Machine learning method for day classification to understand thermostatically controlled load demand

An accurate estimation of the thermostatically controlled loads (TCLs) demand is important for the load control to provide demand-side ancillary services, such as peak- demand reduction. CSIRO developed a mathematical model that provides the aggregate demand of a population of TCLs for a given set of model parameters and a given ambient temperature profile. The model parameters, however, need to be identified (estimated) from experimental data. By recognizing the importance of accurate and efficient identification of model parameters, in this paper, we developed a machine learning methodology using support vector machines (SVMs) to classify days into “hot”, “cold” or “mild” days using very limited aggregated information. This classification is the most important part of the identification and is based on experimental evidence combined with ambient temperature readings and aggregate TCL demand. . The simulation results on real datasets across Australia indicate over 88% accuracy of successful classification.