Forecasting China’s charging-pile demand by an improved Bayesian grey model with sparse structural identification

Accurate forecasting of China’s charging-pile demand is challenging due to limited historical data, rapid market evolution, and strong policy-driven uncertainty. To address these issues, this study proposes an Improved Bayesian Grey Model (IBGM), which integrates structural identification and parameter inference within a unified probabilistic framework. By introducing adjustable dictionary functions, a unified structural expansion approach applicable to diverse grey model structures is established. A Bayesian pruning strategy is further proposed to quantitatively evaluate the contributions of candidate structural terms based on posterior probabilities, thereby promoting grey modeling from experience-based specification toward statistical inference-driven construction. In addition, a structural length constraint parameter is introduced to regulate model complexity, achieving an effective balance among forecasting accuracy, model sparsity, and interpretability. An empirical study on China’s charging-pile industry demonstrates that the proposed IBGM achieves superior forecasting accuracy and structural adaptability compared with existing benchmark models. Forecasting results indicate that the total number of charging piles in China is expected to reach approximately 29 million by 2027, representing an increase of about 40% compared with 2025, highlighting the urgent need for coordinated infrastructure investment and intelligent energy management.