Prediction and Comparative Study of PM_2.5 Concentration Based on Multi-stage Clustering

A deep neural network (DNN) prediction model based on multi-stage clustering is proposed for multi-step PM_2.5 concentration prediction. The proposed model includes decomposition, clustering and prediction. In the part of clustering, the first stage uses HDBscan density clustering to eliminate the noise, and then carries on the second stage clustering. In the second stage, Kmeans, AHClomerative, Gaussian mixture and birch clustering algorithms are used. In the prediction part, the deep neural network (DNN) is used as the predictor, and the hourly data of 11 air quality monitoring stations in Shenzhen are selected to verify the effectiveness of the model. The experimental results show that the prediction model based on multi-stage clustering is suitable for multi-step high-precision prediction of PM concentration, and its performance is better than DNN model and single-stage clustering prediction model.