V2G Demand Prediction Based on Daily Pattern Clustering and Artificial Neural Networks

This paper presents how to manage the power consumption history in a microgrid, clusters days according to their time series patterns, and develops a prediction model for next day demand. Daily consumption patterns, each of which consists of quarter-hourly records, are grouped into 6 clusters, taking advantage of the dynamic time warping method in measuring the similarity between all feasible pairs of days. We select 3 main parameters for the cluster prediction of the next day, namely, month, day-of-week, and day-high temperature given by the weather forecast. For machine learning, learning patterns are generated after joining tables of power consumption, weather archive and day-group association on a daily basis. The next step builds an artificial neural network model using well-known open software. The model shows the accuracy of 67%, making it possible to estimate next day behavior, select the best demand model, and estimate power demand for vehicle-to-grid trades.