Energy Demand Load Forecasting for Electric Vehicle Charging Stations Network Based on ConvLSTM and BiConvLSTM Architectures

The electrification of transport has proved to be a breakthrough to uplift the sustainable and eco-friendly platform in the global sector in which electric vehicles (EVs) are considered indispensable. In particular, creating intelligent energy management in the power distribution system integrated with electric vehicle charging stations (EVCS) as a new entity is one of the most important challenging tasks. The implementation of the EVCS network infrastructure should facilitate the adoption of the spatiotemporal electricity demand for EVs. The intelligent decision for the transmission, distribution, energy allocation and charging station placement by the control center or central aggregator is only possible by correctly forecasting its usage, occupancy, and energy or charging demand. Techniques like data analytics have enabled to extract data from the EVCS on a daily basis to store and process all the recorded data. To overcome the above-mentioned challenges related to energy demand forecasting for EVCS network, this work proposes two encoder-decoder models based on convolutional long short-term memory networks (ConvLSTM) and bidirectional ConvLSTM (BiConvLSTM) in combination with the standard long short-term memory (LSTM) network. Data on energy demand from EVCS located in four different cities is used in the proposed models. All datasets are preprocessed to make them suitable for the multi-step time-series learning models in order to make the framework data-centric. The suggested architectures are built on the ConvLSTM and BiConvLSTM to extract the key features from the spatiotemporal data of the energy demand data of the EVCS distributed over the time and space. The predicted outcomes generated by the suggested strategy are compared with conventional deep learning models and traditional machine learning techniques.