Forecasting of Electric Vehicles Charging Load Using Deep Learning Methods

Cadete, Eugenia
Journal Title
Journal ISSN
Volume Title

The objective of this research study was to create deep learning models to predict the energy consumption of 3 universities' charging stations given a forecasting horizon. The methodology section addresses the methods used to achieve such objectives. These methods are the RNN and LSTM. Both of these methods are based on deep learning neural networks. To evaluate the performance of these methods, four evaluation metrics will be applied: MAE, MSE, RMSE, and R2. Moreover, the analysis and findings section addresses how the methodology described above will be applied to the energy consumption datasets. The datasets will be analyzed, and the curve pattern will be studied. Based on this analysis, some filtering will be performed. The filtered datasets will be pre-processed by normalization and divided into input and output components. The experimental setup will focus on finding the model hyper-parameters using an algorithm and by trial and error. After finding the optimal hyper-parameters, the models will run 50 times. The loss function of the best model will be checked for convergence and then based on the evaluation metrics the model predictions will be plotted. Finally, the 1-step LSTM method performed better for the 3 datasets. However, as the timesteps increased, the RNN performed better.

This item is available only to currently enrolled UTSA students, faculty or staff.
Charging Load, Consumption, Deep Learning, Electric Vehicles, Forecasting, Prediction
Electrical and Computer Engineering