PM2.5 Forecasting Using Pre-trained Components

This work investigates the efforts of composite neural network in PM2.5 prediction. A composite neural network is a combination of several pre-trained and non-instantiated neural networks and according to [1], under certain condition, it is almost guaranteed that the performance of composite network is better than any of its pre-trained component. In this study, we designed a series of systematic experiments to examine the effect of the theory towards PM2.5 prediction. The empirical results support the theory and identify a reasonably good composite neural network design for PM2.5 prediction.