Prediction of Chaotic Time Series Based on SALR Model with Its Application on Heating Load Prediction

Abstract

With the development of nonlinear science, improving the prediction performance of chaotic time series is of great significance in industrial production and daily life. Now, researchers have to develop effective models to achieve accurate prediction performance. The echo state network (ESN) has been proven to be an excellent prediction tool. However, the ESN has been criticized for not being principled enough. Thus, a novel ESN model namely self-join adjacent-feedback loop reservoir (SALR) is proposed. This model achieves the simplest topology structure on the premise of ensuring that all the connection modes of the classic ESN are available. In addition, in order to ensure the prediction performance of the network, the whale optimization algorithm was used to solve the parameter selection problems in the traditional cycle reservoir (SCR) model, the adjacent-feedback loop reservoir (ALR) model, and the SALR model. Finally, we use the proposed SALR model to solve classic benchmark chaotic time series as well as practical heating load prediction problems, and compare the SALR with the ESN, SCR, and ALR, respectively. Experimental results show that the proposed model can obtain higher accuracy with relatively low complexity than the ESN, SCR, and ALR.