A.V. Elshin, A.I. Mirzaev

In the article one has (it was) conducted on a qualitative level a comparative analysis of two main approaches to predictive modeling (of situations): the first one is based on extrapolation of time series of observations, the second - on the analysis of historical/retrospective precedents. In spite of relatively short-term prognosis, the second approach provides predictive modeling/forecasting of non-typical situations, what determines its application area in informational and analytical activity. The main contents of the article include modification of precedent-related and forecasting method, based on a quantitative determination of similarity degrees for retrospective and investigated/under study situations. The method peculiarity that not one-short characteristics measurements are compared, but its dynamics at a certain time interval, made us choose it as a prototype. This dynamics is determined with short-term stationarity of the investigated system transient processes and its environment. The research showed the necessity to provide representative input and output characteristics of analysis object as input materials/data for method realization. To implement the method under conditions of having access mainly to poor informative characteristics it was offered to establish/institute forecast system education by time intervals indication in retrospective data sets that correspond to the classes of state which are of interest to the researcher. Respectively, as a prototype we use situation data that belongs to the time intervals of researcher-s interest and beyond of them. Thus, the task of investigated complex system condition predictive modeling is reduced to the task of classification that allows us to use mathematical tools of nonparametric statistics to generate decision rules.