A clustering-based ensemble approach with improved pigeon-inspired optimization and extreme learning machine for air quality prediction.

In this paper, a novel hybrid learning method is carried out to forecast urban air quality index (AQI). Wavelet packet decomposition (WPD) is firstly performed to decompose the original AQI data into lower-frequency subseries. Then, we improve the pigeon-inspired optimization through using the particle swarm optimization algorithm. The improved pigeon-inspired optimization (IPIO) approach is applied to optimize the initial weights and thresholds of extreme learning machine (ELM) and then the modified ELM (MELM) is employed to forecast the subseries respectively. Moreover, multidimensional scaling and K-means (MSK) clustering methods are utilized to cluster the forecasting outcomes into high frequency, medium–high frequency, medium–low frequency and low frequency subseries. Finally, MELM, as an ensemble approach, is applied to ensemble the subseries together and achieve the final results. To test the predictive precision of the proposed hybrid WPD-MELM-MSK-MELM learning method, AQI of Harbin in China is adopted to make short-term, middle-term and long-term predictions separately. Different decomposition approaches are utilized to compare with WPD, and the non-clustering hybrid model is also compared with the proposed method. The forecasting outcomes indicate that WPD is more suitable for predicting AQI and the proposed WPD-MELM-MSK-MELM learning method has better predictive performance on horizontal precision, directional precision and robustness than some existing methods and benchmark models in this paper. • A novel hybrid learning approach is presented. • Pigeon-inspired optimization is improved through using particle swarm optimization. • Extreme learning machine is modified by pigeon-inspired optimization. • Based on the multidimensional scaling and K-means clustering, the ensemble approach can enhance the predictable precision of AQI. [ABSTRACT FROM AUTHOR]