Research On Baldwin Effect-based Particle Swarm Optimization

Particle Swarm Optimization (PSO) is an intelligent optimization algorithm basedon searching in group. It is a mathematics-based application technique used to solvecombinatorial optimization problems. Over the last decade, more and more scholars arepaying attention to PSO. This algorithm needs fewer parameters and is easy to realize,what’s more complicated problems can be solved by this algorithm effectively.Therefore, PSO has been widely used in function optimization, image processing,pattern recognition and engineering fields. Meanwhile, a lot of PSO variants have beenproduced based on the primitive PSO. However, PSO is still not mature enough both intheory and practice. Similar to other optimization algorithms, PSO is easy to fall intolocal optimal value, meanwhile its convergence speed is slow in later phase and itsaccuracy of convergence result is poor. These problems are focused on by most ofexperts in this field these years.According to the problems of slow convergence speed, poor convergence resultsand easy to fall into premature convergence, a Baldwin Effect-based PSO is proposed inthis paper. The concept of good point set in number theory is used to construct the initialpopulation in this algorithm, which makes the distribution of the particles is moreevenly. Thus the diversity of the population is improved, and the probability of findingthe global optimal solution is creased greatly. During the different phase of the runningprocess of this algorithm, a cosine function is used to adjust the value of the inertiaweight factor adaptively, which can make the variation of the inertia weight factor ismore in line with the characteristics of the population search. Not only the convergencespeed is improved, but also the convergence results are ameliorated by this way.Baldwin effect is used to get the feedback information during iteration. Hence, particlescould perform reinforcement learn, and then better directions which can improve thesearch results are got. Parameters involved in reinforcement learning are determined bythe comparison of the results of experiments. It is naturally to choose the value whichcould make the results better.Some classic functions are used to test the performance of Baldwin Effect-basedPSO. Meanwhile several PSO variants are compared with Baldwin Effect-based PSO bycomputing these test functions. Results show that, not only the convergence speed, butalso the convergence results are improved obviously, and meanwhile the prematureconvergence is ameliorated too.Some complex multi-valued functions and multi-objective functions are always difficult to solve in the engineering applications. However, PSO has advantage to solvethis kind of problems for its special character. According these problems, somemathematical models using PSO have been proposed although they are still not matureenough and should be further perfected. This will be my future research direction.