Distributed V2G Dispatching via LSTM Network within Cloud-Edge Collaboration Framework

Abstract

The bidirectional energy flow between plug-in electric vehicles (PEVs) and power grids enables load flatting and self-consumption of the photovoltaic (PV) output. However, two critical issues should be addressed. One is how to conduct the gap between the decision makings optimized with predictive data and the reality, and the other is how to ensure efficiency of V2G dispatching. In order to tackle these problems, this work proposes a distributed V2G dispatching via long short term memory (LSTM) network within cloud-edge collaboration framework. In the cloud side, the LSTM network is applied merely utilizing the present data to obtain the prediction models of V2G dispatching. Then, these models are sent to the edge side and updated in a regular time. In edge side, the distributed dispatching is conducted to decrease the computational complexity. The proposed framework is verified by numerical analysis, which illustrates that the effectiveness, efficiency and applicability of the V2G operation.