Short-term Electric Vehicle Charging Load forecasting using Transfer and Meta-learning

Load forecasting for electric vehicle (EV) charging hubs has emerged as a crucial task for demand-side management and planning of electric grid frameworks with the increasing number of EVs on road. Existing load forecasting methods like long short-term memory (LSTM) neural networks rely on large amounts of historic data which can be expensive, difficult, and time-consuming to collect in some circumstances. In this work, we propose using Transfer Learning (TL) and Model-Agnostic Meta-Learning (MAML) for short-term EV charging load forecasting by utilizing data from just 10 or 20 days. This is achieved by pretraining models on a larger EV charging dataset from Boulder, Colorado and fine-tuning the models on 10 and 20day subsets of EV charging data from Pasadena, California and Trondheim, Norway. We compare the accuracy of TL and MAML models with traditional LSTM models on their ability to make accurate forecasts using extremely small datasets across different geographies and scopes.