On the V2G capacity of shared electric vehicles and its forecasting through MAML‐CNN‐LSTM‐Attention algorithm

Abstract

Abstract Shared electric vehicles (shared EVs) could be treated as energy storage and participate in electricity market services. In that case, the schedulable capacity that shared EV provide to grid in future time needs to be predicted accurately. Based on open data provided by a shared vehicle rental project, this research firstly proposes a method to build up schedulable capacity data set to prepare sample data for the deep learning‐based forecasting task. Secondly, a schedulable capacity forecasting model by MAML‐CNN‐LSTM‐Attention is proposed. Through the model, ultra‐short prediction (1 h) of aggregated schedulable vehicle‐to‐grid capacity of shared EVs is performed. The proposed model uses model‐agnostic meta‐learning (MAML) to optimize initial parameters of network to quickly adapt to changes in characteristics caused by different travel habits of different functional communities; uses two‐layer convolutional neural network (CNN) with long short‐term memory neural network (LSTM) and Attention mechanism to extract temporal‐spatial features of schedulable capacity for important historical moments. Simulation results show that the proposed model has a mean absolute percentage error (MAPE) of 1.13% and goodness of fit of 0.976, the MAML decreases the MAPE by 1.04%, its stability and accuracy were better than previous used models. The research can act as useful decision aids for shared EV operators to participate in market services.

Keywords

• attention mechanism
• convolutional neural network
• demand response
• long short‐term memory neural network
• model‐agnostic meta‐learning
• shared electric vehicles