More than half of the energy consumption in European residential building is utilized for
space heating, which raises the importance of accurate heating demand quantification.
Several combinations of modern machine learning approach and conventional Degree
Days approach were analyzed in this research, focusing on the prediction accuracy and
flexibility. Additionally, a set of different training datasets was also analyzed, examining
the influence of multi-location building cases and external weather variable to the prediction accuracy. The obtained result indicates that indirect heating demand quantification
through predicting the building base temperature has the highest accuracy. Moreover,
utilizing multi-location training dataset with external climate inclusion further enhances
the prediction result, creating an accurate and stable prediction behavior. However, single-location algorithm has a higher prediction accuracy on certain testing location, which
makes it more suitable for local prediction model. Thus, the final selection of the best
quantification model is greatly dependent on the application scope.
Collections
Show Collections
