Forecasting the consumption evolution of battery electric vehicles under dynamic market conditions: The case study of Jiangsu Province, China

Predicting the future market size of battery electric vehicles (BEVs) and their market share is essential for analyzing transport externalizations and optimizing charging infrastructure deployment. Current smooth-curve models, the system dynamics, and agent-based models for BEV market forecasting are usually static functions or rely on market interactions. Still, they hardly quantify the influencing effects and changes of covariates under dynamic market conditions. Given the above-mentioned, the BEV cumulative sales are forecasted under dynamic market conditions using the artificial neural network and the bidirectional short- and long-term memory models. The samples of five covariates are derived from available data about BEV sales, price changes, fuel-to-electricity ratio, charging piles, driving range, and incentive effects from the priorities of BEV license plates in Jiangsu province. Different evolutionary analyses are set the three future scenarios of the BEV sale market based on the Time-Series Multi-Layer Perceptron model, and the marginal effect of a single covariate was further analyzed. Finally, our results show the advantages of machine-learning methods over smooth-curve models used to generate market predictions, further providing insights on covariates effects for market managers to promote the BEV sale market.