Data-Driven Approach to Forecast Heat Consumption of Buildings with High-Priority Weather Data

By increasing the penetration of renewable energies in district heating (DH), the intermittency of the supply-side increases for heating service providers. Therefore, forecasting the energy consumption of buildings is needed in order to hedge against renewable power intermittency. This paper investigates the application of data-driven approaches to forecast the heat consumption of buildings in the winter, using high-priority weather data. The residential buildings are connected to mixing loops of DH to supply space heating and hot water. The heating consumption of the building is calculated using sensor data, including inflow/outflow temperature and mass flow. Principal component analysis (PCA) is applied to determine the key weather data affecting heat energy consumption. Then, the study compares the competences of artificial neural networks (ANNs), linear regression models (LRM), and k-nearest neighbors (k-NN) in forecasting heat consumption, using informative data. Based on the PCA analysis, ambient temperature and solar irradiation are shown to be the highest priority weather data, contributing to 40.6% and 29.2% of heat energy forecasting, respectively. Furthermore, the ANN exhibits a forecasting accuracy of more than 50% higher than LRM and k-NN.