Forecasting Electric Vehicle Supply Equipment Availability

The advent of electric driving poses novel scientific challenges. One such challenge is predicting the availability of electric vehicle supply equipment (EVSEs). Previous work addressing this question made use of insufficient data sources, limiting the available methods. In this paper, we make use of a much larger amount of data, opening the door to methods previously unused in this domain. The data used is especially suited for prediction using deep learning models, because of the high number of similar units of EVSEs present in the data. Specific deep learning architectures specialised in learning temporal dependencies are compared in their ability to predict the minutes of availability in a given hour for an EVSE. Long short-term memory and gated recurrent unit-based models are combined with a temporal convolution layer and compared in their performance on unseen periods and units. Of these models, the convolutional long short-term memory architecture was found to perform the best with a root mean squared error of 1.2 minutes per hour. However, we conclude that both architectures are similar in their performance, as both models were able to generalise well to unseen periods and to unseen EVSEs in those periods.