Deep learning-based long-medium prediction of Chinese new energy vehicle sales and air quality towards 2035

Abstract

Accurately forecasting new energy vehicle (NEV) sales carries significant policy and strategic implications. This study proposes a deep learning framework to assess policy incentive intensity for NEV and employs natural language processing techniques to quantify media sentiment index. A deep learning algorithm named P-PLSTM is developed, and Monte Carlo simulation and scenario analysis are embedded, constructing a dynamic scenario forecasting model. It predicts the dynamic development trajectories of Chinese NEV sales and the corresponding air quality between 2024 and 2035 under different scenarios. The results indicate that, compared to traditional machine learning models and commonly used LSTM-based deep learning models, the P-PLSTM model achieves higher predictive accuracy for Chinese new energy vehicle sales, with an R² of 0.98. Under current conditions, NEV sales exhibit a steady upward trend, surpassing 15 million units by 2030 and reaching 21.29 (±2.87) million units by 2035. With the increase in NEV promotion, the concentrations of major pollutants exhibit a declining trend. By 2035, the atmospheric concentrations of NO₂ and PM_2.5 are expected to decrease by 10.38 and 7.17 %, respectively.