Hybrid model for short-term forecasting of loads and load control responses

A hybrid model for short-term forecasting of aggregated thermal loads and their load control responses is studied in this paper using field test data. Inputs include temperature measurement and forecast, measured power and control signals. The hybrid model comprises 1) partly physically based forecasting of the responses of the controlled thermal loads and the non-controlled power, and 2) forecasting the residual using Support Vector Machine (SVM). Their summation gives the hourly interval power forecast. Here partly physically based means that the model structure models thermal dynamics of houses. The response forecasting needs as inputs the daily heating energy demand and controllable power range, which are forecast by partly physically based models. Weekly load pattern is used to forecast the non-controllable power. SVM forecasts the residual. Alternative configurations are compared. The response forecasting model improved the forecasting accuracy most and adding the SVM for the residual further improved the accuracy.