AQI multi-point spatiotemporal prediction based on K-mean clustering and RNN-LSTM model

Abstract

The short term air quality index can usually be predicted by statistical and numerical methods, but for the multi-point prediction of AQI, the traditional methods are often inaccurate. In this paper, a new hybrid multi-point prediction method was proposed by combining K-means clustering with the circulating neural network long and short time memory (RNN-LSTM) model. Based on this prediction method, the air quality index in Dezhou was predicted 1-5 days in advance by using 28 multi-point pollution monitoring sensor data from January 1, 2018 solstice to August 31. The prediction results show that the model not only improves the accuracy and effectiveness of the prediction, but also reveals the relationship between land use patterns and air quality index (AQI), which provides important information for land use planning, air pollution mitigation and urban intelligent governance.