▶The MOIS has developed a model for predicting congestion on Seoul Metro and Gimpo Goldline subway platforms in real time.
▶The model detects crowd surge with 90% accuracy, allowing the authority to respond to the situation according to the manual.
▶After a pilot deployment for Seoul Metro stations in November, the ministry plans to standardize and deploy the model across Korea.

The Korean government has developed an artificial intelligence (AI)-based data analysis model capable of detecting congestion levels at subway platforms in real time. The pilot deployment will begin this November.

According to the Ministry of the Interior and Safety (Minister: Lee Sang-min, MOIS), the ministry has been developing the AI-based Subway Platform Congestion Prediction Model since June, using data from Seoul Metro and Gimpo Goldline stations as samples. The development process was recently completed, and the ministry will launch a pilot deployment across Seoul Metro stations.

The developed model calculates passenger density and congestion at subway platforms, taking account of the floor area of each platform and the number of passengers calculated using AI. With this information, the model rates the congestion of each platform on a four-level scale.
 
The model was jointly developed by the Integrated Data Analysis Center, Seoul Metro, and Gimpo Goldline, using around 8 million data, including subway entry/exit tag data, transport pass payment data, and train departure/arrival data.

The model calculates the number of passengers on each platform in real time by comprehensively considering passengers going through the gates in the station, passengers who went through the gates in the previous station, and the number of passengers who previously embarked/disembarked.

The developers conducted two performance tests, which rated the accuracy of the model at 90.1%. The congestion ratings from the developed model are being displayed on the dashboard of the Seoul Metro Electronic Control Room. Seoul Metro monitors the real-time congestion levels at two stations using this model.

The MOIS and Seoul Metro expect the model to facilitate an effective response to crowd surge at subway stations. The model is also expected to boost accident prevention at subway stations, as the scientific monitoring of congestion on subway platforms will allow the authority to proactively respond to situations based on congestion levels 

After a pilot deployment this year, the MOIS plans to standardize it and expand it to the subway station in the metropolitan area and four other cities such as Busan, Daegu, Gwangju, and Daejeon.

 

Source: Integrated Data Analysis Center, MOIS