CN-121998433-A - Electromechanical installation engineering safety management early warning method and system based on large model

CN121998433ACN 121998433 ACN121998433 ACN 121998433ACN-121998433-A

Abstract

The invention discloses an electromechanical installation engineering safety management early warning method and system based on a large model, which are characterized in that multimode data are acquired, the multimode data are subjected to space-time synchronization and feature level fusion by utilizing a cross-modal alignment algorithm to obtain fusion multimode characteristic data, dynamic space-time diagram structural data are built based on the fusion multimode characteristic data and are input into STGCN time sequence diagram convolution networks, high-risk behavior characteristics are distinguished through a diagram attention mechanism, space-time characteristic data are output, the space-time characteristic data are input into MANN memory enhancement neural networks, installation engineering risk indexes are output, evaluation is carried out according to the installation engineering risk indexes, and hierarchical early warning is triggered to generate an intervention strategy. The safety management efficiency is improved, the labor cost is reduced, and the occurrence rate of safety accidents is reduced.

Inventors

LI JIANZHAO
CHEN YUNXUAN
XUE LELE
LIANG JUGUANG
YU SHAOHUI
LUO QIAOJUAN
SONG KAI
REN QIJUN
CUI HONGFANG
LI WEIYING

Assignees

贵州金刚智造技术有限公司

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The electromechanical installation engineering safety management early warning method based on the large model is characterized by comprising the following steps of: Visual data, point cloud data, sensor data and management document data of the electromechanical installation engineering construction site are obtained, and the collected data are preprocessed to obtain multi-mode data; Performing space-time synchronization and feature level fusion on the multi-modal data by using a cross-modal alignment algorithm to obtain fused multi-modal feature data; based on the fusion of the multi-modal feature data, the dynamic space-time diagram structural data is constructed and input into STGCN time sequence diagram convolution network, high-risk behavior features are distinguished through a diagram attention mechanism, and space-time feature data is output; And inputting the space-time characteristic data into MANN memory-enhanced neural network, outputting an installation engineering risk index, evaluating according to the installation engineering risk index, and triggering grading early warning to generate an intervention strategy.
2. The method for pre-warning safety management of electromechanical installation engineering based on large model as claimed in claim 1, wherein the steps of obtaining visual data, point cloud data, sensor data and management document data of the construction site of the electromechanical installation engineering, preprocessing the collected data to obtain multi-mode data, and comprises the following steps: visual data, point cloud data, sensor data and management document data of the electromechanical installation engineering construction site are acquired; Eliminating noise caused by illumination change and dust shielding in visual data by using an image enhancement algorithm, extracting frames of the visual data, reserving key action frames, positioning human bodies and equipment targets by using a target detection algorithm, and unifying image sizes after cutting invalid areas; Removing environmental interference points and equipment scanning noise points in the point cloud data through a straight-through filtering algorithm, performing downsampling on the processed point cloud data, and uniformly converting the point cloud data into a construction site global coordinate system through coordinate calibration; Filling missing data of sensor data by using an interpolation method, eliminating errors caused by sensor shake by using a Kalman filtering algorithm, and carrying out standardized processing on the values of different types of sensors; And carrying out word segmentation and stop word removal processing on unstructured text in the management document data based on an NLP natural language processing technology, extracting information and converting the information into structured data to obtain multi-modal data.
3. The method for pre-warning safety management of electromechanical installation engineering based on a large model according to claim 1, wherein the step of performing space-time synchronization and feature level fusion on the multi-modal data by using a cross-modal alignment algorithm to obtain fused multi-modal feature data comprises the following steps: aligning the frame time of the video data, the scanning time of the point cloud data and the updating time of the management document data by taking the time stamp of the sensor data as a reference to obtain time alignment data; And converting pixel coordinates of the visual data into space coordinates through camera calibration parameters by taking a global coordinate system of point cloud data in the time alignment data as a reference, accurately aligning with the space position of the point cloud data, and simultaneously mapping various data to corresponding construction areas by combining construction site partition information in management document data to obtain space-time alignment data.
4. The method for pre-warning safety management of electromechanical installation engineering based on a large model according to claim 1, wherein the step of performing space-time synchronization and feature level fusion on the multi-modal data by using a cross-modal alignment algorithm to obtain fused multi-modal feature data comprises the following steps: extracting human body action features, equipment appearance features and environmental scene features in the space-time alignment data through a CNN convolutional neural network; extracting spatial structure features, equipment installation precision features and operation space interval features in space-time alignment data through PointNet ++ algorithm; Extracting equipment running state characteristics, personnel physiological characteristics and environment abnormal characteristics in the time-space alignment data by an MLP (multi-layer perceptron) machine; extracting safety rule features, risk threshold features and personnel qualification features in the space-time alignment data through a BERT large model; And carrying out weighted fusion on various modal characteristics by using an attention mechanism-based cross-modal fusion algorithm, and distributing weights according to the importance of the modal data under different construction scenes to obtain fused multi-modal characteristic data.
5. The method for pre-warning safety management of electromechanical installation engineering based on a large model according to claim 1, wherein the step of constructing dynamic space-time diagram structural data based on fusion of multi-modal feature data, inputting STGCN time sequence diagram convolution network, distinguishing high risk behavior features through a diagram attention mechanism, and outputting space-time feature data comprises the following steps: Inputting the fused multi-modal characteristic data into STGCN time sequence chart convolution network, and constructing a dynamic time-space chart by taking human joint coordinates of constructors as core nodes; Taking each operator in a construction site as an independent individual, extracting three-dimensional space coordinates of 18 key joints of a human body in visual data and point cloud data, taking each joint coordinate as an independent node, adding corresponding characteristic information for each node, and defining the node; Two types of edges are constructed, the physical connection edges are connected with adjacent joint nodes of the same person according to the physiological structure of the human body, the physical relevance of human body actions is represented, and the space-time relevance edges are connected with the same joint nodes of different persons.
6. The method for pre-warning safety management of electromechanical installation engineering based on a large model according to claim 1, wherein the step of constructing dynamic space-time diagram structural data based on fusion of multi-modal feature data, inputting STGCN time sequence diagram convolution network, distinguishing high risk behavior features through a diagram attention mechanism, and outputting space-time feature data comprises the following steps: Calculating the attention weight of each node through a diagram attention mechanism of STGCN, and distributing high attention weight to nodes and edges related to high risk behaviors; Extracting time sequence change characteristics of nodes and edges through a STGCN time sequence convolution module, extracting space correlation characteristics through a graph convolution module, fusing the time sequence characteristics and the space characteristics, and outputting space-time characteristic data.
7. The method for pre-warning safety management of electromechanical installation engineering based on large model as set forth in claim 1, wherein the inputting MANN the space-time characteristic data into the memory-enhanced neural network, outputting an installation engineering risk index, evaluating according to the installation engineering risk index, and triggering the hierarchical pre-warning to generate an intervention strategy comprises: The space-time characteristic data are input into MANN memory-enhanced neural networks, and the memory module is used for storing historical construction risk data of the electromechanical installation engineering, a construction site safety specification threshold value and risk assessment standard information; the calculation module compares the space-time characteristic data with the historical data in the memory module, matches the most similar historical risk cases through similarity calculation, extracts corresponding risk weights, combines the high risk identification in the current space-time characteristic, and calculates the installation engineering risk index in a weighted summation mode.
8. The electromechanical installation engineering safety management early warning system based on the large model is characterized by comprising the following modules: The multi-mode data acquisition module is used for acquiring visual data, point cloud data, sensor data and management document data of the electromechanical installation engineering construction site, preprocessing the acquired data and obtaining multi-mode data; The feature data fusion module is used for carrying out space-time synchronization and feature level fusion on the multi-mode data by using a cross-mode alignment algorithm to obtain fused multi-mode feature data; The space-time feature extraction module is used for constructing dynamic space-time diagram structural data based on fusion of multi-modal feature data, inputting the dynamic space-time diagram structural data into STGCN time sequence diagram convolution network, distinguishing high-risk behavior features through a diagram attention mechanism and outputting space-time feature data; And the engineering safety management module is used for inputting the space-time characteristic data into the MANN memory-enhanced neural network, outputting an installation engineering risk index, evaluating according to the installation engineering risk index, and triggering hierarchical early warning to generate an intervention strategy.
9. The large model based electromechanical installation engineering safety management early warning system of claim 8, wherein the spatio-temporal feature extraction module comprises the following sub-modules: the computing sub-module is used for computing the attention weight of each node through a map annotation force mechanism of STGCN and distributing high attention weights to the nodes and edges related to the high risk behaviors; And the extraction submodule is used for extracting time sequence change characteristics of the nodes and the edges through the STGCN time sequence convolution module, extracting space correlation characteristics through the graph convolution module, fusing the time sequence characteristics with the space characteristics and outputting space-time characteristic data.
10. The large model based electromechanical installation engineering safety management early warning system of claim 8, wherein the engineering safety management module comprises the following sub-modules: the input sub-module is used for inputting the space-time characteristic data into the MANN memory-enhanced neural network, and the memory module is used for storing historical construction risk data of the electromechanical installation engineering, a construction site safety specification threshold value and risk assessment standard information; And the evaluation sub-module is used for comparing the space-time characteristic data with the historical data in the memory module, matching the most similar historical risk cases through similarity calculation, extracting corresponding risk weights, combining high risk identification in the current space-time characteristic, and calculating the installation engineering risk index in a weighted summation mode.

Description

Electromechanical installation engineering safety management early warning method and system based on large model Technical Field The invention relates to the technical field of safety management of constructional engineering, in particular to an electromechanical installation engineering safety management early warning method and system based on a large model. Background The construction site of the electromechanical installation engineering has complex environment, large personnel flow, multiple equipment types, high safety risks in links such as high-altitude operation, equipment hoisting and electricity utilization operation and the like, and the safety management difficulty is extremely high. The traditional safety management mode mainly relies on manual inspection and paper recording, has the outstanding problems of lag response, low recognition precision, limited coverage range and the like, and is difficult to cope with multi-scene and dynamic potential safety hazards. The existing early warning technology depends on single type data, can only realize local hidden trouble identification, lacks the integrated utilization of multidimensional data such as vision, space, equipment operation, management standards and the like, is easy to generate the conditions of missed judgment and misjudgment, and is difficult to realize the advanced prejudgment and hierarchical management and control of risks. Disclosure of Invention The invention aims to solve the problems and designs an electromechanical installation engineering safety management early warning method and system based on a large model. The technical scheme for achieving the purpose is that in the electromechanical installation engineering safety management early warning method based on the large model, the electromechanical installation engineering safety management early warning method comprises the following steps of: Visual data, point cloud data, sensor data and management document data of the electromechanical installation engineering construction site are obtained, and the collected data are preprocessed to obtain multi-mode data; Performing space-time synchronization and feature level fusion on the multi-modal data by using a cross-modal alignment algorithm to obtain fused multi-modal feature data; based on the fusion of the multi-modal feature data, the dynamic space-time diagram structural data is constructed and input into STGCN time sequence diagram convolution network, high-risk behavior features are distinguished through a diagram attention mechanism, and space-time feature data is output; And inputting the space-time characteristic data into MANN memory-enhanced neural network, outputting an installation engineering risk index, evaluating according to the installation engineering risk index, and triggering grading early warning to generate an intervention strategy. Further, in the electromechanical installation engineering safety management early warning method based on the large model, the acquiring visual data, point cloud data, sensor data and management document data of the electromechanical installation engineering construction site, preprocessing the acquired data to obtain multi-mode data includes: visual data, point cloud data, sensor data and management document data of the electromechanical installation engineering construction site are acquired; Eliminating noise caused by illumination change and dust shielding in visual data by using an image enhancement algorithm, extracting frames of the visual data, reserving key action frames, positioning human bodies and equipment targets by using a target detection algorithm, and unifying image sizes after cutting invalid areas; Removing environmental interference points and equipment scanning noise points in the point cloud data through a straight-through filtering algorithm, performing downsampling on the processed point cloud data, and uniformly converting the point cloud data into a construction site global coordinate system through coordinate calibration; Filling missing data of sensor data by using an interpolation method, eliminating errors caused by sensor shake by using a Kalman filtering algorithm, and carrying out standardized processing on the values of different types of sensors; And carrying out word segmentation and stop word removal processing on unstructured text in the management document data based on an NLP natural language processing technology, extracting information and converting the information into structured data to obtain multi-modal data. Furthermore, in the electromechanical installation engineering safety management early warning method based on the large model, the performing space-time synchronization and feature level fusion on the multi-mode data by using a cross-mode alignment algorithm to obtain fused multi-mode feature data comprises the following steps: aligning the frame time of the video data, the scanning time of the point cloud data and the u