A novel forecasting algorithm for electric vehicle charging stations

Abstract

In this paper, a recently proposed time series forecasting algorithm, Modified Pattern-based Sequence Forecasting (MPSF), is compared with three other algorithms. These algorithms have been applied to predict energy consumption at individual EV charging outlets using real world data from the UCLA campus. Two of these algorithms, namely MPSF and k-Nearest Neighbor (kNN), are relatively fast and structurally less complex. The other two, Support Vector Regression (SVR) and Random Forest (RF), are more complex and hence require more time to generate the forecast. Out of these four algorithms, kNN with k=1 turns out to be the fastest, MPSF and SVR were the most accurate with respect to different error measures, and RF provides us with an importance computing scheme for our input variables. Selecting the appropriate algorithm for an application depends on the tradeoff between accuracy and computational time; however, considering all factors together (two different error measures and algorithm speed), MPSF gives reasonably accurate predictions with much less computations than NN, SVR and RF for our application.