Abstract: | The paper considers the application of self-organizing models, specifically, the method of grouped arguments consideration (MGAC), to forecast short and non-stationary time series of observations in the ocean. A sequence of operations for the treatment of observational series is suggested. To assess its efficiency, we have used mean monthly oxygen concentration data collected in the surface and near-bottom layers of the Taganrog Bay. It is shown that the application of the MGAC model allows one to reduce by two times the root-mean-square error of that of the series prediction by five points, in comparison with the Jenkins-Box regressional model. It has been concluded that the predictors' non-linear functions may be effectively used in the treatment of short samplings. Translated by Vladimir A. Puchkin. |