CN109726200B - Power grid information system fault positioning system and method based on bidirectional deep neural network - Google Patents

Abstract

The invention discloses a power grid information system fault location system and method based on a bidirectional deep neural network, including a fault monitoring module, an inference engine, a database, and a deep learning module; the forward and reverse training of the deep neural network is performed on the fault tree analysis results to obtain The corresponding relationship between fault features and fault locations is stored in the expert knowledge base, and the grid information system fault monitoring is carried out. The fault features obtained according to the monitoring data are written into the feature database, and the fault location results are deduced according to the expert knowledge base and fault features. . The present invention can effectively solve the problems of difficult, slow and inaccurate fault location in the power grid information system, so that the system can send an alarm at the first time a fault occurs, and combined with the fault self-processing system, it can realize fault intelligent positioning and intelligent fault detection of the information system. The processing function can further improve the operation and maintenance efficiency of the power grid information system and ensure the safe and stable operation of the information system.

Description

Fault location system and method of power grid information system based on bidirectional deep neural network

technical field

The invention belongs to the technical field of information system fault diagnosis, and in particular relates to a power grid information system fault location system and method based on a bidirectional deep neural network.

Background technique

With the development of science and technology, information systems are more and more widely used in enterprises. At present, the number of first-level deployment application systems such as coordination office, unified authority, and ERP of State Grid Corporation has reached 76 sets, and the number of second-level deployment of core systems such as marketing, finance, and production has increased year by year. These information systems provide guarantee for the production, operation and management of the power grid. Once the information system fails, it will have a catastrophic impact on the business of the power grid.

The power grid information system has the characteristics of high complexity, strong real-time performance, high dynamics, and high security requirements. At the same time, the old system machines and gradually declining performance lead to an increase in the failure rate of the information system year by year, a late fault discovery time, and a long fault recovery period. Reliability issues are becoming increasingly prominent. Although fault diagnosis has entered the stage of automation, it still relies on the experience of operation and maintenance and experts, and lacks the control of intelligent brains. Therefore, how to use new technologies to create intelligent fault diagnosis of power grid information system is particularly important, and fault location is the basis of fault diagnosis.

With the development of fault location technology, there are many methods to solve the problem of location. There are software fault location methods based on data link, fault location methods based on configuration management database combined with fault tree, fault location technology based on immune algorithm, and fault location method based on BP neural network and fault tree.

A fault location method based on BP neural network and fault tree is proposed for transformer faults. This method first collects and sorts out transformer fault information as training and identification samples, establishes a transformer fault diagnosis model based on BP neural network, and then uses the fault tree analysis method. Classify the transformer fault level and severity. However, the power grid information system is more complex than the transformer, and the possible location of the fault is variable and difficult to eliminate. Therefore, it is necessary to invent a new method for accurate fault location according to the fault characteristics of the power grid information system.

The existing technical solutions are not designed specifically for the fault location of the power grid information system, and most of them are aimed at a specific mechanical or simple system. The current fault location guarantee of the power grid information system is mainly manual and lacks automatic means. The current operation guarantee system is oriented to ensure that the system does not go wrong. It mainly relies on people for overall management and control. It takes a lot of manpower to follow up and find faults, and then deal with them afterwards. There is no effective means for automatic fault location by information systems. As the "nerve" of State Grid Corporation, the power grid information system plays an increasingly important role, and the reliability requirements of the information system are also increasing day by day. The existing technical solutions cannot meet the reliability requirements of the power grid information system, nor can it solve the intelligent requirements of the fault location and processing of the information system.

At the same time, the traditional BP neural network is a local search optimization method, which has local minimization problems, slow convergence speed, different choices of BP neural network structure, contradictions between application examples and network scale, BP neural network prediction ability and The contradictory problem of training ability and the sample dependence problem of BP neural network.

Contents of the invention

Purpose of the invention: In view of the above problems, the present invention proposes a fault location system and method for power grid information systems based on a bidirectional deep neural network, which monitors and analyzes the power grid information system, thereby realizing accurate fault location for the entire process of the power information system.

Technical solution: In order to achieve the purpose of the present invention, the technical solution adopted in the present invention is: a fault location system for power grid information systems based on a two-way deep neural network, including a fault monitoring module, an inference engine, a database, and a deep learning module; The database includes an expert knowledge base, a feature database and a location database, the expert knowledge base stores the correspondence between fault features and fault locations, the location database stores the fault locations obtained from fault tree analysis of historical faults, and the feature database stores The fault characteristics obtained by the fault tree analysis of historical faults; the deep learning module carries out forward and reverse training of deep neural network to the fault location and fault characteristics obtained by the fault tree analysis of historical faults, and obtains the corresponding relationship between fault characteristics and fault locations , and stored in the expert knowledge base; the fault monitoring module is used for real-time fault monitoring of the power grid information system to obtain fault characteristics of information system faults; Based on the relationship and fault characteristics of information system faults, the fault location results can be obtained by reasoning.

Further, the fault location system also includes a man-machine interface module, which timely feeds back the fault location result to the operation and maintenance personnel, and sends out a fault alarm signal.

A method for locating a fault in a power grid information system based on a bidirectional deep neural network, comprising the steps of:

(1) Carry out the fault tree analysis of the historical accidents of the power grid information system, store the fault location obtained by the fault tree analysis in the location database, and store the fault characteristics in the feature database;

(2) Perform forward and reverse training of the deep neural network on the fault tree analysis results to obtain the corresponding relationship between fault features and fault locations, and store them in the expert knowledge base;

(3) Monitor the faults of the power grid information system to obtain the fault characteristics of the faults of the information system;

(4) According to the corresponding relationship between the fault features and fault locations stored in the expert knowledge base and the fault features of the information system faults, the fault location result is deduced.

Further, step 5 is also included, through the constructed system man-machine interface of the fault location result, the fault location result is fed back to the operation and maintenance personnel in time, and a fault alarm signal is issued.

Further, said step 1 includes:

(1.1) According to the composition of the power grid information system and the relationship between each part, a system architecture diagram is given;

(1.2) Investigate the historical failures and causes of the system;

(1.3) Find out accidents with serious consequences and relatively easy occurrences from historical faults as top events;

(1.4) From the top event, find out the direct cause events step by step, and draw the fault tree according to its logical relationship;

(1.5) According to the fault tree analysis results, a database based on historical data is constructed, and the fault location obtained from the fault tree analysis is stored in the location database, and the fault characteristics are stored in the feature database.

Further, said step 2 includes:

(2.1) Extract fault features and location information according to fault tree analysis results;

(2.2) Carry out forward propagation training of deep neural network from fault feature to fault position;

(2.3) Carry out deep neural network backpropagation training from fault position to fault feature;

(2.4) Store the training results, the correspondence between fault features and fault locations in the expert database.

Beneficial effects: the present invention can effectively solve the problems of difficult, slow, and inaccurate fault location in the power grid information system, enabling the system to issue an alarm at the first time a fault occurs, and combined with the fault self-processing system, the fault intelligence of the information system can be realized The positioning and intelligent processing functions can further improve the operation and maintenance efficiency of the power grid information system and ensure the safe and stable operation of the information system.

Description of drawings

Figure 1 is a schematic diagram of a fault location system for a power grid information system based on a bidirectional deep neural network.

Detailed ways

The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

As shown in Figure 1, the grid information system fault location system based on bidirectional deep neural network (DNN) and fault tree according to the present invention includes a fault monitoring module, an inference engine, a database, a deep learning module and a man-machine interface module.

The fault monitoring module is used to monitor the real-time status of each component of the power grid information system. Through the system monitoring software deployed at the hardware layer, software layer, network layer, and application layer, it completes multi-granularity cross-layer joint perception of the information system and provides real-time monitoring of the system. Operating data.

The reasoning machine, based on the real-time monitoring data combined with the rules of the expert knowledge base, repeatedly infers and draws conclusions. It is the brain that realizes fault location and determines the accuracy and agility of fault location. Reasoning is the process of obtaining results from collected data analysis according to established rules. The database and reasoning machine are the two cores of fault location. The reasoning machine of the present invention analyzes the fault characteristics and fault locations of the grid information system according to the fault tree analysis rules.

Database, including expert knowledge base, feature database, and location database. The expert database is used to store operation and maintenance experience and fault location rules trained by DNN, that is, the correspondence between fault features and fault locations; the feature database is used to store power grid information systems. According to the fault characteristics obtained by the fault tree analysis and the fault characteristics of the information system obtained by the fault monitoring module, the location database is used to store the fault location points obtained by the power grid information system according to the fault tree analysis.

Deep learning module, including forward propagation training and back propagation training of information system fault characteristics and location data, applying fault tree analysis and bidirectional deep neural network (DNN), analyzing and self-learning the characteristic data given by the monitoring system , give the location of the fault and determine the source of the fault.

The man-machine interface module, including the positioning result and fault alarm, is used to inform the operation and maintenance personnel of the fault positioning result and send out the fault alarm signal in time. The conclusion drawn by combining the reasoning machine with the results of DNN positive direction training is fed back to the operation and maintenance personnel through the man-machine interface, and the fault location is clarified through fault location and fault alarm.

As shown in Figure 1, the fault location method of power grid information system based on bidirectional deep neural network and fault tree includes the following steps:

Step 1: Carry out fault tree analysis of historical accidents in the power grid information system, store the fault location obtained from the fault tree analysis in the location database, and store the fault characteristics in the feature database; and build an expert knowledge base based on historical data according to the fault tree analysis results;

S11. Familiar with the power grid information system, give a system architecture diagram according to the system composition and the relationship between each part;

S12. Investigate the historical faults and causes of the system, collect accident cases, conduct accident statistics, imagine possible accidents for a given system, and investigate all causal events and various factors related to accidents;

S13. Determine the top event: the object to be analyzed is the top event, conduct a comprehensive analysis of the investigated accidents, and find out the accidents with serious consequences and relatively easy occurrences as the top events;

S14. Draw a fault tree: from the event on the top, find out the events of the direct cause step by step until the depth to be analyzed, and draw a fault tree according to its logical relationship;

S15. According to the fault tree analysis result, a database based on historical data is constructed.

Step 2, grid information system fault monitoring, and writing the monitoring data into the feature database;

S21. The monitoring system is responsible for the collection of various probe data, and generates an alarm according to the monitoring rules;

S22. Writing the monitoring data into the feature database;

S23. Summarize and format the alarm information according to the rules.

Step 3, carry out the forward and reverse training of the deep neural network of fault characteristics and location information, and output the training results to the fault reasoning machine;

S31. Extract feature and location information from the grid information system fault feature database and fault location database;

S32. Carry out forward propagation training of the deep neural network from the fault feature to the fault location, and use the output of the previous layer to calculate the output of the next layer, which is the forward propagation algorithm of the deep neural network;

S33. Perform deep neural network backpropagation training from the fault location to the fault feature. Before performing the deep neural network backpropagation algorithm, it is necessary to select a loss function to measure the difference between the output calculated by the training sample and the real training sample output. The loss, the present invention chooses to use the sigmoid activation function and the cross-entropy loss function. In DNN, the most common process of optimizing the extreme value of the loss function is to iteratively solve it through the gradient descent method. The process of iteratively optimizing the loss function of the DNN with the gradient descent method to find the minimum value is the backpropagation algorithm.

S34. Output the training result to the fault inference engine.

Step 4, realize the fault inference engine, and analyze the fault characteristics and fault location of the power grid information system;

S41. Obtain fault location reasoning rules according to the historical data of the power grid information system and fault tree analysis;

S42. Combining the expert knowledge base and the fault features, a fault location result is obtained through reasoning.

Step five, build the system man-machine interface of fault location features and results.

Build a man-machine interface to assist in the visualization of fault location results. The fault location results can be displayed directly on the client terminal, and relevant responsible persons can also be notified in a timely manner through SMS or other methods.

The invention can effectively solve the problems of difficult, slow and inaccurate fault location in the power grid information system, so that the system can send an alarm at the first time when a fault occurs, and combined with the fault self-processing system, it can realize fault intelligent positioning and intelligent fault detection of the information system. The processing function can further improve the operation and maintenance efficiency of the power grid information system and ensure the safe and stable operation of the information system. On the basis of manual troubleshooting, the quality and efficiency of information system fault diagnosis can be improved to a higher level, which can reduce the workload of operation and maintenance personnel and realize intelligent fault location and diagnosis of power grid information system. At the same time, the application of this method accelerates the pace of scientific and technological innovation of State Grid Corporation, improves the technological level and management level of information construction of power grid enterprises, improves the information service level of power grid enterprises, and further enhances the corporate image of power grid companies.

Claims (4)

1. The power grid information system fault positioning system based on the bidirectional deep neural network is characterized by comprising a fault monitoring module, an inference engine, a database and a deep learning module;

the database comprises an expert knowledge base, a characteristic database and a position database, wherein the expert knowledge base stores the corresponding relation between fault characteristics and fault positions, the position database stores fault positions obtained by fault tree analysis of historical faults, and the characteristic database stores fault characteristics obtained by fault tree analysis of historical faults;

the deep learning module performs forward and reverse training of the deep neural network on the fault position and the fault characteristic obtained by analyzing the fault tree of the historical fault to obtain a corresponding relation between the fault characteristic and the fault position;

the fault monitoring module is used for carrying out real-time fault monitoring on the power grid information system to obtain fault characteristics of the information system faults;

the inference engine infers and obtains a fault positioning result according to the corresponding relation between the fault characteristics and the fault positions and the fault characteristics of the information system faults;

the database comprises:

(1.1) providing a system architecture diagram according to the composition of the power grid information system and the relation among all parts;

(1.2) investigating system history faults and reasons;

(1.3) finding out accidents which have serious consequences and are more likely to occur from the historical faults as overhead events;

(1.4) from the top event, progressively finding out the event of the direct cause, and drawing out a fault tree according to the logic relation;

(1.5) constructing a database based on historical data according to the analysis result of the fault tree, storing the fault position obtained by the analysis of the fault tree in a position database, and storing the fault characteristics in a characteristic database;

the deep learning module includes:

(2.1) extracting fault characteristics and position information according to a fault tree analysis result;

(2.2) performing forward propagation training of the deep neural network from the fault feature to the fault location;

(2.3) performing deep neural network back propagation training from the fault location to the fault feature;

and (2.4) training to obtain the corresponding relation between the fault characteristics and the fault positions.

2. The system for positioning faults of a power grid information system based on a bidirectional deep neural network according to claim 1, further comprising a man-machine interface module, feeding back fault positioning results to operation and maintenance personnel in time and sending out fault alarm signals.

3. The fault positioning method of the power grid information system based on the bidirectional deep neural network is characterized by comprising the following steps of:

(1) Performing fault tree analysis of historical accidents of the power grid information system, storing fault positions obtained by the fault tree analysis in a position database, and storing fault characteristics in a characteristic database;

the step (1) comprises:

(1.1) providing a system architecture diagram according to the composition of the power grid information system and the relation among all parts;

(1.2) investigating system history faults and reasons;

(1.3) finding out accidents which have serious consequences and are more likely to occur from the historical faults as overhead events;

(1.4) from the top event, progressively finding out the event of the direct cause, and drawing out a fault tree according to the logic relation;

(1.5) constructing a database based on historical data according to the analysis result of the fault tree, storing the fault position obtained by the analysis of the fault tree in a position database, and storing the fault characteristics in a characteristic database;

(2) Training a bidirectional deep neural network on the analysis result of the fault tree to obtain a corresponding relation between fault characteristics and fault positions;

the step (2) comprises:

(2.1) extracting fault characteristics and position information according to a fault tree analysis result;

(2.2) performing forward propagation training of the deep neural network from the fault feature to the fault location;

(2.3) performing deep neural network back propagation training from the fault location to the fault feature;

training to obtain a corresponding relation between the fault characteristics and the fault positions;

(3) Performing power grid information system fault monitoring to obtain fault characteristics of the information system faults;

(4) According to the corresponding relation between the fault characteristics and the fault positions and the fault characteristics of the information system faults, a fault positioning result is obtained in a reasoning mode;

and (5) feeding back the fault positioning result to operation and maintenance personnel in time through a constructed system man-machine interface of the fault positioning result, and sending out a fault alarm signal.

4. A method for locating a fault in a grid information system based on a bi-directional deep neural network according to claim 3, wherein the correspondence between the fault signature and the fault location is stored in an expert knowledge base.