Data Requirements for a Reliable Demand Decomposition in Sparsely Monitored Power Networks

Abstract

This paper discusses data requirements for an efficient demand decomposition at the aggregation level considering a limited number of monitoring points. Two methods are compared: an artificial neural network (ANN) based method and the autoregressive integrated moving average (ARIMA) method, followed by the validation of the superior approach against the data coming from an actual pilot site. The influence of data types, such as the weather and type of day, is investigated, as well as the size of the historical data required. The analysis concludes that the ANN based approach is superior, and that using appropriately trained ANN, even with only 5% of end-users whose per-appliance consumption is being monitored, it is possible to estimate or predict, with high accuracy, the demand composition of the overall aggregation of users.