Local Smoothing Approximation in Hybrid Algorithm of Optimization of Hydromechanical Systems

Problems of global optimization of hydromechanical systems with continuous not everywhere differentiable criterion-functions are considered. The hybrid algorithm is proposed which combines the PCA stochastic algorithm and linearization method. A variant of the linearization method for local optimization problems is implemented on the basis of construction of smoothing approximations of criteria, and necessary conditions of optimality are formulated. The algorithm's local convergence is investigated. A numerical example is given of solving problems to diagnose the phase composition of the heat-transport medium in the reactor plant loop using spectral data.