Development of synthesis method of predictive schemes based on basic predictive models

Abstract

The task of modeling and prediction of processes of various natures is important. Known models and prediction techniques based on the use of integrated information on prehistory of predictive processes. Among the tasks of prediction an important place takes time series prediction. There are many different methods for the prediction of technical, economic processes. In this paper, the method of synthesis of predictive schemes based on key predictive models, which is based on determining the weighting coefficients of the models included in the resulting model. The best step is determined by the background conditions minimizing functional standard deviation at optimum parameters of autoregression models. In the synthesis of predictive schemes for each base model is determined the weighting with which it includes in the final predictive scheme.

Comparison of the results of predictive schemes with the results of basic techniques: autoregressive method, the method of least squares with weights, Brown’s linear model, Brown’s quadratic model, Winters method. It is used mean square error and average relative error of prediction of the various steps to assess the quality of the predictive models. Predictive scheme mainly improved the results of basic models for the studied time series, while scheme results coincided with the results of Winters model at some steps of prediction.

An important feature of predictive scheme is that it allows adding new time series prediction models, removing it from the models or groups of models, that is, the scheme is flexible to use.