OPTIMIZATION OF PARAMETERS OF COMPLEX SYSTEMS BASEDON A GENETIC ALGORITHM WITH STATE PREDICTION USINGTHE KALMAN FILTER

Abstract

The paper addresses the problem of parameter optimization for complex dynamic systems operating under uncertainty, measurement noise, and limited computational resources. A novel method is proposed that combines a genetic algorithm with Kalman-filter-based state estimation and prediction. This integration improves the fitness function's reliability by using filtered and predicted state estimates rather than directly measured, noisy data. The study presents a mathematical model of the system, formulates the optimization objective function with constraints and stochastic disturbances, and describes the mechanism of integrating the Kalman filter into the evolutionary optimization process. A generalized algorithm for the proposed method is developed, including population initialization, state estimation, fitness evaluation, selection, crossover, and mutation. Experimental validation is conducted in a simulated environment that models a dynamic system with varying noise levels. The results demonstrate reductions in mean-squared error and faster convergence speed compared to classical approaches, including a standard genetic algorithm without prediction and gradient-based optimization methods. It is shown that applying the Kalman filter significantly enhances the robustness of the optimization process against noise and improves the algorithm's stability. The proposed method exhibits high adaptability to changing environmental conditions, reduces the probability of convergence to local minima, and remains computationally efficient. The practical value of the research lies in the method's applicability to embedded and cyber-physical systems, particularly for real-time control, monitoring, and data processing tasks.