SHORT-TERM LOAD FORECASTING USING ARTIFICIAL NEURAL NETWORKS AND SOCIAL MEDIA DATA

Abstract

Evolving practices around energy generation, storage and trading within the UK have made it more necessary than ever to provide accurate means of forecasting electricity demand. This paper considers deep neural networks with convolutional and recurrent layers to investigate the inclusion of various data types as inputs to a load forecasting model, by evaluating 24-hour ahead predictions of electricity demand. Using two case studies in Durham, UK, this paper evaluates the benefits of including temporal and meteorological data, and proposes a novel approach to incorporating social media data to a load forecasting model. Performance is assessed using traditional measures of Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE), whilst also considering the standard deviation across repeats. Results indicate that Twitter data containing the number of tweets matching a specific query is capable of improving forecasting accuracy for large-scale residential loads.