The "Power version GPT" increases the efficiency of safety hazard alarms by six times

Tech 2023-09-18 09:09:30 Source: Network

Reporter Ye Qing, Correspondent Yang JingjingWhat will be the effect of using ChatGPT in the power industry? On September 7th, the reporter saw at the Shenzhen Power Supply Bureau of China Southern Power Grid (hereinafter referred to as the Shenzhen Power Supply Bureau) that a staff member lightly clicked the mouse, and a hidden danger image of a crane arm adjacent to the transmission line immediately appeared on a construction site; Continuing to click on the image, the system will provide text like ChatGPT, describing the hidden danger.This is the first multimodal pre trained large model in the power industry launched by Shenzhen Power Supply Bureau - "Zhurong 2

Reporter Ye Qing, Correspondent Yang Jingjing

What will be the effect of using ChatGPT in the power industry? On September 7th, the reporter saw at the Shenzhen Power Supply Bureau of China Southern Power Grid (hereinafter referred to as the Shenzhen Power Supply Bureau) that a staff member lightly clicked the mouse, and a hidden danger image of a crane arm adjacent to the transmission line immediately appeared on a construction site; Continuing to click on the image, the system will provide text like ChatGPT, describing the hidden danger.

This is the first multimodal pre trained large model in the power industry launched by Shenzhen Power Supply Bureau - "Zhurong 2.0". It endows traditional power grid AI technology with logical reasoning and textual expression capabilities similar to ChatGPT, improving the efficiency of power grid safety hazard warning by six times.

Enable the power inspection system to reason

Traditional power grid AI technology identifies objects through representations such as color, shape, and texture, but it is difficult to determine the degree of danger that the object poses to power facilities. For objects with similar shapes, traditional power grid AI technology is also difficult to accurately distinguish them. "Zhang Yunxiang, a technical expert at Shenzhen Power Supply Bureau, introduced, For example, in traditional mode, although the system can filter out alarm screens with bright lights near the line, it still requires inspection personnel to further determine whether the light is caused by a fire or emitted by street lights. If the light is emitted by street lights, the alarm is invalid.

In response to the existing problems, Shenzhen Power Supply Bureau released the first AI pre training model based on Ascension Ecology in the power industry, "Zhurong", early last year, laying the foundation for the transformation and upgrading of artificial intelligence in the power grid. At the end of last year, the bureau upgraded "Zhurong" to version 2.0 using a new artificial intelligence model similar to ChatGPT, enabling the power inspection system not only to read, record and analyze, but also to understand, reason, and express.

Zhurong 2.0 "is a multimodal model that utilizes both image and language models to extract and fuse image and text features, resulting in better performance than traditional image single modal models. The addition of language models enhances the logical reasoning ability of image models, thereby improving the accuracy of judgments.

Taking the scene of mountain fire monitoring in the power grid as an example, at night, the image forms of lighting and fire are very similar, and it is difficult to distinguish the two solely based on image models, which can easily lead to a large number of false positives; Now with the support of language models, the system will interpret the overall image. For example, when the model discovers that the light is neatly arranged and on one side of the road, it can be inferred that it is light, not mountain fire. Nowadays, the inspection system can transmit alarm information through images and text, and accurately describe hidden dangers and their degree of danger to power grid equipment, without the need for further manual investigation. This greatly improves the effectiveness of alarms and adds a strong line of defense for the safe operation of the power grid, "said Zhang Yunxiang.

Huge potential for further promotion and application

At present, single modal models are difficult to achieve good results in many scenarios in the power grid industry. For example, when climbing a ladder, it is necessary to check whether the operator is wearing a safety rope and whether someone is helping to support the ladder. The use of image models alone cannot make logical judgments on multiple simultaneous actions mentioned above. It is necessary to overlay language models and conduct comprehensive analysis in comparison with safety standards. In addition, in the field of marketing, the analysis of factors such as electricity demand also requires data from other modes such as current and voltage. Using data from different modalities to provide comprehensive judgments or contingency plans is the next development direction of "Zhurong 2.0".

It is reported that "Zhurong 2.0" can be further promoted in the field of power production and marketing.

At present, the Shenzhen Power Supply Bureau has completed the development of a multimodal pre training large model for scenarios such as transmission mountain fires and smoke, external damage hazards, and safety supervision violations. It is expected that 300000 invalid alarms can be reduced annually, saving 125 people a day's workload.

Shenzhen Power Supply Bureau has also collaborated with Yunnan Power Grid Information Center and Kunming Power Supply Bureau to carry out verification of transmission mountain fire smoke models, and has preliminarily applied this technology to patrol tasks such as transmission and transformation mountain fire smoke and external damage hazards in both regions, with a recognition accuracy rate of up to 98%.

In addition, the power supply bureau has established a joint research and cooperation mechanism with Yunnan Power Grid to accelerate the effective implementation of "Zhurong 2.0" in fields such as transmission, transformation and distribution, safety supervision, and marketing. Next, Shenzhen Power Supply Bureau plans to develop a pre trained large model that can distinguish sounds, adding new assistance to the investigation of external damage and other hidden dangers, and promoting the improvement of digital inspection efficiency in complex outdoor environments of the power grid.

Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])

Mobile advertising space rental

Tag: The Power version GPT increases the efficiency of safety