A hybrid energy-saving prediction model based on SSA-DNN for district heating system

Abstract

ABSTRACT Accurate heat load prediction is a prerequisite for feed-forward control and on-demand heat supply in district heating system. However, considering that the experimental data used to train the prediction model are often not optimal or most energy efficient, accurate prediction is difficult to achieve effective energy-saving. This paper proposes a hybrid energy-saving prediction model that combines similar sample selection approach (SSA) and deep neural network. A new weighted Euclidean norm (WEN) is used to select suitable similar sample datasets, and a novel energy-saving strategy is proposed to reduce energy consumption. To make the prediction performance more stable, a low-pass filter is used to filter the prediction results. In the case study, real data from a heat exchange station in Tianjin are used to verify the prediction performance of the hybrid model for 1 test day, 3 test days, and 7 test days. Experimental results show that: (a) the proposed model is able to capture the change trend of heat load, with Pearson correlation coefficient of 0.971, 0.969, and 0.954 on different test days, respectively; (b) the proposed model is able to effectively reduce energy consumption, with energy-saving of 5.4%, 7.6%, and 4.8% on different test days, respectively.