Work number - M 49 FILED
V.V. Moskalenko. –Sumy State University.
The aimof this work is to increase the functional efficiency of the self-learning automated control system for the technological processes (PCS) operating in conditions of a priori uncertainty due to random initial conditions and the influence of uncontrolled disturbing factors on the process.
Scientific noveltyof the results is the creation of a new information-extreme technology of analysis and synthesis of self-learning decision support system (DSS) within the process control system and it includes: complex of categorical models and information criterion for assessing functional efficiency allowing to optimize time-spatial parameters of DSS for getting adaptability of control system when it is functioning under a priori uncertainty; methods for build the hierarchical, nested hyper-spherical and multimodal hyper-ellipsoidal decision rules that improve the efficiency of machine learning and accordingly reliability of recognition a process's functional states; prognostic classification method, which allows to determine the time-point of relearning DSS for the correction of decision rules during operation of the control system; algorithms for optimization the time-intervals of observation, quantization step in time of observations and retrospective shift of time-point of reading each features to improve the accuracy of classification of functional states of the non-stationary process; method of information-extreme learning machine from imbalanced and heterogeneous dataset that allows to execute machine learning without pretreatment of training data; automation method of forming input mathematical description of DSS via information-extreme cluster analysis of input dataset with optimizing feature set that allows to increase the functional efficiency of DSS at an undefined set of classes.
The practical significanceof the work results is using of intelligent technology in the control system that allows to model cognitive processes inherent in human decision-making for improve the accuracy and efficiency of control system, competitiveness of products and reduce the influence of the human factor in the production process. Models, methods, algorithms and software packages of self-learning DSS for solving problems of growth a large scintillate single crystals from the melt are used in the Institute for scintillation materials NAS of Ukraine.
Publication. The total number of publications on this subject is 24, among them: 15 articles in scientific journals of Ukraine; 2 articles in international journals indexed in scientometric database SCOPUS; 6 proceedings of international conferences; 1 ukrainian utility patent.