Modeling method of convolutional neural network model for transformer substation behavior monitoring

Abstract

A modeling method of a convolutional neural network model for transformer substation behavior monitoring comprises the following steps: intercepting real-time video streams and video frames, performing data set labeling on collected transformer substation behavior images, and constructing an image model data set; dividing the data set into a training set and a verification set; carrying out improvement and optimization based on a target detection algorithm framework yolov4, and training the training set; performing preliminary evaluation on the obtained training model on the verification set; acquiring a video through an ip address, detecting each frame of picture in a video stream, framing out an illegal behavior, and giving the category and confidence of the illegal behavior; and writing the detected frame with the illegal behavior into a video file, and writing video information into a database. The method has the advantages that all-weather and real-time transformer substation behavior monitoring can be achieved, various illegal behaviors of a transformer substation can be detected, detected illegal videos are automatically stored in a hard disk, video information is input into a database, and a basis is provided for accident analysis and performance evaluation.

Description

technical field [0001] The invention relates to a modeling method of a convolutional neural network model used for substation behavior monitoring, and belongs to the field of substation safety artificial intelligence monitoring and identification. Background technique [0002] According to incomplete statistics, there are currently more than 20,000 substations in the country, and the number is still growing. The substation is the hub of the power system and belongs to the high-risk industrial field. Any violation of relevant regulations may bring great potential safety hazards to people's lives and property safety. The safety behavior monitoring of substation infrastructure is monitored through manual "staring". However, due to human physiological characteristics, there will always be fatigue, slack, etc., and it is impossible to monitor all violations in real time. At the work conference of State Grid in 2019, State Grid has clearly proposed to greatly improve the intellig...

AI Technical Summary

Problems solved by technology

However, directly applying target detection technology to substation behavior monitoring cannot meet the requirements of violation monitoring. If the violation video frame is not intercepted in real time, it is impossible to judge whether it is a violation or a mistake to trigger the violation detection according to the length of the violation duration.

Images