CEEMD-MR-hybrid model based on sample entropy and random forest for SO2 prediction

As the main output of industrial waste gas, SO2 is harmful to environment and human health. For reducing the hazards to human from SO2, it is urgent and necessary to control the atmospheric pollution. Atmospheric pollution forecasting can provide effective information for controlling atmospheric pollution, so this research proposes a CEEMD-MR-Hybrid model for SO2 forecasting, which employs CEEMD, machine learning models (CNN, PSOSVR, PSOBP) and mode refactor system (MR). The main feature of proposed model is to reconstruct the decomposed series using MR including random forest and sample entropy for hybrid forecasting. Sample entropy can combine decomposed series with similar characteristics by measuring the complexity of distinct decomposed series. Random forest can obtain decomposed sub-sequences high correlated with the original sequence by ranking characteristics importance. Five geographically diverse cities in China are selected to test the validation of the proposed model. Compared with previous CEEMD-RN-Hybrid models, the results of proposed hybrid models have higher consistency with actual data. Therefore, the innovative model can be utilized as an effective model for SO2 forecasting. Taking Shenzhen as an example, the MAPE values of all CEEMD-MR-Hybrid models are smaller than 10% for SO2 forecasting. Therefore, the innovative model provides a machine learning system which can efficiently refactor the decomposition series to improve the accuracy for hybrid model.

Automated summary

This study proposes a CEEMD-MR-Hybrid model for SO2 forecasting, which includes CEEMD, machine learning models, and mode refactor system. By reconstructing the decomposed series using sample entropy and random forest, this model improves the accuracy of the forecasting.