Identification of frequency domain maximum likelihood system using genetic algorithm

Float point genetic algorithms were used to solve the start value and convergence problems of frequency domain maximum likelihood system identification, based on errors-in-variables model. Simulations showed that it is difficult to get accurate result only using the genetic algorithm, while the traditional non-linear iterative optimization methods may lead to convergence to local minimum in some cases. By taking the advantages and overcoming the defects of the above two methods, an improved algorithm was proposed which can give the start value of delay directly and can find the global minimum precisely within a rather short time, even when cost function has a lot of local minima. In addition, the improved algorithm exhibited broader adaptability than the old ones.