Non-physical Model Research Articles

Statistical Prediction of Air Pollution Levels Based on Kalman Filtering Method

On-line predictions at a short range of air pollution levels by non-physical models have been developed[1][2]. In this paper, first, the comparisons of the predicted values of air pollution levels based on the following non-physical models are done; (i) linear auto-regressive model, (ii) multiple linear regression model, (iii) Box-Jenkins' method, (iv) a principle of persistence, and (v) a method of adaptive transformation network theory. The actual data measured in Tokyo and Tokushima, Japan, are used. Our methods shown in (i) and (ii) are found to be significantly more accurate and useful than other methods. Secondly, the methods for estimating the confidence intervals of the mean and the variance of the pollution levels at a fixed time are described by using the theory of the modeling and estimation[3]. It is shown that applications of the idea of the modeling to the prediction give a good accuracy to the statistics of the population of the pollution levels. Lastly, a program package for on-line predictions of air pollution levels is presented.

The Predictions of Air Pollution Levels by Nonphysical Models Based on Kalman Filtering Method

We describe the applications of multiple linear regression model and auto-regressive model which may be of use for the on-line prediction and control of concentration levels of pollutants of air pollution. At the beginning of these researches, in this paper, are presented the prediction of air pollution levels at a few hours in advance. The state variables of the multiple linear regression model are determined by considering the contribution of the component analysis. Practical data measured in Tokyo and Tokushima prefecture in Japan are used, respectively. Kalman filtering method is utilized for the prediction by the multiple linear regression model. Auto-regressive model is fitted to the time series which is processed by subtracting the moving average from the original observed data sequence. Accuracy and characteristic of the prediction by the models presented here are compared with the model of the Box and Jenkins, and with that obtained by the principle of persistence, respectively. Both are found to be significantly more accurate and useful than these models.