Ziborov I. Information technology for decision support in complex production processes based on the evolutionary method

The dissertation solves the relevant scientific and applied problem of creating decision-making information technology for effective planning and management of complex production processes based on the evolutionary method. The purpose of the work is to increase the effectiveness, speed of obtaining, and quality of management decisions in planning complex manufacturing processes under uncertainty by developing an information technology for decision support using heuristic search methods. Scientific novelty of the obtained results: 1. For the first time, a hybrid method of conditional optimization based on particle swarm and artificial immune system is proposed, which, unlike existing methods, involves the division of the population into groups, intergroup competition, and population compression mechanism for the adaptation of the compression operator. The application of the method as part of the information technology for decision support allows reducing the time to obtain and improving the quality of these decisions. 2. For the first time, a modular information technology for decision-making in planning and managing complex production processes is proposed, where each module solves a specific optimization problem using a hybrid evolutionary method. Its application allows increasing the efficiency of decision-making in production processes and certification of end products. 3. The improved method of statistical certification of products is based on restoring individual models of optimal complexity for each product type, which, unlike existing methods, uses constraints on predictor degrees that can take negative values. 7 The improved method allows for product certification without direct measurements, with a relative error not exceeding 5%. 4. The method of planning the need for ingredients based on the chemical analysis of the produced product has been improved, which, unlike existing methods, involves restoring predictive models using the developed hybrid evolutionary method. This allows creating individual predictive models for each type of product with a relative error of no more than 8.5% and requiring an order of magnitude fewer training examples in the training sample. The practical significance: The developed information technologies and methods in the dissertation research have been implemented as software modules that can be used jointly or independently to support decision-making in complex production processes. In the first chapter, the problem of information-analytical support of control decision-making processes in complex production processes was analyzed, using the example of a metallurgical enterprise. The relevance of research, which should result in the creation of information technology to support decision-making in batching, deoxidation, and forecasting the mechanical properties of finished products, was noted. It is shown that within the proposed information technology, a number of optimization problems with significant constraints, which can be considered multi-criteria, need to be solved. The effectiveness of the evolutionary approach to solving multi-criteria optimization problems in real space, particularly methods based on the particle swarm principle and artificial immune systems modeling, is emphasized. In the second chapter, a hybrid method of conditional optimization is developed, which uses particle swarm and artificial immune system approaches and involves dividing the population into groups with individually variable agent numbers, intergroup competition, and population compression to counter convergence. The application of the developed HIPSO method as part of the information technology for decision support allows reducing the time and increasing the reliability and accuracy of the obtained solutions. During the study of the proposed method, it was found that the coefficients of the speed components do not depend on the problem space dimension; the number of groups into which the swarm is divided should increase with the increasing dimension of the problem, and the number of defeats of a certain group of the swarm after which the given group dissolves does not depend on the problem dimension; the total size of the swarm should increase proportionally to the number of partition groups as the space dimension increases; the periodicity of intergroup compression should increase with the problem dimension and can be chosen as a simple number within wide limits, while the periodicity of intragroup compression does not depend on the dimension and can be taken as a simple number from 7 to 13.