A novel long-term air quality forecasting algorithm based on kNN and NARX

In this paper, we propose a novel approach of forecasting long-term air quality. The methodology of our solution employs similar sequence search (using kNN) of historical data as reference to make up for the lack of information of the unknowing future conditions. Then these reference values as well as time series features, meterological data, air pollutant concentrations, neighbor station's air quality and weather forecast are utilized to forecast target station's future air quality (using NARX). We Compared with baseline approaches, and our approach shows superior performance.

I. Introduction

Precisely forecasting the value of air quality has a great significance for social decision making. The governments can set the environment policies about pollutant emission referring to the future values of air quality, and local residents might make their out-going plans according to accurate and trustworthy air quality predictions. Many cities have built air quality monitoring stations to inform citizens the real-time (hourly) air quality index (AQI), and provide hourly or daily forecasting services. A long-term prediction is required to support superior planning and decision making.