Predicting solar radiation for space heating with thermal storage system based on temporal convolutional network-attention model

Abstract

Multi-source heating systems have great potential for energy-saving and renewable energy source integration in buildings. The flexibility of electrical heating and thermal energy storage (TES) devices can contribute to overcoming the challenges caused by increasing shares of renewable energy sources in the heating system. However, from the perspective of system thermal management, the aleatory and intermittency of renewable energy sources are still detrimental to heating system reliability. Therefore, this study proposes a novel thermal storage control strategy that considers solar energy uncertainty to improve the operation of a space-heating system integrated with solar energy, electric boiler, and thermal storage system. Further, it is imperative to investigate effective methods to predict solar radiation. To improve the accuracy of short-term prediction for solar radiation, this paper proposed a novel temporal convolutional network model based on the attention mechanism (TCN-Attention) as a prerequisite for formulating optimal system control strategies. The TCN-Attention algorithm is conducive to expressing and modeling nonlinear characteristics of solar radiation data, which provides optimal solar energy estimates. Detailed experiments show that the proposed TCN-Attention model achieves a prediction accuracy of 45.07 W/m2 of root mean square error (RMSE) and 0.922 of Nash-Sutcliffe efficiency coefficient (NSE). The predictive performance of TCN-Attention is superior to other compared algorithms such as the recurrent neural network (RNN), long-short term memory (LSTM), gated recurrent unit (GRU), etc. In addition, the implementation results of the predictive control strategy based on the TCN-Attention model in an actual building for space heating (SH) were presented, showing the potential for cost-saving.