Knowledge Based Prediction Model: A Case Study of Urban Air Pollutant Concentrations

This paper proposes a model that adaptively predicts the hourly concentrations of nitrogen dioxide in the central urban area of Seoul, Korea. In order to consider the hourly variations of air dispersion condition with limited information, an expert system methodology is used. The knowledge base about atmospheric dispersion has been organized by interviewing seven experts in the field. The variables in the knowledge base are wind direction and speed, cloud height and cover, stability and inversion strength. A statistical time series model, in this case a state space model that characterizes air pollutant dispersion is combined with the knowledge base. The statistical part produces the prediction value using the parameters from knowledge inference. The results of empirical study show that the proposed prediction model performs better than general time series models.