Experimental Validation of a Probabilistic Deep Learning Framework for Electric Vehicle Charging Behaviour Prediction

Abstract

The growing penetration of plug-in electric cars (EVs) is altering the transportation and electrical infrastructures, leading to highly variable and stochastic charging demands. An experiment is proposed in this article to experimentally test an electric vehicle-charging patterns (EV-CP) probabilistic deep learning architecture developed to forecast automotive uptake. The proposed method simulates the non-linear sequential interactions by using historical charging-session data, driver behaviour, time and other variables important to EV charging. We apply deep-learning techniques to improve the predicting capabilities. We also employ probabilistic forecasting to measure uncertainty and produce a range of potential future pricing pathways, not just single deterministic projections. Methodological Framework for Data Extraction, Preparation, and Targeted Sampling for the Development of Supervised Deep Learning (DL-SD), Probabilistic Uncertainty Assessment, and Experimental Validation Using Separate Testing Data Sets The evaluation of the model is done using probabilistic metrics like Prediction Interval Coverage Probability (PICP) and Prediction Interval Width (PIW) and point-forecast metrics like Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The statistical test results are mostly positive: high positive correlations between Probabilistic Deep Learning Performance (b =. 672, p < 0. 001). The realised benefits from UP and UCE are small. The illustrated regression model supports all three hypothesised bivariate relationships (R = 0.846, R² = 0.716). The suggested method demonstrates how deep learning and probabilistic uncertainty estimation may result in more reliable and valuable EV charging.