High-Accuracy Particulate Matter Prediction Model Based on Artificial Neural Network

This paper presents a low-cost high-accuracy method for the prediction of the air pollutant Particulate Matter 2.5 (PM2.5). The PM2.5 pollutant is very harmful to humans, animals, and vegetation, and its index depends on many factors. As the existing PM2.5 monitoring methods are mostly expensive, and PM2.5 values are usually not measured at every meteorological station, the PM2.5 prediction is of great importance. The cost-effective and efficient method proposed in this paper is based on a Multilayer Perceptron Artificial Neural Network (MLP-ANN). The PM2.5 level is predicted using the meteorological factors that are easy to measure. The prediction accuracy has been tested at two locations: one at which the training data were collected, and another 250 km away from the first. Excellent prediction accuracy is achieved, showing a great practical significance of the proposed prediction method.