A new technique for short-term residential electric load forecasting including weather and lifestyle influences

A stochastic physically based residential load forecasting technique that accounts for random influences on the load is proposed. The proposed forecasting approach is based on forecasting the weather-dependent and weather-independent components separately. The weather-dependent component is modeled via a stochastic physically based model. From the physically based model, the maximum likelihood duty cycle is deduced and used to predict the weather-dependent component. The weather-independent component is modeled by an autoregressive (AR) model. Addition of the two predictions yields the total residential load forecast. This algorithm overcomes the problem of obtaining the weather-load relationship that has placed most existing forecasting methods at a disadvantage. From the data gathered in the Athens load control experiment, one set of data is used to estimate the parameters of the proposed load prediction method. Other sets of data are used to verify the prediction and provide a measure of the success of the proposed forecasting technique. Very good correlation is observed between the prediction and measurement