Heuristics free group method of data handling algorithm of generating optimal partial polynomials with application to air pollution prediction

Abstract In this paper a revised GMDH (Group Method of Data Handling) algorithm is developed in which heuristicsare not required such as dividing the available date. into training data and checking data, predetermining the structure of the partial polynomials, or predetermining the number of intermediate variables. In this algorithm the prediction error criterion, such as PSS (Prediction Sum of Squares) or AIC (Akaike's Information Criterion) evaluated from all the available data, in used as a criterion for generating optimal partial polynomials, for selecting intermediate variables and for stopping the multilayered iterative computation. This heuristics freeGMDH algorithm is applied to non-linear modelling for short-term prediction of air pollution concentration. By using the time series data of SO2, concentration, the wind velocity and the wind direction in Tokushima; Japan, a suitable model for predicting SO2concentration at a few hours in advance is developed. The predicted results obtained by the rev...