Linear and Neural Network-based Models for Short-Term Heat Load Forecasting

摘要: Successful operation of a district heating system requires optimal scheduling resources to satisfy demands. The operation, therefore, accurate short-term forecasts future heat load. In this paper, forecasting load in Ljubljana is presented. Heat data and weather-related influential variables for five subsequent winter seasons are applied study. Various linear models nonlinear neural network-based developed forecast the daily with horizon one day ahead. evaluated based on generalization error, obtained an independent test set. Results demonstrate importance outdoor temperature as most important variable. Other inputs include solar radiation extracted features denoting population activities (such week). Comparison reveals good performance stepwise regression model (SR) that utilizes only subset relevant input variables. SR was improved by using network (NN) models, also NN direct link (NNLL). latter showed overall best performance, which suggests or proposed NNLL structures should be considered solutions markets.