The widespread deployment of Advanced Metering Infrastructure has madegranular data of residential electricity consumption available on a largescale. Smart meters enable a two way communication between residentialcustomers and utilities. One field of research that relies on such granularconsumption data is Residential Demand Response, where individual users areincentivized to temporarily reduce their consumption during periods of highmarginal cost of electricity. To quantify the economic potential of ResidentialDemand Response, it is important to estimate the reductions during DemandResponse hours, taking into account the heterogeneity of electricity users. Inthis paper, we incorporate latent variables representing behavioral archetypesof electricity users into the process of short term load forecasting withMachine Learning methods, thereby differentiating between varying levels ofenergy consumption. The latent variables are constructed by fitting ConditionalMixture Models of Linear Regressions and Hidden Markov Models on smart meterreadings of a Residential Demand Response program in the western United States.We observe a notable increase in the accuracy of short term load forecastscompared to the case without latent variables. We then estimate the reductionsduring Demand Response events conditional on the latent variables, and discovera higher DR reduction among users with automated smart home devices compared tothose without.
展开▼
