Predictive control for multi-market trade of aggregated demand response using a black box approach

Aggregated demand response for smart grid services is a growing field of interest especially for market participation. To minimize economic and network instability risks, flexibility characteristics such as shiftable capacity must be known. This is traditionally done using lower level, end user, device specifications. However, with these large numbers, having lower level information, has both privacy and computational limitations. Previous studies have shown that black box forecasting of shiftable capacity, using machine learning techniques, can be done accurately for a homogeneous cluster of heating devices. This paper validates the machine learning model for a heterogeneous virtual power plant. Further it applies this model to a control strategy to offer flexibility on an imbalance market while maintaining day ahead market obligations profitably. It is shown that using a black box approach 89% optimal economic performance is met. Further, by combining profits made on imbalance market and the day ahead costs, the overall monthly electricity costs are reduced 20%.