A Socio-Climatic Deep Learning Framework for EV Charging Infrastructure Demand and Operational Stress Prediction under Extreme Arid Climates

Abstract

There is a rise in electric vehicles (EVs) causing several problems related to the grid, particularly in places with high ambient temperatures. Although the current models were designed for temperate regions, the accuracy of the forecast models is significantly impacted in extreme environments such as those in the Middle East region. The core contribution of the study is in the innovative integration of the multi-source NASA POWER satellite-based hydro-thermal stress parameters in a deep neural prediction pipeline for managing extreme temperature electric vehicle infrastructure in arid regions. In order to address this challenge, the study proposes to build a socio-climatic deep learning model that predicts the usage of EV charging stations in environmental distress, using Iraq as a relevant country in the Middle Eastern region. Using the socio-climatic model, combining 20,569 transaction records along with the NASA POWER climate data helps to define the relationship between charging station usage and heat constraints. According to the results of the experiment, it has been found that due to the application of the enhanced framework, the issue of convergence was successfully avoided, and prediction accuracy increased from 56.8% to 91.68%. Moreover, as a result of conditional probability simulations, the proactivity of the system was proven as there was strong load reversal, meaning that probability of "High Demand" increased from 23.5% to 72.8% under peak intensity.