Structural Classification of Phase Space Projections with Neural Network-Based Optimization of Sliding Mode Existence Regions for Automated Control System Synthesis

This paper proposes a method for the structural classification of phase space projections with neural network-based optimization of sliding mode existence regions to facilitate the automated synthesis of control systems based on qualitative analysis of the phase space. Using the Cuk converter as an example, the features of phase space projections in multidimensional systems are studied, along with the limitations inherent to such projections from the perspective of subsequent control law synthesis. A method for circumventing these limitations using sliding mode control is proposed. A technique for generating switching lines for sliding mode control based on phase space projections is introduced, allowing for the classification of projections in terms of their suitability for organizing sliding mode control. Special attention is given to maximizing the area of sliding mode existence regions using a neural network - a multilayer perceptron. It is shown that such neural network-based optimization significantly enhances the potential for selecting control algorithms. The proposed method extends the applicability of phase space projections in the qualitative analysis of multidimensional systems.