CN116522188A - Equipment abnormality identification and treatment method and system based on deep belief network - Google Patents

Abstract

The invention discloses a method and a system for identifying and disposing equipment anomalies based on a deep belief network. Information feature extraction is carried out on various equipment exception handling schemes, an equipment handling scheme rule base is constructed according to the logical relations of handling object information, handling behavior features and handling methods, and equipment exception handling scheme making efficiency is improved, so that equipment exception handling efficiency is improved, monitoring personnel monitoring and handling efficiency is improved, long-term sickness operation of equipment caused by missed monitoring signals is avoided, and safety operation level of a power grid is improved.

Description

Equipment abnormality identification and treatment method and system based on deep belief network

Technical Field

The invention relates to the technical field of power systems, in particular to a method and a system for identifying and disposing equipment abnormality based on a deep belief network.

Background

In the field of power system automation, signals are the main means for monitoring the running state of power equipment, are the main reference basis for monitoring personnel to judge equipment conditions and fault reasons and to formulate on-site disposal schemes, and with the development of transformer substation scale and data acquisition technology, the power monitoring system acquires monitoring signals to rapidly increase in scale, and particularly when a power grid is abnormal or equipment has faults, massive monitoring information flows into the monitoring system, so that great challenges are brought to power monitoring and abnormal disposal.

The traditional power monitoring means mainly has three problems, namely a power system monitoring signal is a discrete signal system, so that monitoring personnel are required to analyze the relationship among the signals to reflect equipment anomalies of different types and different granularities, and after large-scale monitoring signals are accessed, the monitoring personnel are subjected to daily monitoring to bring great challenges, the situation of false monitoring and leakage monitoring is easy to cause, and the equipment anomalies are not easy to be timely analyzed and found; secondly, when a power grid accident occurs, the monitoring system receives a large number of monitoring signals, the manual judgment mode cannot accurately judge the fault reason, and the fault recovery time is prolonged; thirdly, equipment types and models are numerous, various exception handling principles and schemes are more, and the efficiency of developing exception handling is low by relying on experience of monitoring personnel.

Disclosure of Invention

The present invention is directed to a method and a system for identifying and handling device anomalies based on a deep belief network, so as to solve the problems set forth in the background art.

In order to achieve the above purpose, the present invention provides the following technical solutions:

the equipment anomaly identification and treatment method based on the deep belief network comprises the following steps:

constructing a typical monitoring object library of the power equipment;

forming a monitoring signal data set by typical power equipment monitoring signals, constructing a word vector model-based improved deep belief network analysis model, classifying the power equipment monitoring signals acquired in real time through the deep belief network analysis model, and establishing an association relation between the monitoring signals and a typical monitoring object library;

constructing an abnormal study and judgment rule base of the power equipment based on the typical monitoring object;

monitoring a power equipment monitoring signal in real time, identifying an abnormal state of the power equipment according to the mapping relation between the monitoring signal and a typical monitoring object and an abnormal research and judgment rule of the power equipment, and generating an abnormal alarm event of the power equipment;

extracting information features of the equipment abnormality handling process, and constructing an equipment handling plan rule base according to the logical relations of the handling object information, the handling behavior features and the handling method;

after the power equipment abnormal event is monitored, the treatment rules corresponding to the typical signals are retrieved by combining the equipment type and the typical characteristic signals contained in the equipment abnormal event, invalid treatment rules are filtered out by utilizing a network reasoning method according to the association relation between the characteristic signals, and a treatment scheme is generated according to the treatment sequence.

Preferably, characteristic extraction is performed on typical monitoring information of the power equipment, and the typical monitoring information is normalized to operation characteristics according to the monitoring objects of the typical monitoring information and the reflected behavior characteristics thereof, so as to construct a typical monitoring object library of the power equipment.

Preferably, the operation characteristics comprise equipment out-of-limit, heavy overload, operation control, overhaul debugging and action resetting.

Preferably, the monitoring signals of the typical power equipment are screened according to the conditions of different types of substations, comprehensive manufacturers and equipment types to form a monitoring signal data set.

Preferably, the monitoring signal data set is subjected to word segmentation, key feature words are selected, and a word vector model is established based on a cyclic neural network; and fusing a deep Boltzmann machine model in the deep confidence network model, and combining the word vector model to obtain typical classification of the monitoring signals in a typical monitoring object library, so as to establish the association relation between the monitoring signals and the typical monitoring objects.

Preferably, the word vector model is used as a text representation, the 2-layer depth confidence network model with higher information reduction degree is used for carrying out primary dimension reduction on the text content of the monitoring signal to obtain a denoising and higher integrity result, the 3-layer depth Boltzmann machine model is used for extracting high-layer features, finally the text representation is obtained, and the typical classification of the monitoring signal is obtained.

Preferably, the modified TF-IDF algorithm screens out the words of the first 40 as key feature words of the word vector model.

Preferably, the relation between the typical power monitoring object and the equipment abnormality is summarized, and a power equipment abnormality research rule base is constructed by adopting a mode of combining a logic expression and power special abnormality research logic.

A deep belief network-based device anomaly recognition and handling system, comprising:

the signal knowledge base is used for solidifying and storing various information of the monitoring signals;

the reasoning rule base disassembles and solidifies and stores the logic relationship among the operation characteristics and the behavior characteristics of the monitoring signals and the logic relationship of fault abnormality;

the treatment rule base is used for establishing the association relation between the abnormal monitoring signals and the treatment plan;

the abnormal signal intelligent identification library loads monitoring signal identification rules through the signal knowledge library and the reasoning rule library;

the treatment auxiliary decision-making library automatically extracts the related measurement operation data before and after the signal generation according to the generation reason of the monitoring signal and generates a treatment plan according to the treatment rule library;

the abnormal signal monitoring and inquiring library is used for receiving and displaying the abnormal alarm signal in real time and providing a monitoring signal searching tool;

and the defect management module is used for automatically extracting defect information of the associated equipment according to the abnormal signal.

A computer readable storage medium having stored thereon a computer program which when executed by a processor implements a deep belief network based device anomaly identification and handling method.

Compared with the prior art, the invention has the beneficial effects that:

the invention provides a device abnormality identification and treatment method based on a deep belief network, and a device abnormality identification and treatment system based on the deep belief network is constructed by the method.

The method mainly solves the problem of accuracy of equipment abnormal alarm event synthesis, and by extracting and establishing a characteristic parameter library of monitoring signals of the electric equipment, introducing a deep belief network technology to design an efficient classifier for an electric signal system, accurately classifying and identifying the monitoring signals of the electric equipment, and improving the accuracy of classification of the monitoring signals, thereby improving the accuracy of equipment abnormal identification.

Secondly, extracting information features of various equipment exception handling schemes, constructing an equipment exception handling scheme rule base according to the logical relations of handling object information, handling behavior features and handling methods, and improving the equipment exception handling scheme making efficiency, thereby improving the equipment exception handling efficiency;

finally, the abnormal intelligent monitoring, the rapid signal retrieval, the intelligent treatment and the intelligent defect extraction of the equipment are realized, the monitoring and treatment efficiency of monitoring personnel is improved, the long-term sickness operation of the equipment caused by the missed monitoring signal is avoided, and the safe operation level of the power grid is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of an intelligent recognition and processing system for abnormal signals of the equipment of the present invention;

FIG. 2 is a flow chart of the intelligent recognition and processing of abnormal signals of the equipment of the invention;

FIG. 3 is a library of device treatment plan rules of the present invention;

fig. 4 is a schematic diagram of a monitoring signal processing measure case of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Because the construction time, the manufacturer of the transformer substation, the model of the adopted equipment and the management requirements are different, the naming standardization of the monitoring signals of the equipment in the transformer substation is uneven, and the naming standardization of the same signals of the same equipment is quite different, so that the abnormal reasoning and judging accuracy of the equipment is not high.

1. Characteristic extraction is carried out on typical monitoring information of different types of equipment, and according to monitoring objects of monitoring signals and behavior characteristics reflected by the monitoring objects, the monitoring information is normalized to running characteristics such as equipment out-of-limit, heavy overload, operation control, overhaul debugging, action resetting and the like, so that a typical monitoring object library of the power equipment is constructed;

2. and taking typical monitoring signals of substations of different construction comprehensive manufacturers and different types as data sets, performing word segmentation on the data sets, selecting the words of the front 40 as key feature words of a word vector model by using an improved TF-IDF algorithm, and establishing the word vector model by using a cyclic neural network (RNNLM) based on a language model.

3. And fusing a deep Boltzmann machine model (DBM) in a deep belief network model (DBN), taking a keyword vector as a text representation, selecting a 2-layer DBM model and a 3-layer DBN model for combination, performing primary dimension reduction on the text content of the monitoring signal by using the 2-layer DBM with higher information reduction degree to obtain a denoising and higher integrity result, extracting high-layer features by using the 3-layer DBN, finally obtaining the text representation, obtaining the typical classification of the monitoring signal, and establishing the association relation between the specific monitoring signal and the typical monitoring object.

Summarizing the relation between the typical power monitoring object and the equipment abnormality, and constructing a power equipment abnormality grinding rule base by adopting a mode of combining a logic expression and a power special abnormality grinding logic.

Monitoring a power equipment monitoring signal in real time, identifying an abnormal state of the power equipment according to the mapping relation between the monitoring signal and a typical monitoring object and an abnormal research and judgment rule of the power equipment, and generating an abnormal alarm event of the power equipment;

1. as shown in fig. 3, by combing various equipment exception handling steps and handling methods, extracting information features for equipment exception handling processes, and constructing an equipment handling plan rule base according to the logical relations of equipment types, component types, exception types, handling behavior features and handling methods;

2. after monitoring an abnormal event of the power equipment, retrieving a treatment rule corresponding to the typical signal by combining the equipment type and the typical characteristic signal contained in the abnormal event of the equipment, filtering out invalid treatment rules by using a network reasoning method according to the association relation between the characteristic signals, and generating a treatment scheme according to a treatment sequence;

as shown in fig. 4, the electrical equipment abnormal event is generated after comparing typical characteristics corresponding to the monitoring signal (such as what equipment, what component, what index, corresponding behavior characteristics, etc. are monitored by the monitoring signal) with abnormal grinding rule, and the abnormal treatment rule essentially establishes a corresponding relation with the treatment method for the characteristics; after an abnormal alarm event is monitored, firstly, possible measures are searched out according to typical characteristics of monitoring signals serving as a basis for research and judgment, and then a final treatment scheme is combined according to the sequence of occurrence of each characteristic signal.

Through the equipment exception identification and handling system of above-mentioned technique construction based on degree of depth confidence network, realize that equipment exception intelligent monitoring, signal retrieve fast, intelligent handling, defect intelligence draw, improve monitoring personnel's control and handling efficiency, avoid causing equipment long-term disease operation because of leaking the monitoring signal, improve electric wire netting safe operation level, satisfy the overall process demand of abnormal signal monitoring and handling, specifically include:

1. knowledge base management

Including a signal knowledge base, an inference rule base, and a disposition rule base.

The knowledge base system provides a monitoring signal knowledge base management function, various typical monitoring signals are defined according to daily monitoring requirements, and information such as object characteristics, meaning explanation, generation reasons, influences and results of the monitoring signals is solidified and stored according to certain rules to serve as a knowledge base of daily work such as monitoring signal analysis, abnormal handling support and training.

Inference rule base: summarizing the relation between the monitoring signals and fault anomalies, and disassembling, solidifying and storing the logic relation between the operation characteristics and the behavior characteristics of the monitoring signals and the logic relation of the fault anomalies, and taking the logic relation as a basic support for identifying the anomalies and intelligently extracting defects.

Treatment rule base: the method is used for constructing an electronic treatment plan library for equipment abnormality treatment, supporting modes such as manual programming, plan introduction and the like, summarizing and solidifying treatment processes after different equipment abnormalities, establishing an association relation between an abnormality monitoring signal and a treatment plan, and carrying out analysis and treatment according to plan rule reminding or automatic flow when equipment abnormalities occur, so as to provide basic support for accident abnormality emergency treatment and automatic operation and maintenance work order generation.

2. Anomaly signal monitoring and interrogation

A comprehensive event alert window: the method is responsible for receiving in real time and displaying the abnormal alarm signals according to the time sequence, and the displaying content comprises the following steps: the information such as the occurrence time, the event description, the event type, the confirmation state, the confirmation person, the confirmation time and the like supports the filtering query according to the conditions such as the operation maintenance class, the voltage level, the transformer substation, the confirmation state and the like. And the method supports quick retrieval of the typical monitoring signal corresponding to the signal through a right-key menu, and opens a monitoring signal detail interface to check signal basic information, paraphrasing, generating reasons, influences, consequences, disposal principles and the like of the typical signal.

3. Monitoring signal retrieval

The system provides a monitoring signal searching tool, a monitoring signal searching tool bar is opened through a shortcut button or a signal searching menu, a monitoring signal keyword is input into a searching frame, and information such as a monitoring signal definition, a generation reason, influence and result, a treatment principle and the like is rapidly searched for a target signal in a searching result list.

4. Abnormal signal intelligent identification library

The system loads monitoring signal identification rules through a monitoring signal knowledge base and an inference rule base, subscribes monitoring signal operation information to the centralized control system, analyzes the objectified characteristics and the behavior characteristics of the monitoring signals in real time, matches and pairs the monitoring signal identification rules, generates abnormal alarm events according to event generation rules aiming at signals conforming to the abnormal behavior characteristics of the signals, and pushes the abnormal alarm events to the comprehensive event alarm window.

5. Treatment aid decision library

The system automatically extracts relevant measurement operation data before and after the occurrence of the signal according to the generation reason of the monitoring signal, searches treatment content and flow defined in a treatment rule base to generate a treatment plan, instantiates according to actual equipment to generate a treatment task, assists monitoring personnel to rapidly perform abnormal treatment, and can support manual recording of a treatment result.

6. Defect management module

The defect extraction method and the defect detection device support the defect extraction on an alarm window (comprising a real-time alarm window and a comprehensive event alarm window) or a device interval diagram, automatically extract relevant telemetry and remote signaling data, identify defect related devices and relevant parts thereof, defect types and other information, generate defect records, and support manual modification and manual closed loop of defect information, wherein defects which are not closed loop are not repeatedly generated.

The following describes the steps of the intelligent recognition and processing flow of the abnormal signal of the equipment in detail with reference to the accompanying figures 1-2:

1) Monitoring signal cleaning process

(1) Loading power grid equipment and a measurement model from a model library;

(2) typical monitoring signals are screened according to conditions of different types of substations, comprehensive manufacturers, equipment types and the like to form a monitoring signal data set;

(3) and performing word segmentation processing on the monitoring signal data set, selecting key feature words, and establishing a word vector model based on a cyclic neural network (RNNLM).

(4) And improving a DBN+DBM model based on the word vector model to obtain typical classification of the monitoring signals, and establishing an association relation between the specific monitoring signals and typical monitoring objects.

2) Equipment abnormality identification process

(1) Loading the association between the power monitoring object and the power monitoring eventing calculation logic object, and establishing a memory mapping relation to form an analysis calculation model;

(2) starting a power special abnormality research judgment, equipment abnormal event synthesis and treatment scheme synthesis thread;

(3) subscribing real-time data to a real-time data center; subscribing real-time matters to a real-time matters service;

(4) starting a real-time data receiving thread, and circularly waiting for receiving real-time operation data;

a) Judging whether the real-time data need to participate in the special power abnormality judgment, if so, pushing the real-time data into a special power abnormality judgment data queue;

b) Otherwise, the real-time data is participated in judging whether the real-time data accords with the abnormal grinding judgment logic rule, if so, the real-time data is pushed to the monitoring object change queue.

(5) Starting a real-time item receiving thread, and circularly waiting for receiving real-time alarm items;

a) Judging whether the real-time alarm event participates in the special power abnormality judgment, if so, pushing the special power abnormality judgment change data queue;

b) Otherwise, the real-time data is participated in judging whether the real-time data accords with the abnormal grinding judgment logic rule, if so, the real-time data is pushed to the monitoring object change queue.

3) Special abnormality studying and judging process for electric power

(1) Reading the power-dedicated abnormality studying and judging configuration information to complete the instantiation of the power-dedicated abnormality logic calculation class;

(2) circularly detecting an abnormal research and judgment data change queue special for electric power;

(3) judging whether the alarm realization needs to be generated or not, if so, generating an alarm event and pushing a monitoring object change queue;

4) Equipment abnormality alarm event synthesis process

(1) Circularly detecting a change queue of the monitoring object;

(2) acquiring an alarm event list which corresponds to a monitoring object and needs to be synthesized;

(3) the monitoring object is not in any alarm event list, generates a new alarm event and pushes the alarm event list;

(4) judging whether the alarm event can be synthesized or not, if so, generating and pushing the alarm event;

(5) judging whether the alarm event is overtime, if yes, the alarm event is clear from an alarm event list;

5) A treatment scheme synthesis process, namely circularly detecting a monitoring object change queue;

(1) generating typical characteristics of the abnormal event according to the operation characteristics contained in the equipment abnormal event synthesis rule;

(2) monitoring and retrieving the treatment rules corresponding to the typical signals, filtering out invalid treatment rules by using a network reasoning method according to the association relation between the characteristic signals, and generating a treatment scheme according to the treatment sequence.

In the description of the present specification, the descriptions of the terms "one embodiment," "example," "specific example," and the like, mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

The preferred embodiments of the invention disclosed above are intended only to assist in the explanation of the invention. The preferred embodiments are not exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention. The invention is limited only by the claims and the full scope and equivalents thereof.

Claims (10)

1. The equipment abnormality identification and treatment method based on the deep belief network is characterized by comprising the following steps of: the method comprises the following steps:

constructing a typical monitoring object library of the power equipment;

forming a monitoring signal data set by typical power equipment monitoring signals, constructing a word vector model-based improved deep belief network analysis model, classifying the power equipment monitoring signals acquired in real time through the deep belief network analysis model, and establishing an association relation between the monitoring signals and a typical monitoring object library;

constructing an abnormal study and judgment rule base of the power equipment based on the typical monitoring object;

monitoring a power equipment monitoring signal in real time, identifying an abnormal state of the power equipment according to the mapping relation between the monitoring signal and a typical monitoring object and an abnormal research and judgment rule of the power equipment, and generating an abnormal alarm event of the power equipment;

extracting information features of the equipment abnormality handling process, and constructing an equipment handling plan rule base according to the logical relations of the handling object information, the handling behavior features and the handling method;

after the power equipment abnormal event is monitored, the treatment rules corresponding to the typical signals are retrieved by combining the equipment type and the typical characteristic signals contained in the equipment abnormal event, invalid treatment rules are filtered out by utilizing a network reasoning method according to the association relation between the characteristic signals, and a treatment scheme is generated according to the treatment sequence.

2. The method for identifying and handling device anomalies based on deep belief networks according to claim 1, wherein: and extracting characteristics of typical monitoring information of the power equipment, and normalizing the typical monitoring information into operation characteristics according to the monitoring objects of the typical monitoring information and the reflected behavior characteristics thereof, so as to construct a typical monitoring object library of the power equipment.

3. The method for identifying and handling device anomalies based on deep belief networks according to claim 2, wherein: the operation characteristics comprise equipment out-of-limit, heavy overload, operation control, overhaul and debugging and action resetting.

4. The method for identifying and handling device anomalies based on deep belief networks according to claim 1, wherein: typical power equipment monitoring signals are screened according to different types of substations, comprehensive vendors and equipment type conditions to form a monitoring signal data set.

5. The method for identifying and handling device anomalies based on deep belief networks according to claim 1, wherein: performing word segmentation on the monitoring signal data set, selecting key feature words, and establishing a word vector model based on a cyclic neural network; and fusing a deep Boltzmann machine model in the deep confidence network model, and combining the word vector model to obtain typical classification of the monitoring signals in a typical monitoring object library, so as to establish the association relation between the monitoring signals and the typical monitoring objects.

6. The method for identifying and handling device anomalies based on deep belief networks according to claim 5, wherein: and the word vector model is used as a text representation, the 2-layer deep belief network model with higher information reduction degree is used for carrying out primary dimension reduction on the text content of the monitoring signal to obtain a denoising and higher integrity result, the 3-layer deep Boltzmann machine model is used for extracting high-layer features, finally the text representation is obtained, and the typical classification of the monitoring signal is obtained.

7. The method for identifying and handling device anomalies based on deep belief networks according to claim 5, wherein: the modified TF-IDF algorithm screens out the words of the first 40 as key feature words of the word vector model.

8. The method for identifying and handling device anomalies based on deep belief networks according to claim 1, wherein: summarizing the relation between the typical power monitoring object and the equipment abnormality, and constructing a power equipment abnormality grinding rule base by adopting a mode of combining a logic expression and a power special abnormality grinding logic.

9. An equipment anomaly identification and handling system based on a deep belief network is characterized in that: comprising the following steps:

the signal knowledge base is used for solidifying and storing various information of the monitoring signals;

the reasoning rule base disassembles and solidifies and stores the logic relationship among the operation characteristics and the behavior characteristics of the monitoring signals and the logic relationship of fault abnormality;

the treatment rule base is used for establishing the association relation between the abnormal monitoring signals and the treatment plan;

the abnormal signal intelligent identification library loads monitoring signal identification rules through the signal knowledge library and the reasoning rule library;

the treatment auxiliary decision-making library automatically extracts the related measurement operation data before and after the signal generation according to the generation reason of the monitoring signal and generates a treatment plan according to the treatment rule library;

the abnormal signal monitoring and inquiring library is used for receiving and displaying the abnormal alarm signal in real time and providing a monitoring signal searching tool;

and the defect management module is used for automatically extracting defect information of the associated equipment according to the abnormal signal.

10. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements a deep belief network based device anomaly identification and handling method according to any one of claims 1 to 8.