A novel dual-scale ensemble learning paradigm with error correction for predicting daily ozone concentration based on multi-decomposition process and intelligent algorithm optimization, and its application in heavily polluted regions of China

Abstract

Accurate prediction of daily ozone concentration is imperative to prevent and control photochemical pollution in China. A novel ensemble learning paradigm is proposed in this paper to perform high-precision ozone concentrations prediction, including model preprocess, optimized dual-scale prediction, and error correction. Firstly, the original ozone series is decomposed into two detailed sequences and two approximate sequences by wavelet packet decomposition (WPD). The detailed sequences are reconstructed into several new subsequences by complementary ensemble empirical mode decomposition (CEEMD) and fuzzy entropy (FE). Secondly, the extreme learning machine (ELM) and the support vector machine (SVM) are both optimized by the bald eagle search (BES) algorithm to carry out dual-scale parallel prediction. Finally, the error series is decomposed and predicted by the variational mode decomposition (VMD) and the optimized ELM to correct the previously predicted ozone and obtain the final ozone prediction. The actual ozone data from two typical cities in China are collected as the inputs of the model for empirical analysis under two scenarios, and one-day ahead prediction results show that: the RMSE values of the proposed model are 2.5319 and 3.2069 at Taiyuan and Shanghai sites when predicting low levels of ozone, respectively, while the RMSE values of the proposed model are 2.8451 and 3.8702 at two sites when predicting high levels of ozone, respectively; the proposed model outperforms the comparison models and four existing models under each scenario, demonstrating its robustness and serviceability. The results of this study can provide a valuable reference for analyzing the tendency of pollution.