Two Neural Network Methods in Estimation of Air Pollution Time Series

摘要

The measurement of air pollution parameters is a costly process. Due to several reasons, the devices may not take measurements for certain days. In such cases robust estimation methods are quite necessary in order to fill the gaps in the time series. Artificial neural networks have been employed successfully for this purpose for hydrometeorological time series, as reported in literature. In this study, modelling of the time series of air pollution parameters was investigated using two ANN methods; a radial basis function algorithm (RBF) and feed forward back propagation method (FFBP). The ANN methods were employed to estimate the PM10 values using the NO and CO values. The data were from a measurement station in Istanbul, Turkey. The results of an initial statistical analysis were considered in the determination of the input layer node number. In the estimation study, values corresponding to other air pollution parameters were included in the input layer. The results were compared to those obtained with a conventional multi-linear regression (MLR) method.