CN-121982813-A - Power plant safety intelligent early warning method based on dynamic electronic fence

CN121982813ACN 121982813 ACN121982813 ACN 121982813ACN-121982813-A

Abstract

The invention discloses a power plant safety intelligent early warning method based on a dynamic electronic fence, which comprises the steps of analyzing historical work ticket data of a power plant through a BERT-BiLSTM-CRF hybrid model, and identifying equipment entities, operation types and risk grades. Combining the space coordinates of a three-dimensional model of power plant equipment, automatically dividing an initial virtual safety boundary according to the operation content of a working ticket, fusing the operation actions of operators, the position relations of multiple persons and the equipment state in real time, dynamically adjusting the buffer distance to form a dynamic electronic fence, establishing a group operation anti-collision model and a safety path planning strategy based on the dynamic electronic fence and high-precision positioning data, constructing a grading early warning mechanism, and triggering corresponding grade early warning according to the real-time distance between the operators and the electronic fence through the linkage of intelligent wearing equipment and a power plant monitoring system. The invention can obviously improve the intelligentized and refined level of the safety management of the power plant and effectively reduce the occurrence rate of operation safety accidents.

Inventors

ZHANG JUNWEI
LIU LIANWEI
LI JUN
CHEN CHENG
CHEN SHAOHUA
GAN HUIMIN
TIAN HENGSHUANG
WANG XUESONG

Assignees

中国长江电力股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260114

Claims (10)

1. The intelligent power plant safety early warning method based on the dynamic electronic fence is characterized by comprising the following steps of: S1, acquiring historical work ticket data of a power plant, establishing a power plant work ticket semantic analysis method based on a BERT-BiLSTM-CRF hybrid model, and identifying equipment entities, operation types and risk grades in the work ticket; S2, combining the space coordinates of a three-dimensional model of power plant equipment, automatically dividing an initial virtual safety boundary according to the operation content of a work ticket, fusing the operation actions of operators, the position relations of multiple persons and the equipment states in real time, and dynamically adjusting the buffer distance to form a dynamic electronic fence; S3, based on the dynamic electronic fence and the high-precision positioning data, a group operation anti-collision model and a grading early warning mechanism are established, a safety path is planned for an operator in real time, and corresponding grade early warning is triggered according to the real-time distance between the operator and the electronic fence through linkage of intelligent wearing equipment and a power plant monitoring system.
2. The intelligent early warning method for power plant safety based on dynamic electronic fence according to claim 1, wherein the step S1 specifically comprises the following steps: s1.1, acquiring historical work ticket text data of a power plant, and cleaning and standardizing the original data; S1.2 based on a general BERT model, incremental pre-training is carried out on the general BERT model by using a work ticket text and electric power domain knowledge expectation of an electric power safety rule, the preprocessed text is input into the BERT model after fine adjustment, 768-dimensional dynamic vector representation of each character is obtained, and a feature matrix is output ; S1.3 the feature matrix of step S1.2 Inputting BiLSTM a model network, analyzing the context relation of the characters based on BiLSTM model to obtain the context feature vector of each character ; S1.4 output vector of step S1.3 based on CRF model Performing label constraint decoding, injecting power safety rule knowledge to ensure that output accords with a physical law and an operation rule; s1.5, carrying out structural output on the identified equipment entity, the identified operation type and the identified risk level label sequence, establishing a structural operation instruction and a coordinate mapping rule base, and associating the work ticket operation instruction with the target equipment.
3. The intelligent early warning method for power plant safety based on dynamic electronic fence according to claim 1, wherein the step S2 specifically comprises the following steps: S2.1, acquiring space coordinate data of different devices, and presetting an initial buffer distance by taking target devices as the center according to the device types and the operation risk level Dividing a polygonal electronic fence; S2.2, fusing multielement real-time data of the positions, actions and the multi-person relations of the operators through a Gaussian Process Regression (GPR) model, and dynamically adjusting the buffer distance D; S2.3, predicting the movement track of the operator through a Kalman filtering model, and adjusting the boundary of the electronic fence in advance.
4. The intelligent early warning method for power plant safety based on dynamic electronic fence according to claim 3, wherein distance parameters of operators in the GPR model input feature vector are obtained based on a multi-source fusion high-precision positioning method constructed by UWB, IMU and visual SLAM.
5. The intelligent power plant safety early warning method based on the dynamic electronic fence according to claim 1, wherein the community operation anti-collision model in the step S3 adopts a TTC algorithm, collision risk is estimated through multisource fusion positioning data real-time sharing of operators, and safety path planning is achieved through global initial path generation, local dynamic adjustment and multi-agent collaborative optimization strategies in combination with dynamic electronic fence constraint.
6. The dynamic electronic fence-based power plant safety intelligent early warning method according to claim 1, wherein the hierarchical early warning mechanism comprises a primary early warning (warning L1), a secondary early warning (warning L2) and a tertiary early warning (emergency stop L3); triggering intelligent wearable equipment to vibrate for prompting when the real-time distance between a worker and the electronic fence is greater than a preset threshold L1; When the real-time distance between the worker and the electronic fence is greater than a preset threshold L2, triggering the intelligent wearable equipment to vibrate and carry out voice prompt, and simultaneously, triggering a power plant monitoring system to pop up a warning picture; When the real-time distance between the worker and the electronic fence is greater than a preset threshold L3, triggering strong vibration and voice warning of the intelligent wearable equipment, and triggering an equipment control system to stop emergently in a high-risk area; the intelligent wearing equipment comprises an intelligent safety helmet, AR protective glasses and an intelligent bracelet.
7. The intelligent early warning method for power plant safety based on dynamic electronic fence according to claim 3, wherein the source fusion high-precision positioning method adopts a tight coupling architecture, an IMU pre-integration residual error, a visual SLAM re-projection residual error and a UWB ranging residual error are incorporated into a unified objective function, and the position, speed, posture and sensor zero offset parameters of an operator are jointly optimized through a sliding window optimization algorithm, wherein the UWB ranging residual error adopts a Huber kernel function to inhibit multipath interference, and the positioning precision under a complex industrial environment is improved.
8. The intelligent early warning method for power plant safety based on dynamic electronic fence according to claim 7, characterized in that the visual SLAM re-projection residual is specifically that environmental images are collected in real time through a visual camera, key points are extracted through an ORB algorithm, camera motion is calculated through hamming distance matching of feature points of adjacent frames, and accumulated drift of IMU is corrected based on the re-projection error, so that the boundary of the dynamic fence is ensured to be consistent with the environment.
9. The intelligent early warning method for power plant safety based on the dynamic electronic fence according to claim 1, wherein the kernel function of the GPR model in the step S2.2 adopts a combination form of square index kernel and white noise kernel, the input feature vector comprises real-time distance between an operator and target equipment, moving speed, action amplitude, minimum distance between multiple operators and real-time state of the equipment, and the posterior mean value is output as the dynamic buffer distance through a reasonable buffer distance training model manually marked in a historical scene, so that the self-adaptive adjustment of the boundary of the electronic fence is realized.
10. The intelligent early warning method for power plant safety based on dynamic electronic fence according to claim 5, wherein the safety path planning adopts a three-level architecture of global initial path generation, local dynamic adjustment and multi-agent collaborative optimization: An environment map constructed based on visual SLAM is adopted firstly Generating an initial global path by a search algorithm; Based on the group state data and the collision risk assessment result, adopting a rapid expansion random tree to locally adjust the global path; And finally, constructing a multi-agent cooperation path planning model aiming at the group operation scene, and sharing group path data through multi-hop communication to realize collision-free cooperation.

Description

Power plant safety intelligent early warning method based on dynamic electronic fence Technical Field The invention belongs to the technical field of power plant safety management, and particularly relates to a power plant safety intelligent early warning method based on a dynamic electronic fence. Background The power plant safety management relates to the life safety of operators and the protection guarantee of important power equipment, and is a core foundation stone for sustainable and healthy development of the power industry. The current power plant safety protection system mainly relies on physical isolation, fixed electronic fence and manual inspection mechanisms. The static electronic fence generates a rigid boundary based on a preset geographic coordinate, and cannot respond to the changing requirements of dynamic operation scenes such as equipment maintenance, multi-person cooperation and the like, so that a safety area is seriously misplaced with an actual risk. The management of the work ticket still takes manual auditing as a main part, operation and maintenance personnel need to consume a large amount of time to analyze unstructured texts (such as equipment coordinates and operation instructions), the identification accuracy is limited by subjective experience, and the critical risk omission rate is high. The existing protection system has the problem of response delay generally, and cannot meet the quick response requirement of high-risk scenes. The personnel positioning technology mostly adopts independent systems, the positioning precision is obviously fluctuated by the interference of complex industrial environments, and the error of a single technical scheme in a metal dense area is higher. The risk early warning mechanism presents fragmentation characteristics, data cooperation is lacking among the positioning terminal, the monitoring system and the wearable equipment, a main stream single-stage warning mode is difficult to match with a multi-stage risk scene, hierarchical intervention cannot be realized, the group operation anti-collision relies on a traditional algorithm, real-time personnel action and equipment state data are not fused, and the sudden track deviation prediction capability is weak. Meanwhile, the prior art framework lacks closed-loop capability of risk identification, dynamic protection and active intervention, and the risk of safety accidents is high due to actual disconnection of a protection system and operation. Disclosure of Invention Aiming at the problems that a static safety boundary of a power plant is difficult to adapt to a dynamic operation scene, the risk response delay is high, and the safety protection and the execution intervention are disjoint, the invention comprehensively considers the real-time performance of the operation personnel, the equipment state coupling and the group cooperation complexity, and provides a power plant safety intelligent early warning method based on a dynamic electronic fence based on a semantic intelligent analysis technology, a multi-metadata fusion prediction algorithm and a hierarchical linkage control framework. In order to realize the technical characteristics, the invention discloses a power plant safety intelligent early warning method based on a dynamic electronic fence, which comprises the following steps: Step S1, acquiring power plant historical work ticket data, establishing a power plant work ticket semantic analysis method based on a BERT-BiLSTM-CRF hybrid model, and identifying equipment entities, operation types and risk grades in the work ticket; step S2, combining the space coordinates of the three-dimensional model of the power plant equipment, automatically dividing an initial virtual safety boundary according to the operation content of the work ticket, fusing the operation actions of operators, the position relations of multiple persons and the equipment state in real time, and dynamically adjusting the buffer distance to form a dynamic electronic fence; Step S3, based on the dynamic electronic fence and the high-precision positioning data, establishing a group operation anti-collision model and a grading early warning mechanism, planning a safety path for an operator in real time, and triggering corresponding grade early warning according to the real-time distance between the operator and the electronic fence through linkage of intelligent wearing equipment and a power plant monitoring system; the step S1 specifically comprises the following steps: s1.1, acquiring historical work ticket text data of a power plant, and cleaning and standardizing the original data; step S1.2, based on a general BERT model, incremental pre-training is carried out on the general BERT model by using a work ticket text and electric power field knowledge expectation of an electric power safety regulation; Model parameter set optimized through pre-training Can be expressed as: ; In the formula, The model parameter set is opt