Method for the prediction of time series using small sets of experimental samples

摘要

The paper is concerned with the method of prediction of time series based on the concepts of system identification. The distinctive property of the method is the use of small sets of experimental samples of data. The latter create some basis for building so-called learning subsets, which are used to construct particular prediction models. Values of variables predicted by different particular models allow calculating the desired variables by using a batch voting technique. The method can be used for short-term prediction of data values at future time instants based on the analysis of a brief history of the process under consideration. It can be useful in cases of processing very large arrays of data samples, when the researcher has to confirm his (her) attention to only a small part of the samples received at the last instants of time, in view of the limited memory of the computer or in cases when very slow processes are analyzed. A special place in the paper is given to the example in which the computational aspects of the proposed method are considered in detail.