Deep charge-fusion model: Advanced hybrid modelling for predicting electric vehicle charging patterns with socio-demographic considerations

This study examines electric vehicle (EV) charging behaviours across 5,898 participants from various socio-demographic backgrounds, using both statistical analysis and deep learning models. Exploratory data analysis (EDA) and Analysis of Variance (ANOVA) show that factors such as gender, education, employment status, household size, income, and housing type significantly influence charging behaviours at home, work, and fast-charging stations. Education notably impacts home charging (F = 2.54, p = 0.04), while household income strongly affects fast charging behaviour (F = 5.34, p = 0.001). A hybrid deep learning model, Deep Charge-Fusion, was developed by combining Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and eXtreme Gradient Boosting (XGBoost). CNNs capture spatial patterns for regional charging, while LSTMs model temporal dependencies in charging patterns. XGBoost adds efficiency by managing structured data and preventing overfitting through regularisation. Dropout layers were introduced in the LSTM network to reduce overfitting. The Deep Charge-Fusion model achieved an R² value of 0.81 for predicting fast-charging behaviours, outperforming standalone models such as Gated Recurrent Unit (GRU)-based (R² = 0.51) and LSTM-based (R² = 0.43) models. It also achieved an R² of 0.81 for home charging predictions. These results underscore the role of socio-demographic factors and demonstrate that hybrid models significantly improve predictive accuracy, with implications for EV infrastructure planning and energy management strategies.