Federated Learning‐Based Deep Stacked Autoencoder—Long Short‐Term Memory Model for Electric Vehicle Energy Demand Prediction

ABSTRACT The rising demand for electric vehicles highlights the global push for green transportation. Predicting energy demand is crucial for managing electricity. This study presents a federated learning model using a deep stacked autoencoder‐based long short‐term memory for electric vehicles energy demand prediction. Initially, electric vehicles network nodes are simulated, and local training occurs at charging stations. Power grids are then upgraded, and model aggregation happens at the power grid. The enhanced training process iteratively predicts energy demand at charging stations using the deep stacked autoencoder‐based long short‐term memory. Here, the merging of deep stacked autoencoder and long short‐term memory designs a deep stacked autoencoder‐based long short‐term memory. The deep stacked autoencoder and long short‐term memory are combined using a fractional concept. Then, local updates and server accumulation are adjusted based on the average technique. The proposed model is analyzed using the metrics resource average, loss function, mean squared error, root mean squared error, false positive rate, mean average precision, computational efficiency, memory usage, and run time and obtained a value of 0.302, 0.090, 0.082, 0.287, 0.068, 0.921, 0.717, 2.684, and 5.139, respectively. The proposed model helps to reduce the cost associated with energy production, distribution, and maintenance by optimizing charging schedules and energy use.