Model and method for evolutionary control of the decision-making structure in intelligent systems

Abstract

The relevance of the study is обусловлена the fact that modern intelligent software systems operate under conditions of incomplete, conflicting, and dynamically changing data, as well as limited computational resources, which necessitates improving the adaptability, robustness, and correctness of decision-making processes compared to traditional approaches focused on tuning individual algorithms. Furthermore, the increasing complexity of information environments and the need to process heterogeneous data sources intensify the requirements for consistency and reliability of analysis results. Under such conditions, the transition to controlling the structure of the decision-making process as a higher-level object becomes particularly important. The aim of the study is to improve the adaptability, robustness, and correctness of decisions by shifting from the tuning of individual algorithms to the controlled modification of the structure of the analysis process, including the composition of active software components, the order of their interaction, and the rules governing transitions between processing modes. The methods of the study are based on the approach of evolutionary control of the decision-making structure, which involves controlled transformations of admissible configurations, integration of fuzzy interpretation and evidence aggregation, assessment of the reliability and consistency of information sources, as well as formal proof of correctness, finite convergence, and local optimality properties. The proposed approach enables adaptive selection of the analysis process configuration depending on the level of uncertainty and available resources. In addition, the method ensures a balance between decision-making accuracy and computational resource costs under dynamic operating conditions. The results of the study consist in the development of a model and a method for evolutionary control of the decision-making structure, as well as their practical implementation in the form of an intelligent medical diagnostic system, where experimental studies have shown that as the level of uncertainty increases, the system automatically activates additional mechanisms for source consistency control and predictive analysis, thereby providing enhanced robustness of results compared to a baseline fixed configuration.