Smart Grid: A Bottom-Up Approach for Consumer Demand Forecasts

Playing a delicate game of balancing grid load, utility operations are increasingly dependent on accurate forecasts of both grid production and consumer demand. Aggressive moves towards integration of weather-dependent renewable energies has exposed vulnerabilities of the current infrastructure to forecast skill – at best, an incorrect forecast results in unplugging solar panels from the infrastructure, at worst, rolling blackouts. With modernization of the grid, a larger integration of variable renewable sources can be assumed, as can a transition from unilateral flow of energy from producing facility to consumer to a bilateral system in which localized consumer production (i.e., rooftop photovoltaic production) plays a significant role. Here we present a bottom-up approach for estimation of one side of the energy equation – consumer demand. Our approach leverages open source data including tax lot information, Census data, TMY3 climatology and NREL building energy consumption simulations to build a geo-located simulation at the address level and on the temporal scale of hours. Consumption patterns were “learned” for building types of varying purpose, climate and demographic zones as a function of imposing weather conditions. The resulting simulation for a given weather forecast is built up at the address level, with each individual building energy profile forecasted via the previously trained model, and knowledge of present structures given by tax records.