The pragmatic aim of this paper is to provide a genetic algorithm for predicting the technical systems state. The research novelty is to represent the genuine approach to forecast the technical systems states. The given approach implies finding future values by extrapolating current observation results. Forecast can be considered as a diagnostic control at zero-time extrapolation, or as a general case of diagnosis. The developed genetic algorithm is based on the classical representation of genetic algorithms with the changes required for forecasting. So, a function that validates alternative solutions outlaying from the geometric representation of the average values of the time series plot is exploited as a fitness function. The method based on the use of Shewhart process-behavior charts is also applied to exclude failures of the sensor collecting measured data and to control the mutation. The algorithm performs a prediction for one time interval ahead for processes that are not affected by external factors or processes, or the influence of external factors on which is not significant within one time interval. Our experiment confirmed the efficiency of the suggested algorithm. It resulted in obtaining a predictive solution.

No abstract

The paper considers the algorithm of diagnostics of the technical system States at future time points by extrapolation of the results of current observations using a genetic algorithm. The novelty of this work lies in the fact that the proposed algorithm is able to generate predictions of technical system States using predictive algorithms and mathematical rules. The paper offers an original view on the use of genetic algorithm as an independent predictive algorithm. The described algorithm is a combination of a modified genetic algorithm and a number of mathematical rules. So, as a modification of the genetic algorithm, its parallel variant (island model) is used, and as a fitness function, a function is used that tests new alternative solutions for distance from the geometric representation of the averaged values of the time series graph. The algorithm performs prediction for a given number of time intervals ahead, for processes that are affected by a limited number of external factors. The efficiency of the algorithm was confirmed by an experiment, which resulted in a predictive solution, the General direction of the process (which was tested in practice).