The Real Number Pmbga Achieve Improvements

PMBGA is a new algorithm derived from GA. It introduces some concepts of statistics into GA, and uses the statistical learning of population to make further evolution and uses the distribution of individuals to analyze the learning result of population according to the new generated population. Therefore gradually it got closed to be the best. Real-coded PMBGA is studied in this paper. Some improving methods are proposed and tested after the frame of PMBGA and the selection of parameters are analyzed.The algorithm flow is discussed first and then the difference between PMBGA and GA, and their comparison are analyzed. PMBGA is validated to be better on the evolution quality and converge efficiency.Population size, the changing rules of probabilistic model, probabilistic model with differ distributions, stop conditions are tested to have effect on the performance of real-coded PMBGA. A small population size and a big learning rate can cause bad convergence.Based on analysis of parameters, a self-adapted PMBGA is proposed with a flexible population size and a learning rate to accelerate converging speed and reduce time consumption.A new model of weighted selected individuals is proposed in order to search more useful information. Fitness factor is added into the process of statistic evaluation. This improvement is validated to be better converging quality and more efficient.Finally, PMBGA is used for direct design of R Digital Filter and the experiment also shows feasibility of this method.