process.

[0212] S52: For three types of abnormal states, namely node timeout, path deviation and security risk, pre-set quantitative graded warning triggering rules, divide the warning into three levels according to the severity of the abnormality, match the corresponding push objects, response time limits and closed-loop handling processes for different warning levels, and automatically trigger the corresponding graded warning when the abnormal state meets the warning triggering rules.

[0213] S53: Once all emergency repair tasks are completed, an automatic full-dimensional statistical report on emergency repair will be generated, including statistics on the entire fault process duration, fault type and cause analysis, resource consumption statistics, personnel performance statistics, abnormal event statistics, and power restoration completion statistics.

[0214] Automatic algorithm switching logic: The system has a built-in automatic selection module. When there is no central computing power for a single fault point, it automatically switches to the Dijkstra algorithm. When there are 2 to 5 fault points with edge computing power, it switches to the A algorithm. When there are more than 5 fault points with central computing power, it adopts the improved genetic algorithm. For emergency faults, the A algorithm is used to quickly output a solution.

[0215] Furthermore, in another embodiment of the invention;

[0216] Based on the location and status data of emergency repair dispatch tasks and personnel terminals, as well as inventory data from the materials management system, six core control nodes are constructed: task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation. The completion status and timestamps of each node are collected in real time through the mobile terminals of emergency repair personnel and the power distribution automation system, enabling real-time tracking of all nodes throughout the process.

[0217] For three types of abnormal states—node timeout, path deviation, and security risk—tiered early warning triggering rules are set, with the core quantitative formula as follows:

[0218] 1. Node timeout graded early warning

[0219] ;

[0220] In the formula, This is the difference between the actual completion time of the node and the standard duration. The actual completion time of the node; The preset standard completion time for this node; This is a warning level; the higher the level, the greater the risk.

[0221] 2. Path deviation warning

[0222] ;

[0223] In the formula, The distance between the real-time location of the repair personnel and the planned route; Provide the real-time geographic coordinates of the repair personnel; The coordinates of the nearest point on the planned path; The preset path deviation threshold; This is the path deviation warning trigger signal; a value of 1 triggers the warning.

[0224] The closed-loop response procedures for different warning levels are as follows:

[0225] Level 3 warning: Only pushed to the team leader of the emergency repair team. The team leader must provide feedback on the rectification within 10 minutes. If the warning is not cleared within 30 minutes, it will be automatically upgraded to Level 2 warning.

[0226] Level 2 warning: The warning is simultaneously pushed to the emergency repair specialist and the safety specialist. A response and supervision are required within 5 minutes. If the warning is not cleared within 1 hour, it will be automatically upgraded to Level 1 warning.

[0227] Level 1 warning: The warning will be simultaneously sent to the person in charge of the power supply center and the operation and maintenance and safety management department. The emergency plan will be activated within 2 minutes and a review will be completed within 3 working days after the incident is resolved.

[0228] Furthermore, in another embodiment of the invention;

[0229] Once the emergency repair task is completed, a comprehensive statistical report on the repair is automatically generated. This report includes statistics on the entire fault process duration, fault type and cause analysis, resource consumption statistics, personnel performance statistics, abnormal event statistics, and power restoration completion status. The report data is synchronously stored in the fault feature database, continuously iterating and optimizing the judgment model and scheduling algorithm to form a closed-loop operation mechanism for system self-learning and self-optimization.

[0230] A distribution network emergency repair command and dispatch system based on multi-source data fusion, used to implement the method, includes a data acquisition unit, a data fusion module, a fault analysis module, a resource scheduling module, and a visualization and control module connected in sequence, with the five modules corresponding to the method described above. The data acquisition unit performs operation S1, outputting a standardized dataset to the data fusion module. The data fusion module performs operation S2, outputting the fused dataset and a distribution network fault feature library to the fault analysis module. The fault analysis module performs operation S3, outputting a fault handling list with priority tags to the resource scheduling module. The resource scheduling module performs operation S4, outputting standardized emergency repair dispatch tasks to the visualization and control module. The visualization and control module performs operation S5, completing the entire emergency repair process control and anomaly early warning.

[0231] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. These embodiments are merely descriptions of preferred embodiments and are not intended to limit the scope or concept of the invention. The specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. Such combinations, as long as they do not violate the spirit of the present invention, should also be considered as part of this disclosure. To avoid unnecessary repetition, the present invention will not further describe the various possible combinations.

[0232] This invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this invention and without departing from the design idea of ​​this invention, all modifications and improvements made by those skilled in the art to the technical solutions of this invention should fall within the protection scope of this invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. A multi-source data fusion-based distribution network repair command intelligent scheduling method, characterized in that, Includes the following steps: S1: Collect raw data from multiple sources of distribution network services, perform standardized preprocessing on the collected raw data, and obtain a standardized dataset in a unified format; S2: Using the unique device identifier as the matching condition, and combining the two dimensions of timestamp and geographic coordinates, perform spatiotemporal correlation fusion on the standardized dataset to obtain a fused dataset that corresponds one-to-one with a single fault event. Then, extract the core fault features from the fused dataset and construct a distribution network fault feature library. S3: Based on the fused dataset and the distribution network fault feature library, the system completes intelligent fault assessment by combining a rule engine and a machine learning model, and outputs the fault type, accurate fault location, and fault impact range; then, based on four preset indicators, namely the importance of fault-related equipment, the scale of affected users, regional meteorological risk, and allowable power restoration time limit, the system completes the fault priority ranking and outputs a fault handling list with priority labels. S4: Based on the fault handling list with priority labels, combined with the original data of multi-source distribution network services collected in S1, with the goal of minimizing the total repair time, set multiple constraints to complete the repair path optimization, generate the optimal repair scheduling scheme, and output the standardized repair scheduling task. S5: Based on the standardized emergency repair scheduling task, the entire process nodes of task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation are tracked in real time. The abnormal status of node timeout, path deviation, and safety risk is automatically triggered with graded early warning. After the task is completed, a full-dimensional statistical report of emergency repair is generated. 2.The multi-source data fusion based intelligent dispatching method for power distribution network repair command according to claim 1, characterized in that, The raw data of the multi-source distribution network business collected in step S1 includes equipment operation and fault alarm data of the distribution network SCADA system and distribution automation system, user request data of the 95598 user repair system, spatial coordinate data of the GIS geographic information system, environmental data of the meteorological monitoring system, location and status data of emergency repair personnel terminals, and inventory data of the material management system; the standardized preprocessing includes format unification, duplicate data removal, and outlier cleaning. 3.The multi-source data fusion based intelligent dispatching method for power distribution network repair command according to claim 1, characterized in that, The spatiotemporal correlation fusion described in step S2 specifically includes the following sub-steps: 1) Using the unique device identifier as the matching condition, perform an initial screening of multi-source data in the standardized dataset, and retain only the data that matches the device identifier for subsequent matching; 2) Extract the timestamp information of the data after initial screening, calculate the time difference between the data to be matched, and control the time difference within the preset time window to obtain the time dimension matching data; 3) Extract the geographic coordinate information of the time dimension matching data, calculate the spatial distance between the data, and control the spatial distance within a preset range to obtain the spatial dimension matching data; 4) The matching results of the three dimensions of device identification, time and space are weighted and calculated to obtain the comprehensive matching degree. Among them, the device identification dimension has the highest weight, the time dimension has the second highest weight, and the space dimension has the lowest weight. The sum of the weight coefficients of the three dimensions is 1. 5) Compare the overall matching degree with the preset matching threshold. When the overall matching degree is greater than or equal to the preset matching threshold, it is determined that multiple sets of data belong to the same fault event, and the spatiotemporal correlation fusion of multi-source data is completed to generate a fusion dataset that corresponds one-to-one with the fault event.

4. The multi-source data fusion-based network distribution repair command intelligent scheduling method according to claim 1 or 3, characterized in that, Step S2, which involves extracting core fault features and constructing a distribution network fault feature library, specifically includes the following sub-steps: 1) Extract six core fault features from the fused dataset: equipment operation exceeding limits amplitude, frequency of user repair reports in the same area, importance level of fault-related equipment, meteorological risk level of the fault area, number of users affected by the fault, and distribution network topology correlation. 2) Normalize the six core fault features and convert them into standardized values ​​in the range of 0 to 1 to form a unified quantified fault feature vector; 3) Associate and bind the fault feature vector with the corresponding fault type, fault location, and fault impact range to form a single fault feature sample; 4) Perform secondary association and binding between a single fault feature sample and the corresponding historical emergency repair and handling plan, standard emergency repair duration, and final power restoration result; 5) Store all the fault feature samples that have been associated and bound into a distributed database to build a distribution network fault feature library that can be called in real time and traced back in history. 5.The multi-source data fusion based intelligent dispatching method for distribution network repair command according to claim 1, characterized in that, The intelligent fault assessment in step S3 specifically includes the following sub-steps: 1) Retrieve the fusion dataset corresponding to the fault event, input it into the preset rule engine, and perform preliminary screening of the data through the preset fault judgment rules to filter out the suspected fault set that meets the fault triggering conditions; the preset fault judgment rules include electrical quantity abnormality triggering rules, user repair report related triggering rules, meteorological risk related triggering rules and distribution network topology related triggering rules. 2) Retrieve the standardized fault feature vectors corresponding to the suspected fault set and input them into the pre-trained and validated random forest machine learning classification model; 3) The fault feature vector is classified and calculated using a random forest machine learning classification model, and the output is the fault type, precise fault location area and fault impact range corresponding to the fault event. 6.The multi-source data fusion based intelligent dispatching method for power distribution network repair command according to claim 5, characterized in that, The fault priority sorting in step S3 specifically includes the following sub-steps: 1) Set corresponding weight coefficients for the four indicators respectively, and the sum of all weight coefficients is 1, among which the importance weight of fault-related equipment is the highest; 2) Standardize and quantify the four indicators, converting them into score values ​​in the range of 0 to 1; 3) Multiply the scores of each indicator by their corresponding weight coefficients and sum them to obtain the total priority score for the fault event; 4) Determine the emergency repair priority of each fault according to the total priority score of all fault events from high to low, and generate a fault handling list with priority tags. 7.The multi-source data fusion based intelligent dispatching method for distribution network repair command according to claim 1, characterized in that, The multiple constraints in step S4 include: a single fault point is handled by only one emergency repair team, the material carrying capacity of the emergency repair team does not exceed its maximum carrying capacity, and the time for the emergency repair team to reach the fault point does not exceed the maximum allowable power restoration time limit. The emergency repair path optimization is implemented using an improved genetic algorithm, with the fitness function being the minimization of the sum of the total emergency repair travel time and the task waiting time, and the multiple constraints being the selection rules, to iteratively solve for the optimal emergency repair scheduling scheme. 8.The multi-source data fusion based intelligent dispatching method for distribution network repair command according to claim 1, characterized in that, The standardized emergency repair scheduling task output in step S4 is simultaneously pushed to the mobile terminal of the corresponding emergency repair personnel. The pushed content includes the fault type, fault location coordinates, and fault impact range. 9.The multi-source data fusion based intelligent dispatching method for distribution network repair command according to claim 7, characterized in that, Step S5 specifically includes the following sub-steps: S51: Based on the standardized emergency repair scheduling task, six control nodes are defined: task issuance, personnel departure, arrival at the site, fault handling, power restoration completion, and site evacuation. The completion status and corresponding timestamp of each node are collected in real time through the mobile terminal of the emergency repair personnel and the power distribution automation system to realize real-time status tracking of the entire emergency repair process. S52: For three types of abnormal states, namely node timeout, path deviation and security risk, pre-set quantitative graded warning triggering rules, divide the warning into three levels according to the severity of the abnormality, match the corresponding push objects, response time limits and closed-loop handling processes for different warning levels, and automatically trigger the corresponding graded warning when the abnormal state meets the warning triggering rules. S53: Once all emergency repair tasks are completed, an automatic full-dimensional statistical report on emergency repair will be generated, including statistics on the entire fault process duration, fault type and cause analysis, resource consumption statistics, personnel performance statistics, abnormal event statistics, and power restoration completion statistics.

10. A multi-source data fusion-based distribution network repair command intelligent scheduling system for implementing the method of any one of claims 1-9, characterized in that, The system includes a data acquisition unit, a data fusion module, a fault analysis module, a resource scheduling module, and a visualization and control module, which are connected in sequence. The data acquisition unit is used to collect raw data from multiple sources of distribution network services, perform standardized preprocessing on the raw data, and output a standardized dataset in a unified format to the data fusion module. The data fusion module is used to perform spatiotemporal correlation fusion of multi-source data with the unique identifier of the device as the core matching condition, combined with timestamps and geographical coordinates, to generate a fusion dataset corresponding to a single fault event, extract core fault features to construct a distribution network fault feature library, and output it to the fault analysis module. The fault assessment module is used to complete intelligent fault assessment by combining a rule engine and a random forest model. It performs fault priority ranking by weighted quantification of four core indicators and outputs a fault handling list with priority labels to the resource scheduling module. The resource scheduling module is used to optimize the emergency repair path based on multiple constraints with the goal of minimizing the total repair time, generate the optimal scheduling scheme, and output the standardized emergency repair scheduling task to the visualization and management module. The visualization and control module is used to realize real-time tracking of all nodes in the emergency repair process, graded early warning of abnormal status, and generate a comprehensive statistical report on emergency repair after the task is completed.

Images