The impact of increased information content on electricity load forecasting

Electricity load forecasting of private households usually aims at predicting the household’s total load by applying artificial intelligence methods to data measured at the household’s grid connection point. This is a challenging task, as the parameter of interest is influenced by climate, seasonal and human factors. With the rising number of implemented digital metering systems, future household data promises to not only contain the total load, but also to have an increased granularity, even covering the load of single household devices. This analysis investigates the extent to which the knowledge of the per-device load will improve the residential load forecasting quality by training an artificial neural network (ANN) on the data taken at the grid connection point and comparing its results to the collective output of ANNs trained on high-granularity data containing every individual electricity load in the household.