Application And Research For Wavelet Mutation-based Binary Particle Swarm Strategy

Particle swarm algorithms get their optimal solutions iteratively, through theparallel search approaches by a group of initial particle swarms. Weak dependenciesto problems, simple concepts, quick convergence, and easy to be implemented aretypical advantages of it. Accordingly, it is generally adopted in function optimization,multiple objective optimizations, automatic objective detection, biological signalrecognition, image partitioning, dynamical environment optimization, schedulingdecision, fuzzy control system, neural network training, etc.Nevertheless, particle swarm algorithms are usually acceptable to theoptimization problems for continuous space, while hardly adaptable to the numerousexisting problems which are discrete, just with limited variables. Consequently, tomake them accommodated to the optimization problems in discrete space, the particleswarm algorithms should be discretized. Additionally, the premature could becommenced in particle swarm algorithms, which will contribute to decrease ofprecision for solutions. Therefore, the precision of solutions should be furtherimproved. This paper focuses on the following issues:At first, the basic principle and evolution mechanism of particle swarmalgorithms are researched, the primary two kinds of discrete approaches are analyzed,and further dedicate to the binary particle swarm algorithms. On the foundation ofbasic principle of binary particle swarm, the concrete evolution rules are transformed.The probability for current position of particle swarm is calculated by introduction ofBayes formulas, based on the result of the combination of the individual optimalsolution, global optimal solution and the result of last iteration. Then, the probabilityis compared to a random value, in order to determine the position for particle swarm.Meanwhile, based on these principles, an optimized binary particle swarm algorithmwas proposed. Experiment results demonstrate that the solutions are effective,convergent and parallel to the expected requirements, via the computation of DeJongtest.At second, on the purpose of deal with the deficiency of improved binary particleswarm algorithms, such as easy premature, low precision of solutions, a waveletmutation operation was introduced to adjust the particle swarms slightly and improvethe diversity of them, then a wavelet mutation-based binary particle swarm algorithm was presented. The results gained from this algorithm are more precise than binaryparticle swarm algorithm.At last, the wavelet mutation-based binary particle swarm algorithm isintroduced for the software/hardware partitioning problems. The objective system areconstrained as bi-dimensional partitioning pattern, say single CPU and single ASICarchitecture. Then, Directed Acyclic Graph (DAG) was used to model the objectivesystem, and transform the software/hardware partitioning problems into snackproblem, with constraints. A improved breadth first search approach was proposed toschedule the tasks, whose efficiencies are verified by the computation of instances.Wavelet mutation binary particle swarm method and binary particle swarm approachare adopted to partition the software and hardware tasks in different scale. Theexperiment results indicate the solution of former is better than later, since its overallexecuting time is shorter and the partitioning result is better, under the condition ofmeeting all the constraints.