High spatial granularity residential heating load forecast based on Dendrite net model

Abstract

With the application of smart meters, more information is available from residential buildings for support heat load forecast. Yet, there is still a lack of an effective method to exploit the value of the high spatial granularity information, particularly for residential communities with high randomness in human behaviors. To fill this gap, this paper proposes a data-driven heat load forecast method based on the field measurements of smart meters A white-box machine learning algorithm, namely Dendritic Network, is employed to aggregate and analyze data obtained from different locations of district heating systems. An online correction mechanism is then proposed to regulate the forecast horizon and ensure the adaptiveness of the machine learning model under different scenarios. The results demonstrate that the dendritic network employed in the proposed machine learning method shows higher forecast accuracy, the root-mean-square error is 0.0029, in some research cases the coefficient of determination can reach 0.99–1, and wide scope of application in the area of heat load forecast.