Consideration is being given to problems of global optimization of main dynamic characteristics of hydromechanical systems. It is suggested that criterion functions are continuous, Lipschitzian, multiextremal and not everywhere differentiable. New hybrid algorithms are proposed with scanning a search space by use of modern stochastic algorithms on base of analogy with absorption and scattering processes for nuclear particles. The local search is implemented using a linearization method with smoothing approximations of criteria for the sequential algorithm or reduction of the problem dimension for the parallel one. Some results on solving model problems in identification of the coolant phase constitution anomalies in the reactor circuit with the use of spectral data are presented.
|
