CN-122022703-A - Intelligent two-ticket auxiliary decision-making method and system for electric power system integrating environment perception and operation risk analysis

CN122022703ACN 122022703 ACN122022703 ACN 122022703ACN-122022703-A

Abstract

The invention provides an intelligent two-ticket auxiliary decision-making method and system for an electric power system integrating environment perception and operation risk analysis, and relates to the technical field of electric power systems. The method comprises the steps of collecting environmental parameters such as temperature, humidity, wind speed, wind direction, gas concentration and the like on an operation site to form a time sequence feature matrix, analyzing an operation ticket and a work ticket, extracting information such as operation type, voltage level, personnel qualification and the like to generate the operation feature matrix, extracting the environment and the operation feature vector through a deep network and generating a joint feature vector based on attention mechanism fusion, inputting a wind control model to output risk probability and level, comparing the risk probability and the rule base, identifying non-conforming items and pushing optimization suggestions, monitoring the operation state in real time, automatically sending a pause instruction and an alarm when the pause threshold is reached, collecting feedback data after operation, screening high-value samples based on a sliding time window, and updating the model in an incremental or federal learning mode to realize continuous improvement of risk identification capability.

Inventors

HUANG CHENGWEI
LIAO LIZHEN
QIN YU
LIANG XINGJI
LU YIFAN
LAN SONGTAO
LIU XINGYAN
HE YOUWEN
ZHONG RUIBIN
He Zhuxiu
JIANG XIGUI
WEN XIANTAO

Assignees

广西电网有限责任公司贺州供电局

Dates

Publication Date: 20260512
Application Date: 20250818

Claims (10)

1. An intelligent two-ticket auxiliary decision-making method of an electric power system integrating environment perception and operation risk analysis is characterized by comprising the following steps: An environment sensing unit is arranged on an operation site, temperature, humidity, wind speed, wind direction, rainfall, combustible gas concentration, infrared thermal image, noise and illumination parameters are periodically collected, and a time sequence environment characteristic matrix is formed after the parameters are converged through an edge gateway; Analyzing an operation ticket to be executed and a work ticket, extracting the operation type, the voltage level, the target equipment, the operator qualification, the safety measure item, the planning period and the tool information, and generating an operation feature matrix by unified vectorization coding; deep feature extraction is respectively carried out on the environment feature matrix and the operation feature matrix to obtain an environment feature vector and an operation feature vector, and correlation calculation and fusion are carried out on the environment feature vector and the operation feature vector based on an attention mechanism to generate a joint feature vector; inputting the combined feature vector into a wind control prediction network, outputting risk probability, and dividing risk grades according to a preset threshold value; carrying out association reasoning on the risk probability and the risk level with a safety rule base, determining non-conforming items of safety measure deficiency, non-conforming operation conditions, abnormal equipment state and human factor risk overrun, generating bill optimization suggestions aiming at the non-conforming items, and pushing the bill optimization suggestions to operation responsible persons, guardianship persons and approvers through at least one mode of a mobile terminal and a monitoring platform; Monitoring environmental parameters and job execution states in real time during the implementation of the job, and automatically triggering a pause instruction and giving an alarm when the risk level reaches a preset pause threshold; And collecting operation feedback data and event logs, adding new samples into a continuous learning queue according to a sliding time window, and updating a wind control prediction network by adopting at least one mode of incremental learning and federal learning so as to improve the self-adaptation performance of the model in different sites and new environment scenes.
2. The method for intelligent two-ticket aided decision-making of a power system integrating environmental awareness and job risk analysis according to claim 1, wherein the ticket optimization advice includes at least risk type, corrective action, advice execution period and priority.
3. The method for intelligently making two-ticket auxiliary decision in an electric power system by combining environment sensing and operation risk analysis according to claim 1, wherein an environment sensing unit is arranged on an operation site, temperature, humidity, wind speed, wind direction, rainfall, combustible gas concentration, infrared thermal image, noise and illumination parameters are periodically collected, and a time sequence environment characteristic matrix is formed after the parameters are converged by an edge gateway, and the method comprises the following steps: The environment sensing units at least comprise a temperature sensor, a humidity sensor, an ultrasonic anemometer, a tipping bucket type rain gauge, a catalytic combustible gas probe, an infrared thermal image module, a decibel meter and an illuminometer; the zero point and the measuring range of the environment sensing unit are calibrated by adopting a reference instrument specified by national metering verification regulations, and the traceability number is recorded; According to a preset partition coverage principle, various environment sensing units are installed on the top of a main transformer, the side surface of a switch cabinet, the vicinity of an overhead insulator, the edges of a pedestrian passageway and easy leakage points of gas, and unique network addresses and sampling passageway numbers are distributed to nodes of each environment sensing unit in a 920MHz wireless ad hoc network mode; triggering network time protocol service in a gateway by using GPS pulse second signals, and performing synchronous timing on all nodes of the environment sensing unit to ensure that time base errors of all nodes are controlled within 10 milliseconds; under the driving of a schedule issued by an edge gateway, each environment sensing unit executes data acquisition in a fixed sampling period, and packages a single sampling value, a node address and a time stamp into a data frame; Each environment sensing unit sends the data frames to an edge gateway through a wireless link, the edge gateway caches the arrived data frames in ascending order according to a time stamp, and an initial environment feature matrix comprising N rows and M columns is constructed according to a preset label sequence; And calling an anomaly detection algorithm based on triple median absolute deviation at the edge gateway, and performing elimination and interpolation complement on outliers in the initial environmental feature matrix to obtain the cleaned time sequence environmental feature matrix.
4. The method for intelligently making two-ticket auxiliary decision in electric power system integrating environment awareness and operation risk analysis according to claim 1, wherein analyzing the operation ticket and the work ticket to be executed, extracting the operation type, the voltage level, the target equipment, the operator qualification, the safety measure item, the planning period and the tool information, generating the operation feature matrix by unified vectorization coding, and comprising the following steps: Receiving file input of an operation ticket and a work ticket to be executed from a preset ticket management platform, respectively converting the file input into standardized text contents through a format recognition module, extracting field titles, paragraph contents, serial number information and form areas by adopting a layout structure analysis model, and constructing a ticket initial data set containing the corresponding relation between field key names and the text contents, wherein the file input comprises three types of PDF format files, microsoft word format files and image scanning pieces; Respectively calling a named entity recognition model obtained based on deep learning training for field content in the bill initial data set, and extracting and labeling eight key semantic information, namely, an operation type, a voltage grade, a target equipment identifier, an operator name, a personnel qualification grade, a safety technical measure item, an operation start-stop period and a tool equipment name, so as to form a semantic tag set with a unique tag identifier; According to the power grid operation term specification, the equipment name standard library, the tool unified coding rule and the operation time expression dictionary, standard term matching and normalized mapping are carried out on the free text extracted from the semantic tag set, all entity items are standardized to corresponding values in a predefined word list, and a structured field vector set is output; Performing vectorization coding on each field in the structured field vector set by adopting at least one of a preset word embedding model, a single-hot coding model and a rule template coding mode, and splicing coding results of all fields according to a field sequence to form the operation feature matrix with consistent dimensionality; And if the conditions of missing fields, field disagreement formats or logic conflict exist, returning to the execution step of receiving file input of an operation ticket and a work ticket to be executed from a preset ticket management platform.
5. The method for intelligently making two-ticket auxiliary decision in electric power system by fusing environmental awareness and operation risk analysis according to claim 1, wherein the method for performing depth feature extraction on an environmental feature matrix and an operation feature matrix to obtain an environmental feature vector and an operation feature vector, performing correlation calculation and fusion on the environmental feature vector and the operation feature vector based on an attention mechanism, and generating a joint feature vector comprises: Performing normalization processing on all numerical fields of the environmental feature matrix E and the operation feature matrix W by adopting a zero mean unit variance method to obtain normalization matrices E n and W n ; A convolution layer with a one-dimensional convolution kernel size of 3 and a GELU activation function are respectively applied to E n and W n , local context information is extracted, and primary feature vectors F e 0 and Fw 0 are output; F e 0 and Fw 0 are respectively input into an independent transform coding tower with a two-layer encoder and a fixed multi-head number h, and after position coding and layer normalization, an environment characteristic vector F e and an operation characteristic vector Fw are obtained; In the same training round, cross attention weight calculation is performed on F e and Fw w , and a joint feature vector is generated, wherein the calculation process meets the following formula: Wherein F e is an environmental feature vector obtained by double-ended transducer coding, F w is an operation feature vector obtained by double-ended transducer coding, W q 、W k 、W v is a query mapping matrix, a key mapping matrix and a value mapping matrix respectively, all automatically learn in the training process, d is a scaling constant of key vector dimension for numerical stability, softmax is a line normalization exponential function, and F f is a final joint feature vector.
6. The method for intelligently making a two-ticket auxiliary decision in an electric power system by combining environment awareness and operation risk analysis according to claim 1, wherein the steps of inputting the combined feature vector into a wind control prediction network, outputting risk probability, and classifying risk levels according to a preset threshold value include: inputting the joint feature vector F f into a feedforward risk prediction network which is finished by offline training, and performing matrix multiplication and offset addition operation between an input layer and a hidden layer to form a linear mapping result; And applying a Sigmoid activation function to the linear mapping result to obtain a risk probability P, wherein the calculation formula is as follows: Wherein W r is a risk network weight matrix, b r is a risk network bias term, sigma is a Sigmoid activation function, P is an output risk probability, t 1 is a low-medium risk demarcation threshold value, t 2 is a medium-high risk demarcation threshold value, and R is a risk grade identification obtained by threshold value judgment; And smoothing the risk probability sequences in three continuous time windows in an exponential moving average mode, improving the robustness of the burst data, comparing the smoothed probability value with the level R of the previous state in consistency, and triggering a model self-checking signal if cross-two-stage jump occurs.
7. The power system intelligent two-ticket auxiliary decision-making method integrating environment awareness and operation risk analysis according to claim 1, wherein the risk probability and risk level are related to a safety rule base, non-conforming items of safety measure missing, operation condition non-conforming, equipment state abnormal and human factor risk overrun are determined, ticket optimization suggestions are generated for the non-conforming items, and the ticket optimization suggestions are pushed to operation responsible persons, guardians and approvers through at least one mode of a mobile terminal and a monitoring platform, and the method comprises the following steps: loading rule items matched with the current operation type, voltage level, equipment name and risk level, comparing operation characteristic content with the rule items, identifying non-conforming items with safety technical measure deficiency, non-conforming operation conditions, abnormal equipment state and insufficient personnel qualification, and classifying and marking the non-conforming items according to the risk type; generating bill optimization suggestions comprising correction suggestions, suggestion execution time periods and priorities according to the risk types and severity levels of each non-conforming item, and configuring mobile terminals through operation responsible persons, operation guardians and bill approvers; And pushing the optimization suggestion in a structured message form, and recording a sending state and a user response result.
8. The power system intelligent two-ticket aided decision-making method integrating environment awareness and job risk analysis according to claim 1, wherein the environment parameters and job execution state are monitored in real time during the implementation of the job, and when the risk level reaches a preset suspension threshold, a suspension instruction is automatically triggered and an alarm is sent out, comprising: Acquiring personnel in and out, tool use and operation step confirmation and equipment start-stop states in the working process in real time from a field working recording terminal, equipment worn by operators and a tool bar code recognition device, and synchronously constructing a working execution state set; collecting the temperature, wind speed, combustible gas concentration, lightning early warning value and noise level of a current operation point at fixed time intervals, and uploading the temperature, wind speed, combustible gas concentration, lightning early warning value and noise level of the current operation point and the operation execution state set to an edge computing node in a combined way to serve as risk assessment input of a current monitoring period; Invoking a deployed wind control prediction model, combining the current environmental parameters and the operation execution state, judging whether a new risk level result is generated, and comparing the new risk level result with a preset suspension threshold value; when the current risk level reaches or exceeds the suspension threshold, a suspension execution instruction is issued to the operation terminal through the local communication bus, and multichannel suspension alarm information is issued simultaneously through the on-site audible and visual alarm, the monitoring platform alarm window and the operator mobile terminal.
9. The method for intelligent two-ticket aided decision-making of an electric power system integrating environment awareness and operation risk analysis according to claim 1, wherein collecting operation feedback data and event logs, adding new samples into a continuous learning queue according to a sliding time window, and updating a wind control prediction network by adopting at least one of incremental learning and federal learning, comprises: The operation process data comprises an operator operation log, an environmental parameter change record, a system alarm record, a bill approval process log, operation completion confirmation information and an operation post-evaluation record; the operation process data is archived and integrated according to the operation number to form a complete operation feedback sample, wherein the sample containing the events of risk rise, illegal operation, rule disagreement or alarm response in the operation is marked as a high-value sample and added into a sample pool to be trained; selecting recent data from the sample pool to be trained according to a sample updating strategy of a preset sliding time window, adding the recent data into a continuous learning queue, and deleting outdated samples beyond the range of the time window to ensure that training data are continuously fresh and accord with the characteristics of the current operation scene; Executing model increment training tasks in a central server or region edge nodes, updating wind control prediction model weights through an increment learning mode or a federal learning mode based on a current continuous learning queue, and completing model version management and deployment synchronization operation.
10. An intelligent two-ticket auxiliary decision-making method of an electric power system integrating environment perception and operation risk analysis is characterized by comprising the following steps: the environment sensing unit is used for arranging the environment sensing unit on an operation site, periodically collecting temperature, humidity, wind speed, wind direction, rainfall, combustible gas concentration, infrared thermal image, noise and illumination parameters, and forming a time sequence environment characteristic matrix after converging through an edge gateway; The bill information analysis unit is used for analyzing the operation ticket to be executed and the work ticket, extracting the operation type, the voltage level, the target equipment, the personnel qualification, the safety measure item, the planning period and the tool information, and generating an operation feature matrix by unified vectorization coding; The multi-mode feature fusion unit is used for carrying out depth feature extraction on the environment feature matrix and the operation feature matrix respectively to obtain an environment feature vector and an operation feature vector, and carrying out correlation calculation and fusion on the environment feature vector and the operation feature vector based on an attention mechanism to generate a joint feature vector; the risk prediction unit is used for inputting the combined feature vector into the wind control prediction network, outputting risk probability and classifying risk grades according to a preset threshold value; The rule checking and suggestion generating unit is used for carrying out association reasoning on the risk probability and the risk level with a safety rule library, determining non-conforming items of safety measure deletion, non-conforming operation conditions, abnormal equipment state and human factor risk overrun, generating bill optimization suggestions aiming at the non-conforming items, and pushing the bill optimization suggestions to operation responsible persons, guardianship persons and approvers in at least one mode of a mobile terminal and a monitoring platform; the operation implementation monitoring and suspending control unit is used for monitoring environmental parameters and operation execution states in real time during operation implementation, and automatically triggering a suspending instruction and giving an alarm when the risk level reaches a preset suspending threshold value; The model self-adaptive updating unit is used for collecting operation feedback data and event logs, adding new samples into the continuous learning queue according to a sliding time window, and updating the wind control prediction network by adopting at least one mode of incremental learning and federal learning so as to improve the self-adaptive performance of the model in different sites and new environment scenes.

Description

Intelligent two-ticket auxiliary decision-making method and system for electric power system integrating environment perception and operation risk analysis Technical Field The invention relates to the technical field of power systems, in particular to an intelligent two-ticket auxiliary decision-making method and system for a power system integrating environment perception and operation risk analysis. Background In the operation, maintenance and overhaul process of the power system in China, operation tickets and work tickets (collectively referred to as two tickets) are mainly paper or static electronic forms, and are completed by manual filling, manual checking and manual filing. Such conventional practices rely on operator experience and lack of real-time data support for the actual environment of the site, resulting in high reliance on personal levels for ticket quality and execution security. With the fusion application of intelligent power grids, internet of things and artificial intelligence technology, electric enterprises are accelerating to advance operation and maintenance digitization, risk intelligent assessment and decision intelligentization, and new technical means such as multisource sensor deployment, 5G/industrial wireless network access, federal learning and the like are becoming important directions for improving two-ticket management refinement, real-time and closed-loop. Although the digitization trend is remarkable, most of the two-ticket systems still exist at present, ① cannot sense extreme environments such as high temperature, strong wind, thunder and lightning in real time, ② risk analysis method stays in experience estimation, quantitative modeling of three-element of human-machine-ring is lacking, and ③ system lacks intelligent examination and dynamic decision capability, so that safety prevention and control lag is caused. Therefore, a two-ticket decision-making method capable of integrating environmental awareness, risk prediction and intelligent decision-making is needed to achieve maximized safety protection and operation efficiency. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide the intelligent two-ticket auxiliary decision-making method and system for the electric power system, which are used for integrating environment perception and operation risk analysis, break through the limitation that the traditional two-ticket system depends on static rules and manual judgment, realize the whole process intelligent auxiliary decision-making of environment perception, risk prediction, rule verification and model self-evolution, and greatly improve the intrinsic safety level and the two-ticket management intelligent degree of electric power operation. In order to achieve the above object, the present invention provides the following solutions: An intelligent two-ticket auxiliary decision-making method of an electric power system integrating environment perception and operation risk analysis comprises the following steps: An environment sensing unit is arranged on an operation site, temperature, humidity, wind speed, wind direction, rainfall, combustible gas concentration, infrared thermal image, noise and illumination parameters are periodically collected, and a time sequence environment characteristic matrix is formed after the parameters are converged through an edge gateway; Analyzing an operation ticket to be executed and a work ticket, extracting the operation type, the voltage level, the target equipment, the operator qualification, the safety measure item, the planning period and the tool information, and generating an operation feature matrix by unified vectorization coding; deep feature extraction is respectively carried out on the environment feature matrix and the operation feature matrix to obtain an environment feature vector and an operation feature vector, and correlation calculation and fusion are carried out on the environment feature vector and the operation feature vector based on an attention mechanism to generate a joint feature vector; inputting the combined feature vector into a wind control prediction network, outputting risk probability, and dividing risk grades according to a preset threshold value; carrying out association reasoning on the risk probability and the risk level with a safety rule base, determining non-conforming items of safety measure deficiency, non-conforming operation conditions, abnormal equipment state and human factor risk overrun, generating bill optimization suggestions aiming at the non-conforming items, and pushing the bill optimization suggestions to operation responsible persons, guardianship persons and approvers through at least one mode of a mobile terminal and a monitoring platform; Monitoring environmental parameters and job execution states in real time during the implementation of the job, and automatically triggering a pause instruction and giving an alarm when t