Daily residential heat load prediction based on a hybrid model of signal processing, econometric model, and support vector regression

•The VMD method decomposite the heat load and extract different components.•Components' independence is achieved based on the FastICA transformation.•The IGARCH and SVR model is adopted to predict the different components.•Detailed experiments' results validate the model's higher prediction accuracy. The heat load prediction algorithm is essential for achieving optimal control in the smart district heating system (SDHS). However, it is challenging to meet the accuracy requirement due to the complex nonlinear heat load characteristics. The traditional forecasting methods ignore the divergence between different heat load compositions. Based on the divide and conquer strategy, the prediction accuracy may be improved by decomposing the heat load into several relatively simple sequences. A novel hybrid model is proposed to predict the daily heat load of households based on the combination of signal processing technology, econometric models, and support vector regression (SVR), which is called VMD-FastICA-SVR-IGARCH. Firstly, the original heat load time series is decomposed to obtain different components based on the variational mode decomposition (VMD) and the fast independent component analysis (FastICA). Then, the integer generalized autoregressive conditional heteroscedasticity model (IGARCH) and SVR model are used to predict high-frequency and other components, respectively. Compared with the state-of-the-art prediction methods, the proposed hybrid model has more significant advantages in the prediction performance of residential heat load data sets, as well as good generalization ability and robustness.