This report presents a study of models for forecasting the load for supermarket refrigeration. The data used for building the forecasting models consists of load measurements, local climate measurements and weather forecasts. The load measurements are from a supermarket located in a village in Denmark. The load for refrigeration is the sum of all cabinets in the supermarket, both low and medium temperature cabinets, and spans a period of one year. As input to the forecasting models the ambient temperature observed near the supermarket together with weather forecasts are used. Every hour the hourly load for refrigeration for the following 42 hours is forecasted. The forecast models are adaptive linear time-series models which are fitted with a computationally efficient recursive least squares scheme. The dynamic relations between the inputs and the load is modeled by simple transfer functions. The system operates in two regimes: one in the closing hours during night and one in the opening hours during the day. This is modeled by a regime switching model in which some of the coefficients in the model depends on the regime. The results show that the one-step ahead residuals are close to white noise, however some dependence on the ambient temperature remains, which is caused by non-linearities in the relation between the load and the ambient temperature. Suggestions for including these non-linearities are given in the discussion of the results.
The report starts with a section in which the data and the NWPs are described. This is followed by a presentation of the modeling approach and the model identification, where a suitable forecasting model is found. Finally, the results are presented, and the method is discussed and conclusions are drawn.
|Number of pages||25|
|Publication status||Published - 2013|
|Series||D T U Compute. Technical Report|