Edge Computing Based Coordinated Energy-saving Train Regulation with Multi-agent Learning

Regenerative braking technology shows great potential in energy conservation and plays an increasingly important role in urban rail transit industry. In order to enhance the operation synergy among multiple trains to make full use of regenerative braking energy and reduce overall energy consumption, train schedule regulation needs to take driving strategy as a whole into account, which asks for higher requirements for the real-time adaptability of urban rail train control system. Based on the analysis of metro train big data environment, this paper puts forward an edge computing architecture combined with multi-agent actor-critic (EC-MAAC) model for real-time train schedule regulation considering energy saving. The deployed cloud server adopts multi-agent actor-critic network for parallel offline training. Each edge server downloads the multi-agent model trained in the cloud to the local for online traffic regulation. The experiments show the improvement of regenerative braking energy using among multiple trains through the real-time adjustment of inter station driving time, dwell time and traction-braking strategy.