Machine Learning methods for clustering and day-ahead thermal load forecasting of an existing District Heating

This work aims at providing a District Heating day-ahead energy demand forecasting method based on unsupervised clustering and artificial neural networks. The results of three different clustering methods are compared: k-means, Hierarchical Agglomerative Clustering and Density Based Spatial Clustering of Applications with Noise. According to the indices used to assess the goodness of clustering, the three clustering methods provided comparable results. Finally, in order to predict the required thermal energy consumption of the District Heating thermal plants, three different strategies training a single neural network are adopted, namely: for all utilities, for each cluster and for each utility. The last strategy showed better results, but not far from the second strategy, which could be effective if data and the number of utilities increase. The analysis is performed on the variables of the thermal plants and weather parameters, recorded in an existing District Heating in the North of Italy.