Hyper-scale AI-Based Urban Rail Transit Safety GPT Service

Overview

For urban rail transit safety, accident prevention and proactive response are crucial, but the scattered nature of specialized materials increases the burden on safety managers and highlights their importance.

Accordingly, Seoul Metro aims to enhance the safety expertise of personnel and reduce decision- making time by enabling them to learn safety-related data, including not only relevant laws such as the Railway Safety Act, Transportation Safety Act, and Fire Safety Act, but also internal regulations and train manuals.

Project Description

The system has been trained on industrial safety-related laws and regulations, Seoul Metro internal regulations, safety operation manuals, website materials, and other relevant sources.

1. Data collection and preprocessing of training data.
2. Creation and fine-tuning of learning model.
3. Implementation of the Safety GPT service and integration/testing of learning models.

Key service features include Q&A support for internal operational manuals and guidelines, providing the latest safety regulations and legal basis for inquiries. The service is accessible as a web service, available anytime and anywhere.

Key technological features

• Domain Specialization: Development of a Semi-Automatic Domain Learning Data Generation Approach to Enable Reasoning in Expert Fields
• Knowledge Retrieval: Developing Context-Aware Chunking for Language Models to Understand

Expected outcomes

The Urban Rail Transit Safety GPT service enables field workers to receive accurate information related to inspection tasks, contributing to maintaining a safe rail environment and preventing safety accidents. Furthermore, the ability to easily access diverse and mixed information enables effective work support, leading to anticipated reductions in social costs. The project will be expanded to related agencies of the Transportation Corporation in the future. Chatbot-based data search screen Accident Site Notification

Launch year: 2023

Attribution

Case study courtesy of the National Information Society Agency, Republic of Korea, Apolitical's Content Partner. Sourced from Use Cases of Public AI Service, Vol.1,  https://eng.nia.or.kr/site/nia_eng/ex/bbs/View.do?cbIdx=31975&bcIdx=28997&parentSeq=29044