MULTI-OUTPUT REGRESSION MODELS FOR CONTROLLING MULTICOMPONENT DYNAMIC SYSTEMS

Abstract

Modern multi-component systems are characterized by the interaction of numerous internal components and external factors, which can exhibit both regular and chaotic behavior. Effective management of such systems requires tools capable of providing accurate state predictions under conditions of uncertainty and limited input data. This article explores the use of multioutput regression models, which enable the consideration of interdependencies among system components, optimization of the parametric space, and improvement in prediction accuracy. Multi-output models allow simultaneous forecasting of several aspects of a system's state, reducing errors and enhancing the generalization ability of the models. The article provides a detailed examination of methods to improve such models, including minimization of noise influence, accounting for the temporal scales of component changes, optimization for small data samples, and increasing the interpretability of predictions. Approaches to addressing data scarcity are proposed, such as knowledge sharing between tasks and the use of generative models. Special attention is given to the challenges of applying multi-output models, including the risks of overfitting, conflicts between optimization objectives, and the impact of correlation biases. Strategies to mitigate these risks are discussed, including adapting multi-criteria optimization, parameter regularization, and developing hierarchical models that can account for system dynamics across different time scales. Ensemble approaches, which integrate the outputs of submodels into a unified architecture, are highlighted for their ability to enhance noise robustness, prediction accuracy, and model adaptability to changing conditions. The approaches proposed in the article have practical significance for automating decision-making processes in complex multi-component systems operating under high variability and data limitations. This provides a comprehensive framework for forecasting, contributing to more effective management of dynamic systems across various domains. Thus, the article makes a significant contribution to the development of methodologies for modeling complex systems, expanding the possibilities for their analysis and management.