A power failure fault research and judgment method based on a knowledge graph and power failure big data analysis

Abstract

The application discloses a power failure fault research and judgment method based on a knowledge graph and power failure big data analysis, reconstructs power grid topological relations through a neo4j knowledge graph, forms triples of entities, attributes and relations, and divides the graph into different sections with switches as nodes; searches and determines power failure equipment and power failure areas through a graph search algorithm combined with a call-measuring algorithm; and corrects the power failure equipment and power failure areas through a backtracking algorithm to query a delay signal of the call-measuring algorithm. In the framework of the knowledge graph, the traceable searchability of the power grid topology is brought out, and the live signals returned by the call-measuring algorithm are combined to check the power grid section by section, so that the live states of dozens of transformers in one section can be checked within 3 minutes, the efficiency of power failure checking is greatly improved, a large amount of manual work is saved, the power failure response time is greatly shortened, and the accuracy is very high. The application has the advantages of clear thought, good universality, high economic value and suitability for popularization and use.

Description

Technical Field

[0001] This invention relates to the field of power outage area sensing technology in power distribution networks, specifically a power outage fault assessment method based on knowledge graphs and power outage big data analysis. Background Technology

[0002] As people's living standards continue to improve, users have increasingly higher requirements for power supply quality and reliability. The power distribution network directly serves a large number of users and is related to their electricity experience. Once a power distribution network system fault occurs, rapid and accurate fault location and isolation, as well as timely assessment of the impact of the fault, play an extremely important role in minimizing power outages and improving power supply safety and reliability.

[0003] As a crucial link in the production, transmission, and use of electricity, the power distribution system is a key connection between the actual user demand side and the power generation and transmission system. Therefore, rapid and accurate fault location after a fault occurs in the distribution network, effectively guiding maintenance work, is a critical aspect of the safe and sustainable development of urban power grids. However, several factors contribute to difficulties in fault assessment during the distribution network process: ① Harsh environments affecting network communication devices impact information flow; ② Inconsistent levels of distribution automation development create numerous uncertainties in fault information, affecting its integrity; ③ In cases of multiple complex faults, numerous faulty and non-faulty components exist in the power outage area, and circuit breaker protection may fail to operate or malfunction, interfering with fault assessment. These factors all lead to an expanded assessment scope and distorted information transmission, impacting the safe and stable operation of the distribution system. With increasingly complex wiring configurations and a growing variety of equipment components, researching big data algorithms for power outage fault assessment and improving fault assessment accuracy is of paramount importance for the comprehensive development of the power system.

[0004] This method proposes a fault outage assessment approach that uses knowledge graphs to reconstruct searchable power grid topology and combines it with big data analytics algorithms. This not only enables automatic assessment of outages based on outage signals in a short time, improving the efficiency of outage investigation and providing timely feedback to affected users, but also frees up significant manpower resources for investigation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a power outage fault assessment method based on knowledge graphs and big data analysis of power outages. This method can quickly respond to power outage signals without manual intervention, relying solely on electrical signals to assess the affected equipment. This improves the speed of power restoration after outages, stabilizes customer sentiment, and enhances the quality of power grid services. Specifically, this invention is implemented as follows:

[0006] A method for power outage fault assessment based on knowledge graphs and power outage big data analysis includes the following steps:

[0007] Step S1: Embed the big data information of the power grid equipment in the transmission line that needs to be analyzed into the neo4 knowledge graph, and perform thinning processing on the original graph to form single-line data of the line, thus obtaining the constructed neo4 knowledge graph.

[0008] Step S2: Number and label the power grid devices in the constructed neo4j knowledge graph, and divide them into several different segments with switches as nodes, and number them accordingly;

[0009] Step S3: Obtain transformer outage signal data for power grid equipment in each line and filter them;

[0010] Step S4: Use Cypher language to build the Neo4j graph search algorithm, combined with the transformer recall algorithm, to search for the outage area in the graph towards the upstream substation through the outage signal. If the recall signal is energized, stop the search, merge the same outage data and return to the outage area equipment; if the recall signal is not energized, proceed to step S5.

[0011] Step S5: After completing the search of a segment, backtrack the recall delay signal of the previous segment. If no delay signal is returned, search for the next segment of the power outage area in the upstream substation direction in the map and execute step S4. If a delay signal is returned, stop the next stage of the search and return to the previous segment.

[0012] Step S6: When a charged area is found, stop the search, correct or merge the power outage areas using the neo4j graph structure, and return to the equipment in the power outage area.

[0013] Furthermore, the filtering step in step S3 includes: discretizing the processed line single-line diagram with switches as nodes, filtering the power outage signal data, using a line clustering algorithm to perform cluster analysis on power outage signals with short time intervals on the same line, and deleting power outage signals with low reliability without processing.

[0014] Furthermore, in step S1, before embedding the big data information of the power grid equipment in the transmission line to be analyzed into the neo4 knowledge graph, entity and relationship recognition is performed using the transformer model lac encapsulated by the Baidu Paddle framework. This also includes: extracting and processing the GIS data of the power grid equipment to make it into triplet data of entity nodes, entity attributes, and relationships between entities, and embedding it into the neo4j graph database.

[0015] Furthermore, step S6 also includes: if the delayed signal of the backtracking is energized, then according to the segment number identified in the backtracking algorithm, all segments from that segment onwards are identified as energized, while the previous segments are still identified as de-energized, and the devices in these segments are returned.

[0016] Furthermore, in step S4, the algorithm is written in Java. When the algorithm receives a transformer power outage alarm signal, it first takes the alarm transformer as the starting point and uses the Cypher Neo4j graph search language to find the direction of the substation of the line, and divides the line between the transformer and the substation into segments with switches as nodes.

[0017] Furthermore, in step S4, if the number of transformers in the section exceeds 10, in order to prevent the recall algorithm channel from being blocked, 10 transformers are randomly selected for recall.

[0018] The working principle and beneficial effects of this invention are as follows: This invention reconstructs the power grid topology using the Neo4j knowledge graph, forming triples of entities, attributes, and relationships. The graph is divided into different segments with switches as nodes. A graph search algorithm combined with a recall algorithm is used to search and determine outage equipment and outage areas. A backtracking algorithm queries the delayed signals from the recall algorithm to correct the outage equipment and areas. Within the framework of the knowledge graph, the traceability and searchability of the power grid topology are fully utilized. Combined with the energized signals returned by the recall algorithm, the power grid is checked segment by segment. The energized status of dozens of transformers within a segment can be checked within 3 minutes, greatly improving the efficiency of outage investigation. This not only saves a lot of manpower and significantly shortens outage response time, but also has very high accuracy. This invention can quickly, accurately, and automatically find all relevant outage equipment for the current outage event, saving manpower for line inspection, accelerating fault repair, and improving the quality of power grid service. This invention has the advantages of clear thinking, good versatility, high economic value, and suitability for widespread application. Attached Figure Description

[0019] Figure 1 This is a flowchart of a power outage fault assessment method based on knowledge graph and power outage big data analysis according to the present invention.

[0020] Figure 2 This is a schematic diagram of the data storage structure in the backtracking algorithm of this invention;

[0021] Figure 3 This is a judgment result of the present invention in a practical application of a power grid. The dark lines represent energized lines, the light lines represent de-energized lines, and the junction of the energized and de-energized lines is the fault point. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0023] Example 1: A method for power outage fault assessment based on knowledge graph and power outage big data analysis

[0024] The GIS data of power grid equipment is extracted and processed into triples of entity nodes, entity attributes, and relationships between entities. These triples are then embedded into the Neo4j graph database using the neo4j-admin import method to form a knowledge graph. The knowledge graph is then thinned to remove some irrelevant entity nodes and their related relationships, simplifying the knowledge graph and forming individual line diagrams for easier operation.

[0025] The processed line neo4j graph single-line diagram is discretized into segmented single-line diagrams with switches as nodes;

[0026] The power outage signal data is filtered, and the line clustering method is used to perform cluster analysis on power outage signals with short time intervals on the same line. Power outage signals with low reliability are deleted and not processed.

[0027] A graph search algorithm written in Neo4j's built-in Cypher language, combined with a transformer recall algorithm, analyzes the returned information. If no transformer signal is returned for a given section, it indicates a high probability of power outage. The search then proceeds upstream towards the next power outage area in the graph. If a signal indicating power is returned, the search stops at this stage. The Cypher search algorithm is as follows:

[0028] Search all devices within a specific range 1:

[0029] MATCH p=(ns{dev_id:'%s'})-[*]-(dev)

[0030] where apoc.coll.duplicates(nodes(p))=[]

[0031] and all(x in nodes(p)where x.feederId='%s')

[0032] and all(x in nodes(p)where not x.dev_id='%s')

[0033] and all(x in relationships(p) where x.feederid = '%s')

[0034] return dev.dev_id AS DeviceID, labels(dev) AS DeviceLabel, dev.feederId AS FeederID, dev.classId AS ClassID, dev.buro AS Buro, dev.subBuro AS SubBur, dev.trans_id AS TransformerID, dev.fromDev AS fromDevID",

[0035] Search for all devices in a certain section 2:

[0036] MATCH p = shortestpath((s{dev_id: '%s'})-[*]-(e{dev_id: '%s'}))

[0037] where apoc.coll.duplicates(nodes(p)) = []

[0038] with [head(relationships(p))] as h_r

[0039] match pp = (ns{dev_id: '%s'})-[*]-(dev)

[0040] where apoc.coll.duplicates(nodes(pp)) = []

[0041] and all(x in nodes(pp) where x.feederId = '%s')

[0042] and all(x in relationships(pp) where not x in h_r and x.feederid = '%s')

[0043] and all(x in nodes(pp) where not x.dev_id = '%s')

[0044] return dev.dev_id AS DeviceID, labels(dev) AS DeviceLabel, dev.feederId AS FeederID, dev.classId AS ClassID, dev.buro AS Buro, dev.subBuro AS SubBur, dev.trans_id AS TransformerID, dev.fromDev AS fromDevID"

[0045] Search for the shortest path:

[0046] MATCH (dev_start {dev_id: '%s'}), (dev_end {dev_id: '%s'}),

[0047] path = shortestPath((dev_start)-[*]-(dev_end))

[0048] WHERE apoc.coll.duplicates(nodes(path)) = []

[0049] RETURN length(path) AS PathLength

[0050] Search for the nearest upstream switch on the path:

[0051] MATCH (dev_start {dev_id: '%s'}), (dev_end {dev_id: '%s'}),

[0052] path = shortestPath((dev_start)-[*]-(dev_end))

[0053] WHERE apoc.coll.duplicates(nodes(path)) = []

[0054] WITH path

[0055] MATCH (dev)

[0056] WHERE dev IN nodes(path)

[0057] AND (dev: %s OR dev: %s OR dev: %s)

[0058] AND NOT dev.dev_id = '%s'

[0059] RETURN dev.dev_id AS DeviceID;

[0060] After completing a phase of the search, backtrack to check the recall delay signal of the previous segment. If the transformer signal of the previous segment does not return a delayed signal, it means that there is likely no power. Then, search for the next segment of the power outage area in the upstream substation direction in the map. If the returned signal indicates that there is power, stop the next phase of the search.

[0061] When a charged area is found, the search stops and the devices in the de-energized area are returned. If the delayed signal from the backtracking is charged, all subsequent segments are marked as charged according to the segment number identified in the backtracking algorithm, while the previous segments are still marked as de-energized. The de-energized areas are corrected or merged based on the neo4j graph structure, and the devices in these segments are returned.

[0062] This invention can quickly, accurately, and automatically locate all relevant power outage equipment in a current power outage event, saving manpower for line inspection, accelerating fault repair, and improving the quality of power grid service. The main features of this invention are:

[0063] 1. Reconstruct the power grid topology using the Neo4j knowledge graph, forming triples of entities, attributes, and relationships, and divide the graph into different segments with switches as nodes;

[0064] 2. Implement a graph search algorithm combined with a recall algorithm using Neo4j's Cypher language to search for and determine power outage equipment and areas;

[0065] 3. Correct outage equipment and merge outage areas by constructing a new neo4j knowledge graph structure.

[0066] The beneficial effects of this invention are that, within the framework of the Neo4j knowledge graph, it leverages the traceability and searchability of the power grid topology. By utilizing the convenience of the Cypher language to write a more easily processed graph search algorithm, and combining it with the energized signals returned by the recall algorithm, it performs segment-by-segment checks on the power grid. It can check the energized status of dozens of transformers within a section within 3 minutes, greatly improving the efficiency of power outage troubleshooting. This not only saves a significant amount of manpower and greatly shortens the power outage response time, but also achieves very high accuracy. This invention has the advantages of clear concept, good versatility, high economic value, and suitability for widespread application.

[0067] Example 2: Case Study

[0068] A power outage fault assessment method based on knowledge graphs and power outage big data analysis is presented. The method presents the assessment process and results of a power grid operation in a specific region, and conducts a nearly year-long assessment of power outage equipment for that power grid from January 2021 to early December 2021. The specific steps of one power outage assessment are as follows:

[0069] (1) The latest GIS data of the power grid is converted into triples of entities, attributes and relationships, and the triples are embedded into the neo4j graph database using the neo4j-admin import method provided by neo4j. The specific statement is: neo4j-admin import --database = neo4j --nodes "import / sto.csv" --nodes "import / con.csv" --relationships "import / rel.csv" --ignore-empty-strings --skip-bad-relationships --skip-duplicate-nodes = true, and a knowledge graph of single-line diagram of transmission lines is constructed.

[0070] (2) The power outage signal filtering algorithm is written in Java. After receiving the alarm signal of transformer power failure, the power outage signal filtering algorithm determines that the power failure alarm is real and transmits the signal to the power failure equipment search and judgment algorithm.

[0071] (3) The power outage equipment search and judgment algorithm is written in Java. When the algorithm receives the transformer power outage alarm signal, it first takes the alarm transformer as the starting point, uses the Cypherneo4j graph search language to find the direction of the substation of the line, and divides the line between the transformer and the substation into sections with switches as nodes.

[0072] (4) Using the graph search algorithm written in the Cypher language provided by Neo4j, the transformers in the next section of the line are searched, and a recall is performed on the search results. If the number of transformers in the section exceeds 10, 10 transformers are randomly selected for recall to prevent congestion of the recall algorithm channel. The Cypher search algorithm is as follows:

[0073] MATCH p=(ns{dev_id:'%s'})-[*]-(dev)

[0074] where apoc.coll.duplicates(nodes(p))=[]

[0075] and all(x in nodes(p)where x.feederId='%s')

[0076] and all(x in nodes(p)where not x.dev_id='%s')

[0077] and all(x in relationships(p)where x.feederid='%s')

[0078] return dev.dev_id AS DeviceID,labels(dev)AS DeviceLabel,dev.feederIdAS FeederID,dev.classId AS ClassID,dev.buro AS Buro,dev.subBuro AS SubBur,dev.trans_id AS TransformerID,dev.fromDev AS fromDevID",

[0079] MATCH p=shortestpath((s{dev_id:'%s'})-[*]-(e{dev_id:'%s'}))

[0080] where apoc.coll.duplicates(nodes(p))=[]

[0081] with[head(relationships(p))]as h_r

[0082] match pp=(ns{dev_id:'%s'})-[*]-(dev)

[0083] where apoc.coll.duplicates(nodes(pp))=[]

[0084] and all(x in nodes(pp)where x.feederId='%s')

[0085] and all(x in relationships(pp)where not x in h_r and x.feederid='%s')

[0086] and all(x in nodes(pp)where not x.dev_id='%s')

[0087] return dev.dev_id AS DeviceID,labels(dev)AS DeviceLabel,dev.feederIdAS FeederID,dev.classId AS ClassID,dev.buro AS Buro,dev.subBuro AS SubBur,dev.trans_id AS TransformerID,dev.fromDev AS fromDevID",

[0088] MATCH(dev_start{dev_id:'%s'}),(dev_end{dev_id:'%s'}),

[0089] path=shortestpath((dev_start)-[*]-(dev_end))

[0090] WHERE apoc.coll.duplicates(nodes(path))=[]

[0091] RETURN length(path)AS PathLength

[0092] MATCH(dev_start{dev_id:'%s'}),(dev_end{dev_id:'%s'}),

[0093] path=shortestpath((dev_start)-[*]-(dev_end))

[0094] WHERE apoc.coll.duplicates(nodes(path))=[]

[0095] WITH path

[0096] MATCH(dev)

[0097] WHERE dev IN nodes(path)

[0098] AND(dev:%s OR dev:%s OR dev:%s)

[0099] AND NOT dev.dev_id='%s'

[0100] RETURN dev.dev_id AS DeviceID

[0101] (5) Analyze the transformer recall return results, remove the recall results in the transformer whitelist and do not analyze them. After the remaining transformer return results are analyzed to be no power, enter the recall backtracking algorithm. The recall backtracking algorithm is written in Java language. Review the delayed return signal of the previous round of recall. If the analysis result is still no power, enter the next stage of search to find the transformer in the next line section and recall it.

[0102] (6) When the recall return signal analysis result or the delayed signal analysis result returned by the recall backtracking algorithm indicates that there is power, the search for upstream substations is stopped. Then, the outage areas are modified or merged using the graph structure constructed by neo4j, and finally all currently searched outage transformers and other line equipment are returned. See the final results. Figure 2 .

[0103] As can be seen from the above calculation process, a power outage fault assessment method based on knowledge graph and power outage big data analysis can quickly, accurately and automatically find all relevant power outage equipment in the current power outage event, save manpower to check lines, speed up fault repair, and improve the quality of power grid services.

[0104] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. A method for judging power outage faults based on knowledge graphs and power outage big data analysis, characterized in that... Includes the following steps, Step S1: Embed the big data information of the power grid equipment in the transmission line that needs to be analyzed into the neo4 knowledge graph, and perform thinning processing on the original graph to form single-line data of the line, thus obtaining the constructed neo4 knowledge graph. Before embedding the big data information of the power grid equipment in the transmission lines that need to be analyzed into the neo4 knowledge graph, it also includes: extracting and processing the GIS data of the power grid equipment, turning it into triple data of entity nodes, entity attributes, and relationships between entities, and embedding it into the neo4j graph database. Step S2: Number and label the power grid devices in the constructed neo4j knowledge graph, and divide them into several different segments with switches as nodes, and number them accordingly; Step S3: Obtain and filter the transformer outage signal data of the power grid equipment in each line. Step S4: Construct a Neo4j graph search algorithm using the Cypher language. The algorithm is written in Java. When the algorithm receives a transformer outage alarm signal, it first uses the alarm transformer as the starting point and uses the Cypher Neo4j graph search language to find the direction of the substation on the line. The line between the transformer and the substation is divided into segments with switches as nodes. Combined with the transformer recall algorithm, the algorithm searches for the outage area upstream of the substation in the graph based on the outage signal. If the recall signal is energized, the search stops, identical outage data is merged, and the outage area equipment is returned. If the recall signal is not energized, proceed to step S5. Step S5: After completing the search of a segment, backtrack the recall delay signal of the previous segment. If no delay signal is returned, search for the next segment of the power outage area in the upstream substation direction in the map and execute step S4. If a delay signal is returned, stop the next stage of the search and return to the previous segment. Step S6: When a charged area is found, stop the search, correct or merge the power outage areas using the neo4j graph structure, and return to the equipment in the power outage area.

2. The power outage fault assessment method according to claim 1, characterized in that, The filtering step in step S3 includes: discretizing the processed line single-line diagram with switches as nodes, filtering the power outage signal data, using a line clustering algorithm to perform cluster analysis on power outage signals with short time intervals on the same line, and deleting power outage signals with low reliability without processing.

3. The power outage fault assessment method according to claim 1, characterized in that, Step S6 further includes: if the delayed signal of the backtracking is energized, then according to the segment number identified in the backtracking algorithm, all segments starting from that segment are identified as energized, while the previous segments are still identified as de-energized, and the devices in these segments are returned.

4. The power outage fault assessment method according to claim 1, characterized in that, In step S4, if the number of transformers in the section exceeds 10, in order to prevent the recall algorithm channel from being blocked, 10 transformers are randomly selected for recall.

Images