Hourly PM2.5 concentration multi-step forecasting method based on extreme learning machine, boosting algorithm and error correction model

Abstract

The multi-step forecasting of PM2.5 concentration is helpful to realize the early warning of air pollution, but the accurate multi-step forecasting of PM2.5 has certain difficulties. In this paper, a novel multi-step forecasting method of hourly PM2.5 concentration is proposed. Two boosting algorithms, Modified AdaBoost.RT and Gradient Boosting, are used to enhance the extreme learning machine (ELM) for ensemble prediction of the PM2.5. Then two multi-step forecasting strategies, multiple-input multiple-output (MIMO) and recursive, are used. Finally, through error correction model (ECM) the prediction error is corrected to obtain the hourly PM2.5 multi-step forecasting results. Corresponding experiments are carried out through the PM2.5 data sets of four cities, and the results show that: (1) the forecasting method proposed in this study can achieve a good multi-step forecasting effect of PM2.5, and changing the forecasting strategy or boosting algorithm has little influence on the forecasting effect; (2) the use of ECM can improve the PM2.5 forecasting accuracy of the model, and as the forecasting steps increase, the improvement effect of ECM is more significant; (3) the forecasting framework proposed in this paper is effective, and the forecasting accuracy of the proposed method is significantly better than the corresponding single models and the existing models.