마코브 디시젼 프로세스 기반 최적 공유 Personal Mobility 배치지역 선정

초록

The proper deployment of shared personal mobility (PM) can increase mobility for residents who have difficulty traveling short distances to their destinations. This study utilizes Geographic Information System (GIS) and reinforcement learning to find the optimal locations for shared PMs. This study focused on Seoul, a city with open source data and a large population and PM usage. Among the data provided by the Seoul Open Data Plaza, we used public transportation-related data, socioeconomic indicators, road and terrain data, and shared mobility data. The data was inputted after subdividing the area to be analyzed into a hexagonal grid structure, which minimizes decision making for computation. The reinforcement learning structure is composed of the commonly used Markov decision process (MDP). The MDP is based on Open AI GYM, which can build a simple reinforcement learning environment. The reward function of the MDP is designed so that the smaller the distance required for PM placement, the higher the reward. The optimization of reinforcement learning was performed through the Q-learning algorithm, which is simple and applicable in various environments. Every hour, PM users generated from random points were moved according to the collected shared mobility start and end points, and the predicted PM deployment areas were visualized using QGIS. The analysis showed that the predicted PM deployment areas are Hwayang Intersection and Gasan Digital Complex, which are predicted to have a large number of commuters near industrial areas. In addition, Jamsil Station and Magoknaru Station, which are predicted to use PM for leisure purposes, are predicted to need PM deployment. The results of this study can improve the operational efficiency of service providers and the mobility of service users, and can help policy developers prepare legal grounds.

키워드