Frequent power outage analysis and root cause identification system for distribution network lines based on intelligent analysis

Through the intelligent analysis system combined with LSTM, graph neural network and Seq2Seq model, the problem of the failure of the existing technology to accurately identify the frequent power outages of distribution network lines is solved, real-time monitoring and fault identification of the power grid is realized, and the efficiency and reliability of the power grid is improved. sex.

CN118885978BActive Publication Date: 2024-11-29ANHUI JIYUAN SOFTWARE CO LTD

Patent Information

Application Number
CN202411393707.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-08
Publication Date
2024-11-29
Estimated Expiration
2044-10-08

AI Technical Summary

Technical Problem

The existing power outage prediction system is based on a static analysis model and cannot accurately describe the characteristics and data sources of complex distribution network power systems, resulting in the inability to effectively identify the root causes of frequent power outages on distribution network lines.

Method used

The distribution network line frequent power outage analysis and root cause identification system based on intelligent analysis, including acquisition and processing module, prediction and identification module, decision optimization module and visualization module, is used to combine LSTM network, graph neural network and Seq2Seq model to perform data preprocessing, Root cause analysis, decision support and early warning.

Benefits of technology

Real-time monitoring of distribution network systems and accurate identification of fault types and causes, improve grid management efficiency and reliability, and reduce the impact of frequent power outages on social life and production.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118885978B_ABST
    Figure CN118885978B_ABST
Patent Text Reader

Abstract

The present invention discloses a system for judging and identifying the root causes of frequent power outages in distribution network lines based on intelligent analysis, and relates to the technical field of power grid fault identification. Compared with previous power outage prediction systems, the present invention solves the problem that most existing power outage prediction systems perform power outage analysis based on static analysis models, and cannot cope with the complex power characteristics and data sources of the power system; it introduces an intelligent root cause identification algorithm for power outage events, conducts comprehensive modeling and analysis on multiple data from different data sources, combines the attention mechanism to improve the degree of attention to key data, and also combines the graph neural network to capture the topological structure and mutual influence between devices, which can detect the location of the fault and classify the type and cause of the fault; at the same time, combining the results of data analysis and root cause identification, the system provides comprehensive decision support; it realizes the ability to monitor the distribution network system in real time, and can immediately issue an alarm when an abnormality or potential power outage risk occurs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power grid fault identification, and in particular to a system for analyzing and identifying root causes of frequent power outages of distribution network lines based on intelligent analysis. Background Art

[0002] The distribution network is the most relevant power grid in the power network to residents' lives. With the continuous improvement of residents' living standards, users have higher and higher requirements for power supply reliability. However, due to the complex operation and grid structure of distribution network lines, the large number of lines and equipment, and the unreliable monitoring and information channels of line conditions in large areas, distribution network lines frequently experience power outages, which usually cause huge economic losses, affect other key infrastructure systems, and seriously disrupt daily life. Therefore, it is necessary to analyze power outage data.

[0003] In order to meet the electricity demand of users, power companies generally build distribution network line monitoring systems for power outage management. However, most of the existing power outage prediction systems are based on static analysis models for power outage analysis. However, with the continuous expansion of distribution network lines and the emergence of various new types of power equipment, the characteristics of power loads and data sources have become more and more complex. Simple static analysis models can no longer accurately describe the characteristics of the power system.

[0004] In order to solve the above problems, the present invention proposes a distribution network line frequent power outage analysis and root cause identification system based on intelligent analysis. Summary of the invention

[0005] The purpose of the present invention is to propose a system for frequently power outages in distribution network lines based on intelligent analysis and root cause identification to solve the problems raised in the background technology:

[0006] Most existing power outage prediction systems perform power outage analysis based on static analysis models. However, with the continuous expansion of distribution network lines and the emergence of various new types of power-consuming equipment, the power characteristics and data sources of the power system have become more and more complex. Simple static analysis models can no longer accurately describe the characteristics of the power system.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] The frequent power outage analysis and root cause identification system of distribution network lines based on intelligent analysis includes: acquisition and processing module: used to collect and pre-process the operation data of distribution network lines based on data acquisition equipment;

[0009] Prediction and Identification Module: used to establish an intelligent root cause identification algorithm based on the collected data and analyze the fault type and cause of the power outage event;

[0010] Decision optimization module: used to provide real-time decision support by combining the root cause identification results of the prediction and identification module;

[0011] Visualization module: used to provide users with a visual interface;

[0012] The prediction and recognition module comprises:

[0013] Root cause analysis unit: used to learn and identify the patterns and trends of frequent power outages in distribution lines based on the optimized LSTM network, and combined with the graph neural network to analyze the root causes of power outages;

[0014] Model training unit: Use unlabeled data to perform self-supervised pre-training on each model, and then perform supervised optimization and adjustment of the model;

[0015] Power outage prediction unit: used to predict potential power outage events and causes of distribution network lines using the Seq2Seq model framework;

[0016] Power outage warning unit: used to issue a power outage warning according to the prediction result of the power outage prediction unit.

[0017] Preferably, the preprocessing operations of the acquisition and processing module include data cleaning, data formatting and time series processing operations.

[0018] Preferably, the optimized LSTM network in the root cause analysis unit respectively builds a short-term optimized LSTM model, a medium-term optimized LSTM model and a long-term optimized LSTM model to process features of different time scales, and introduces an attention mechanism for output after the optimized LSTM network outputs, and then combines the output value with the output value of the graph neural network, and performs the final root cause classification prediction through the fully connected layer.

[0019] Preferably, the optimized LSTM model adopts a nested network, optimizes the memory unit into an LSTM network, and couples the forget gate, the input gate and the memory unit, as follows:

[0020] The input gate is as follows:

[0021] I t =s ( oh I · [ h t-1 , x t ]+ b I )

[0022] in, I t is the activation value of the input gate; s is the activation function; oh I is the input gate weight; h t-1 is the output of the memory unit at the previous moment; x t for t The input of the moment memory unit; b I is the bias of the input gate;

[0023] The forget gate is as follows:

[0024] F t =s ( oh F · [ C t-1 , h t-1 , x t ]+ b F )

[0025] in, F t is the activation value of the forget gate; oh F is the forget gate weight; C t-1 is the state of the memory unit at the previous moment; b F is the bias of the forget gate;

[0026] The output gate is as follows:

[0027] O t =s ( oh O · [ C t-1 , h t-1 , x t ]+ b O )

[0028] in, O t is the activation value of the output gate; oh O is the forget gate weight; b O is the bias of the forget gate;

[0029] The memory units are as follows:

[0030] C t =F t ·C t-1 + I t ·s ( oh C · [ h t-1 , x t ]+ b C )

[0031] in, C t for t The state of the memory unit at each moment; oh C is the memory unit weight; b C is the bias of the memory unit;

[0032] Among them, each weight in the optimized LSTM network is obtained by optimizing based on the optimized particle swarm algorithm, and the optimized particle swarm algorithm is specifically as follows:

[0033] u i,d,j+1 =αu i,d,j +s 1 r 1( p i,d -v i,d,j )+ s 2 r 2( g i,d -v i,d,j )

[0034]

[0035] v i,d,j+1 =v i,d,j +u i,d,j+1

[0036] in, u i,d,j+1 For the i The particle in d Vidi j+The position change rate of 1 iteration; α is the inertia factor; r 1. r 2 is a random number between (0,1); p i,d For the i Particles v i The individual optimal value of g i,d is the global optimal value of the particle swarm; v i,d,j For the i The particle in d Vidi j The position of the iteration; v i,d,j+1 For the i The particle in d Vidi j+ The position of 1 iteration; s 1. s 2 is the acceleration factor.

[0037] Preferably, the attention mechanism introduced in the root cause analysis unit is specifically as follows;

[0038] Output of the optimized LSTM network y , using the trained attention weights l The weighting is as follows:

[0039] Y=λy =( oh λ ·y + b λ )

[0040] in, Y is the attention weight output; oh λ is the attention mechanism weight; b λ is the bias of the attention mechanism.

[0041] Preferably, the graph neural network is combined with GIS to obtain the distribution network line topology information.

[0042] Preferably, the Seq2Seq model in the power outage prediction unit uses an optimized LSTM network as encoder and decoder components, and the model framework is specifically as follows:

[0043] Input layer: Based on the optimized particle swarm algorithm, the length of the input sequence is dynamically adjusted according to the input power outage event related data as the input of the Seq2Seq model;

[0044] Encoding layer: The input power outage event-related data sequence is analyzed by the LSTM encoding layer, and the input power outage event-related data information is encoded into background variables;

[0045] Decoding layer: decodes the background variables output by the encoding layer through the LSTM decoding layer and generates an output sequence;

[0046] Output layer: Each time step of the output sequence corresponds to a Dense layer.

[0047] Preferably, the decision optimization module recommends maintenance strategies, optimizes equipment configuration, and formulates preventive maintenance plans based on the analysis results of the prediction and recognition module.

[0048] Compared with the prior art, the present invention provides a system for judging and identifying the root cause of frequent power outages in distribution network lines based on intelligent analysis, which has the following beneficial effects:

[0049] The present invention introduces an intelligent root cause identification algorithm for power outage events, conducts comprehensive modeling and analysis of multiple data from different data sources, combines the attention mechanism to increase the focus on key data, and combines the graph neural network to capture the topological structure and mutual influence between devices. It can not only detect the location of the fault, but also classify the type and cause of the fault, thereby helping operation and maintenance personnel to quickly locate and solve the problem; at the same time, combining the results of data analysis and root cause identification, the system provides comprehensive decision support; it realizes the ability to monitor the distribution network system in real time, and can immediately issue an alarm when an abnormality or potential power outage risk occurs. By accurately analyzing and identifying power outage events, it helps to improve the efficiency and reliability of power grid management and reduce the impact of frequent power outages on social life and production. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a system block diagram mentioned in Embodiment 1 of the present invention;

[0051] Figure 2 This is a schematic diagram of the optimized LSTM model mentioned in Example 1 of the present invention;

[0052] Figure 3 This is a schematic diagram of the network Seq2Seq model framework mentioned in Example 1 of the present invention.

[0053] Meaning of the symbols in the figure:

[0054] 100, acquisition and processing module; 200, prediction and identification module; 210, root cause analysis unit; 220, model training unit; 230, power outage prediction unit; 240, power outage warning unit; 300, decision optimization module; 400, visualization module. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0056] The present invention introduces an intelligent root cause identification algorithm for power outage events, conducts comprehensive modeling and analysis of multiple data from different data sources, combines the attention mechanism to increase the focus on key data, and combines the graph neural network to capture the topological structure and mutual influence between devices. It can not only detect the location of the fault, but also classify the type and cause of the fault, thereby helping operation and maintenance personnel to quickly locate and solve the problem; at the same time, combining the results of data analysis and root cause identification, the system provides comprehensive decision support; it realizes the ability to monitor the distribution network system in real time, and can immediately issue an alarm when an abnormality or potential power outage risk occurs. By accurately analyzing and identifying power outage events, it helps to improve the efficiency and reliability of power grid management and reduce the impact of frequent power outages on social life and production. Specifically include the following contents.

[0057] Embodiment 1:

[0058] See also Figure 1-3 The present invention provides a system for analyzing and identifying the root cause of frequent power outages in distribution network lines based on intelligent analysis, comprising:

[0059] Collection and processing module 100: used to collect historical data and real-time data from the distribution network system and perform preprocessing. This includes data cleaning, data formatting and time series processing to ensure data quality and availability. Among them, historical data includes past power outage event records, equipment maintenance records, meteorological data, power load data, etc. Such as equipment model, installation date, maintenance records, fault history, temperature, humidity, wind speed, rainfall, lightning strikes, floods, earthquakes, switch operation records, protection device action records, load data, voltage and current monitoring data, etc.; real-time data is real-time operating status data obtained by sensors and monitoring systems, such as real-time voltage, real-time current, real-time temperature, real-time humidity, etc. Data preprocessing operations include:

[0060] Cleaning: Process missing values, outliers, and noisy data to ensure data quality.

[0061] Feature engineering: Extract effective features, such as current fluctuations, equipment aging, meteorological conditions, etc., to improve the model's predictive ability.

[0062] Feature fusion: Fuse data from different sources to form a comprehensive feature vector for subsequent analysis.

[0063] Normalization: Standardize or normalize the data to suit different machine learning algorithms.

[0064] Prediction and identification module 200: used to establish an intelligent root cause identification algorithm based on known power outage events, and analyze the fault type and cause of the power outage event;

[0065] The data of known power outage events can be found in the following table:

[0066] Table 1: Number and causes of power outages in different regions

[0067]

[0068] The prediction and recognition module 200 also includes:

[0069] Root cause analysis unit 210: The LSTM network used for optimization builds short-term optimization LSTM model, medium-term optimization LSTM model and long-term optimization LSTM model to process features of different time scales; refer to Figure 2 , the optimized LSTM model adopts a nested network and optimizes the memory unit to an LSTM network, which can improve the modeling ability of the neural network, speed up the training speed, and reduce the risk of overfitting; and the forget gate, input gate and memory unit are coupled to more fully select the historical information to be retained and forgotten, and improve the recognition efficiency of the model; and the attention mechanism is introduced after the output of the optimized LSTM network for output. The attention mechanism helps the model focus on important time steps and can improve the recognition ability of key events; the output value is then combined with the data output by the graph neural network, and the final root cause classification prediction is performed through the fully connected layer; among them, the graph neural network is combined with GIS to obtain the topological structure information of the distribution network line; the use of the graph neural network to process the topological structure information of the power grid, combined with the LSTM network, can provide a more comprehensive analysis of the causes of power outages.

[0070] Model training unit 220: Use unlabeled data to perform self-supervised pre-training on each model, and then perform supervised optimization adjustment of the model.

[0071] The power outage prediction unit 230 uses the Seq2Seq model framework to predict potential power outage events and causes of distribution network lines; Figure 3 ,The Seq2Seq model framework combined with the above optimized LSTM network can effectively consider the ,connections between the output time series sequences and improve the ,accuracy of prediction for power outages.

[0072] The intelligent root cause identification algorithm built by combining LSTM, attention mechanism, graph neural network and Seq2Seq model can be used to identify the root cause of power outages. This multi-level, multi-model combination method can not only capture the complex dependencies in the time series, but also use the information of the power grid topology to provide more comprehensive and accurate analysis and prediction of power outage causes.

[0073] Based on the prediction and identification module 200, power outage prediction and root cause analysis are performed for 6 hours in a day, 6 days in a month, and 6 months in a year. The specific results can be seen in Table 2:

[0074] Table 2: Power outage event prediction results

[0075]

[0076] The power outage warning unit 240 issues a power outage warning based on the prediction result of the power outage prediction unit 230 so that relevant personnel can take maintenance measures in advance.

[0077] Decision optimization module 300: It is used to combine the results of intelligent analysis and root cause identification to provide real-time decision support for operation and maintenance personnel and managers. It can recommend maintenance strategies, optimize equipment configuration, and formulate preventive maintenance plans to reduce the occurrence of future power outages.

[0078] For example: During periods of high frequency of power outages: strengthen patrol inspections, strictly prohibit the use of high-power electrical appliances; strengthen the inspection of line heaters; strengthen the inspection of switches in use.

[0079] For areas prone to power outages: conduct insulation tests on equipment and apply for scrapping of unqualified equipment; strengthen inspection and maintenance of lines and switches in high-risk areas; strengthen inspection of equipment maintenance records and supervise regular maintenance.

[0080] For short circuits in electrical equipment: Check the electrical equipment in each area of ​​the distribution network, focus on monitoring obsolete equipment that has exceeded the reference service life, and gradually replace equipment with potential safety hazards; readjust areas where equipment wiring is unreasonable; monitor and replace electrical appliances with excessive loads; and conduct on-site verification of the current carrying capacity of high-power equipment.

[0081] Incomplete power outage records: Improve power outage records through training of engineering personnel, including improving records of power outage duration of distribution network lines, power outage fault locations, etc.

[0082] Visualization module 400: used to provide an intuitive user interface and visualization tools, so that operation and maintenance personnel can easily view system status, reports and analysis results. This helps real-time decision-making and team collaboration.

[0083] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A system for identifying frequent power outages in distribution networks and their root causes based on intelligent analysis. It is characterized in that include: A collection and processing module (100): used to collect and pre-process the operation data of the distribution network line based on the data collection equipment; Prediction and identification module (200): used to establish an intelligent root cause identification algorithm based on the collected data, and analyze the fault type and cause of the power outage event; A decision optimization module (300): used to provide real-time decision support in combination with the root cause identification result of the prediction and identification module (200); Visualization module (400): used to provide a visualization interface for users; The prediction and recognition module (200) comprises: A root cause analysis unit (210): used to learn and identify the pattern and trend of frequent power outages in distribution network lines based on the optimized LSTM network, and combine with the graph neural network to analyze the root cause of the power outage event; Model training unit (220): using unlabeled data to perform self-supervised pre-training on each model, and then performing supervised optimization adjustment of the model; A power outage prediction unit (230): used to predict potential power outage events and causes of distribution network lines using a Seq2Seq model framework; A power outage warning unit (240): configured to provide a power outage warning according to the prediction result of the power outage prediction unit (230); The optimized LSTM network in the root cause analysis unit (210) respectively builds a short-term optimized LSTM model, a medium-term optimized LSTM model and a long-term optimized LSTM model to process features of different time scales, and introduces an attention mechanism to output after the output of the optimized LSTM network, and then combines the output value with the output value of the graph neural network, and performs the final root cause classification prediction through a fully connected layer; The optimized LSTM model adopts a nested network, optimizes the memory unit into an LSTM network, and couples the forget gate, input gate and memory unit, as follows: The input gate is as follows: I t =σ ( ω I · [ h t-1 , x t ]+ b I ) in, I t is the activation value of the input gate; σ is the activation function; ω I is the input gate weight; h t-1 is the output of the memory unit at the previous moment; x t for t The input of the moment memory unit; b I is the bias of the input gate; The forget gate is as follows: F t =σ ( ω F · [ C t-1 , h t-1 , x t ]+ b F ) in, F t is the activation value of the forget gate; ω F is the forget gate weight; C t-1 is the state of the memory unit at the previous moment; b F is the bias of the forget gate; The output gate is as follows: O t =σ ( ω O · [ C t-1 , h t-1 , x t ]+ b O ) in, O t is the activation value of the output gate; ω O is the forget gate weight; b O is the bias of the forget gate; The memory units are as follows: C t =F t ·C t-1 + I t ·σ ( ω C · [ h t-1 , x t ]+ b C ) in, C t for t The state of the memory unit at each moment; ω C is the memory unit weight; b C is the bias of the memory unit; Among them, each weight in the optimized LSTM network is obtained by optimizing based on the optimized particle swarm algorithm, and the optimized particle swarm algorithm is specifically as follows: u i,d,j+1 =αu i,d,j +s 1 r 1 ( p i,d -v i,d,j )+ s 2 r 2 ( g i,d -v i,d,j ) v i,d,j+1 =v i,d,j +u i,d,j+1 in, u i,d,j+1 For the i The particle in d Vidi j+ The position change rate of 1 iteration; α is the inertia factor; r 1 , r 2 is a random number between (0,1); p i,d For the i Particles v i The individual optimal value of g i,d is the global optimal value of the particle swarm; v i,d,j For the i The particle in d Vidi j The position of the iteration; v i,d,j+1 For the i The particle in d Vidi j+ The position of 1 iteration; s 1 , s 2 is the acceleration factor.

2. According to the intelligent analysis-based distribution network line frequent power outage analysis and root cause identification system of claim 1, It is characterized in that The pre-processing operation of the acquisition processing module (100) includes data cleaning, data formatting and time series processing operations.

3. According to the system for frequent power outages and root cause identification of distribution network lines based on intelligent analysis in claim 1, It is characterized in that The attention mechanism introduced in the root cause analysis unit (210) is specifically as follows; Output of the optimized LSTM network y , using the trained attention weights λ The weighting is as follows: Y=λy =( ω λ ·y + b λ ) in, Y is the attention weight output; ω λ is the attention mechanism weight; b λ is the bias of the attention mechanism.

4. The system for determining and identifying the root cause of frequent power outages in distribution network lines based on intelligent analysis according to claim 3, It is characterized in that The graph neural network is combined with GIS to obtain the distribution network line topology information.

5. According to the system for frequent power outages and root cause identification of distribution network lines based on intelligent analysis in claim 1, It is characterized in that The Seq2Seq model in the power outage prediction unit (230) uses an optimized LSTM network as an encoder and decoder component, and the model framework is specifically as follows: Input layer: Based on the optimized particle swarm algorithm, the length of the input sequence is dynamically adjusted according to the input power outage event related data as the input of the Seq2Seq model; Encoding layer: The input power outage event-related data sequence is analyzed by the LSTM encoding layer, and the input power outage event-related data information is encoded into background variables; Decoding layer: decodes the background variables output by the encoding layer through the LSTM decoding layer and generates an output sequence; Output layer: Each time step of the output sequence corresponds to a Dense layer.

6. The system for determining and identifying the root cause of frequent power outages in distribution network lines based on intelligent analysis according to claim 1, It is characterized in that The decision optimization module (300) recommends maintenance strategies, optimizes equipment configuration, and formulates a preventive maintenance plan based on the analysis results of the prediction and recognition module (200).

Citation Information

Patent Citations

  • Substation distribution network line fault analysis and prediction method and system

    CN118501608A

  • KR20230038352A

Cited By

  • Zone area frequent power failure early warning method

    CN122414783A