An innovative ensemble learning air pollution early-warning system for China based on incremental extreme learning machine

Abstract

Air pollution has lots of adverse effects on industrial production and public life. Thus, it is an urgent task to construct an efficient air quality early-warning system to guide public life and production. This paper proposes an innovative air pollution early-warning system, including four main modules: clustering, preprocessing, forecasting and evaluation. In the clustering module, with the aim of building an efficient air pollution warning system, the air pollution situation of 31 provincial capitals is clustered and the study areas of the current study are selected based on the clustering result. A new data preprocessing algorithm is conducted to excavate the potential characteristics of the raw time series in the first place in the preprocessing module. Then, the length-changeable incremental extreme learning machine is used to forecast each component. In the evaluation module, the air quality is qualitatively analyzed by the fuzzy evaluation method. Moreover, the DM test and the SPA test are employed to test the accuracy of the forecasting model. The experimental results of eighteen data sets from three cities show that the hybrid air quality early-warning system establish in the study not only has higher accuracy and generalization ability than other benchmark models, but can provide sufficient air quality information, which is essential to control air pollution.