In order to reduce the prediction error of Air Quality Index (AQI) by Extreme Learning Machine (ELM), an Intelligent Composite Prediction Model (ICPM) is proposed. ICPM uses an Improved Whale Optimization Algorithm (IWOA) to find the ELM parameters. IWOA introduces logarithmically varying nonlinear control factors and cosine varying adaptive weighting factors to balance local exploitation with global search capabilities. Prediction of AQI combined with daily historical data of air quality in Henan Province (2019), it is proved that ICPM has better prediction performance and generalization performance than ELM and other models.