A Transformer Attention‐Based Convolutional Network for Solar <scp>PV</scp> Integrated <scp>EV</scp> Charging Demand Prediction

ABSTRACT Accurate prediction of EV charging demand is crucial in developing solar photovoltaic integrated EV charging systems. In this regard, intermittency and fluctuation from solar PV systems are added to the process used in predicting charging demand by EVs. Integrating renewable energy sources into EV charging stations requires advanced models, which can account for variable energy production, consumption patterns, and external factors like weather and time of day. While there are improvements in the present models, most of them forecast the charging demand poorly for systems dependent on renewable energy sources. In consideration of such challenges, this paper proposes a new method combining Models of TACNet and the CSLS optimization technique for solar PV‐integrated EV charging demand prediction: the first one is called Transformer Attention‐based Convolutional Memory Network (TACNet), which combines both the transformer's attention mechanisms and convolutional memory networks in extracting long‐term dependencies and local features of the time‐series data. The second one is Chaotic Slime Levy Search (CSLS) optimization, applied for learning rate optimization during model training using a chaotic Levy search approach to improve training efficiency and speed up convergence. To the best of our knowledge, this is the first study to integrate TACNet with CSLS to ensure the model performs optimally using advanced forecasting techniques and efficient optimization. The attention‐based mechanism in TACNet could learn the complicated patterns present in charging demand data. On the other hand, CSLS improves the optimization process by avoiding local minima and speeding up convergence. The proposed methods give a way to improve efficiency in solar PV‐integrated EV charging stations, reduce dependence on the grid, and increase the use of renewable energy in transportation infrastructure by way of accurate demand prediction for EV charging and optimization of energy use. According to the findings, it is observed that the proposed model provides 98.4% of efficiency, 99% of accuracy with low computational time of 5.9 s.