CN-121980494-A - Multisource supervision data fusion method and system

CN121980494ACN 121980494 ACN121980494 ACN 121980494ACN-121980494-A

Abstract

The invention relates to the technical field of power supervision and discloses a multisource supervision data fusion method and system. The method comprises the steps of obtaining multi-source heterogeneous data generated in a power engineering supervision process, converting the multi-source heterogeneous data into standardized unified data through format conversion, carrying out time synchronization and space alignment on the standardized unified data to obtain collaborative association data with unified space-time references, carrying out layered fusion on the collaborative association data according to different functional dimensions to obtain fusion characteristics of multiple levels, inputting each fusion characteristic into a pre-built deep learning model after being weighted in real time to carry out cross-modal association reasoning to obtain supervision decision support information, carrying out risk threshold comparison on the supervision decision support information, generating structured early warning information according to comparison results, and carrying out real-time display on the structured early warning information. The invention provides effective technical support for power engineering construction safety and quality control.

Inventors

LIN ZHOUYOU
XU XUDONG
HONG CHENSONG
Ye Shangnan
ZHENG PU
WANG TAO
LIN YUEJIN
XU SISI
YING WEIJUN
CHI CHAOFAN
WU GUANG
LI YI

Assignees

温州电力设计有限公司监理分公司

Dates

Publication Date: 20260505
Application Date: 20251226

Claims (10)

1. The multi-source supervision data fusion method is characterized by comprising the following steps of: the method comprises the steps of obtaining multi-source heterogeneous data generated in the power engineering supervision process, and converting the multi-source heterogeneous data into standardized unified data through format conversion; Performing time synchronization and space alignment on the standardized unified data to obtain collaborative association data with unified space-time reference; Layering and fusing the collaborative correlation data according to different functional dimensions to obtain a plurality of layers of fusion features, inputting each fusion feature into a pre-constructed deep learning model after real-time weighting to perform cross-modal correlation reasoning to obtain supervision decision support information, wherein the deep learning model learns the correlation mode between each functional dimension and multi-modal data through historical power engineering supervision data in the training process; And comparing the risk threshold value of the supervision decision support information, generating structural early warning information according to a comparison result, and displaying the structural early warning information in real time.
2. The method for fusing multi-source management data according to claim 1, wherein the steps of obtaining multi-source heterogeneous data generated in the power engineering management process, and converting the multi-source heterogeneous data into standardized unified data through format conversion include: The method comprises the steps of obtaining multi-source heterogeneous data generated in a power engineering supervision process, wherein the multi-source heterogeneous data comprise space data, sensor data, service data and video image data; And carrying out format standardization, coordinate system unification and semantic mapping on the space data, the sensor data, the service data and the video image data to obtain standardized unified data containing a unified format, a unified coordinate system and a unified semantic tag.
3. The multi-source supervision data fusion method according to claim 2, wherein the spatial data comprises a point cloud model generated by SLAM real-time modeling, a BIM design model and physical coordinates of a two-dimensional code positioning anchor point, the sensor data comprises environment sensor data, equipment sensor data and construction machinery state data, the service data comprises structured records, supervision logs and construction progress plans after paper file electronization, and the video image data comprises a site real-time picture acquired by an AR terminal and a guide image marked by a remote expert.
4. The method for merging multisource supervision data according to claim 1, wherein the performing time synchronization and spatial alignment on the standardized unified data to obtain collaborative association data with unified space-time reference includes: And carrying out time synchronization on the standardized unified data through network time protocol and local clock calibration, and carrying out space alignment on the standardized unified data with unified time stamp by combining space association rules and space positioning calibration to obtain collaborative association data with unified space-time reference.
5. The multi-source supervision data fusion method according to claim 1, wherein the step of hierarchically fusing the collaborative association data according to different functional dimensions to obtain a plurality of levels of fusion features, and inputting each fusion feature into a pre-constructed deep learning model for cross-modal association reasoning after real-time weighting to obtain supervision decision support information comprises the steps of: Layering and fusing the collaborative association data according to a space positioning dimension, an equipment state dimension, a business compliance dimension and a visual defect dimension to obtain fusion characteristics of a plurality of layers; And after weighting each fusion feature according to the reliability score and the timeliness score calculated in real time, inputting the weighted reliability score and the timeliness score into a pre-constructed deep learning model to perform cross-modal correlation reasoning so as to obtain supervision decision support information.
6. The method for fusing multisource supervision data according to claim 5, wherein the hierarchically fusing the collaborative association data according to a spatial location dimension, a device state dimension, a business compliance dimension, and a visual defect dimension to obtain a plurality of hierarchy fusion features includes: fusing the collaborative association data according to the space positioning dimension to obtain the space coordinate characteristics of the corresponding dimension level; Fusing the collaborative association data according to the equipment state dimension to obtain equipment state characteristics of a corresponding dimension level; fusing the cooperative association data according to the service compliance dimension to obtain a procedure compliance judging feature of a corresponding dimension level; and fusing the collaborative correlation data according to the visual defect dimension to obtain defect positioning labeling features of corresponding dimension layers.
7. The multi-source proctoring data fusion method of claim 5, wherein the deep learning model is based on a fransformer architecture, and uses MobileNetV as a feature extraction backbone network.
8. The multi-source proctoring data fusion method of claim 5, wherein the proctoring decision support information comprises a risk event occurrence probability, a process compliance decision, and a defect localization result.
9. The multi-source proctoring data fusion method of claim 1, wherein the performing risk threshold comparison on the proctoring decision support information, generating structured early warning information according to a comparison result, and displaying the structured early warning information in real time comprises: Comparing the quantized risk index in the supervision decision support information with a predefined power engineering supervision risk threshold, if the quantized risk index exceeds the power engineering supervision risk threshold, generating structural early warning information, and displaying the structural early warning information in real time through a visual interface.
10. A multi-source proctoring data fusion system, comprising: The acquisition and conversion module is used for acquiring multi-source heterogeneous data generated in the power engineering supervision process and converting the multi-source heterogeneous data into standardized unified data through format conversion; the space-time alignment module is used for carrying out time synchronization and space alignment on the standardized unified data to obtain collaborative association data with unified space-time reference; The fusion reasoning module is used for carrying out layered fusion on the collaborative correlation data according to different functional dimensions to obtain a plurality of layers of fusion features, inputting each fusion feature into a pre-constructed deep learning model after being weighted in real time to carry out cross-modal correlation reasoning to obtain supervision decision support information, wherein the deep learning model learns the correlation mode between each functional dimension and multi-modal data through historical power engineering supervision data in the training process; And the comparison display module is used for comparing the risk threshold values of the supervision decision support information, generating structural early warning information according to a comparison result, and displaying the structural early warning information in real time.

Description

Multisource supervision data fusion method and system Technical Field The invention relates to the technical field of power supervision, in particular to a multisource supervision data fusion method and system. Background The electric power engineering construction has the characteristics of large scale, complex process links and strict safety control requirements, engineering supervision is used as a key link for guaranteeing construction quality, safety compliance and progress control, and the working mode of the electric power engineering construction is gradually updated from traditional manual inspection to an intelligent and data-driven direction. With the deep application of synchronous positioning and map construction (SLAM), building Information Model (BIM), internet of things sensor, augmented Reality (AR) and other technologies in the field of electric power supervision, multi-type and multi-source heterogeneous mass data can be generated in the supervision working process. How to realize the effective fusion and cooperative application of mass data has important significance for the intelligent management and control of the power engineering supervision. At present, in a multisource data fusion link, most of the prior art can only realize simple association of single type data such as sensor data and video data, and aiming at scene characteristics of power engineering supervision, a technical scheme capable of solving accurate alignment and dynamic fusion of multi-type and multi-scale data is not provided, and real-time and intelligent management and control requirements of supervision work on all elements of a field 'man-machine-environment' cannot be supported. Disclosure of Invention Aiming at the bottleneck existing in the prior art, the invention provides a multi-source supervision data fusion method and system. In a first aspect, an embodiment of the present invention provides a method for fusing multisource supervision data, including: the method comprises the steps of obtaining multi-source heterogeneous data generated in the power engineering supervision process, and converting the multi-source heterogeneous data into standardized unified data through format conversion; Performing time synchronization and space alignment on the standardized unified data to obtain collaborative association data with unified space-time reference; Layering and fusing the collaborative correlation data according to different functional dimensions to obtain a plurality of layers of fusion features, inputting each fusion feature into a pre-constructed deep learning model after real-time weighting to perform cross-modal correlation reasoning to obtain supervision decision support information, wherein the deep learning model learns the correlation mode between each functional dimension and multi-modal data through historical power engineering supervision data in the training process; And comparing the risk threshold value of the supervision decision support information, generating structural early warning information according to a comparison result, and displaying the structural early warning information in real time. Preferably, the obtaining the multi-source heterogeneous data generated in the power engineering supervision process, and converting the multi-source heterogeneous data through format to obtain standardized unified data includes: The method comprises the steps of obtaining multi-source heterogeneous data generated in a power engineering supervision process, wherein the multi-source heterogeneous data comprise space data, sensor data, service data and video image data; And carrying out format standardization, coordinate system unification and semantic mapping on the space data, the sensor data, the service data and the video image data to obtain standardized unified data containing a unified format, a unified coordinate system and a unified semantic tag. Preferably, the space data comprises a point cloud model, a BIM design model and physical coordinates of a two-dimensional code positioning anchor point, which are generated by SLAM real-time modeling, the sensor data comprises environment sensor data, equipment sensor data and construction machinery state data, the service data comprises structured records, supervision logs and construction progress plans after paper file electronization, and the video image data comprises a site real-time picture acquired by an AR terminal and a guiding image marked by a remote expert. Preferably, the performing time synchronization and spatial alignment on the standardized unified data to obtain collaborative association data with unified space-time reference includes: And carrying out time synchronization on the standardized unified data through network time protocol and local clock calibration, and carrying out space alignment on the standardized unified data with unified time stamp by combining space association rules and space positioning calibration to obt