Utilising numerical weather forecast for planning electricity production in cogeneration plant

Production of heat and electricity in the cogeneration plant depends on weather, thus forecasting production is dependent on weather forecasts. Here we present the models of the heat production based on two weather forecast models, COAMPS and UM. The linear models that are based on the predicted air temperature can explain up to 90% of variability of production and deteriorate slowly with the range of forecast. The models of heat productions that are based on UM weather forecasts significantly outperforms those that are based on the models based on the COAMPS weather forecasts. The machine learning algorithm random forest is used to improve the basic models. To this end the residuals from the linear models are predicted using various meteorological variables along with variables governing activity of city inhabitants, such as hour of the day or day of the week. This machine learning approach leads to small but significant improvement in comparison to the original model.