An Attention-Seq2Seq Model for New Energy Vehicle Sales Prediction

With worsening energy and environmental issues, new energy vehicles (NEVs) have emerged as the future of the automotive industry, as they aim to address the high energy consumption and carbon emissions of traditional fuel vehicles. However, due to the industry’s short development history, limited available data, and incomplete supporting systems, most existing NEV research focuses on theoretical analysis, which hinders the achievement of accurate sales predictions. Today, online reviews influence consumer decisions and thus provide a new perspective for sales forecasting. Based on consumer behavior theory and neural network principles, our research selects factors influencing NEV sales (covering economics, technological, policy, and consumer dimensions, including preprocessed crawled online reviews), constructs an index system screened via grey relational analysis, and establishes five models (SARIMA, GRU, Seq2Seq, Attention-GRU, Attention-Seq2Seq) for training and testing. The study supports the use of online reviews in NEV sales prediction and proves that the model based on cutting-edge technology of Attention-Seq2Seq can outperform the other four methods presented above. Through this, the current contributions advance marketing innovation by helping NEV stakeholders understand relevant information using a predictive model from online reviews, which leads to precise product improvement and optimal distribution of resources as well as precise adoption of marketing strategies.