Cooperative Co-evolution Particle Swarm Computing And Its Application

Enlightened by natural phenomena and evolution, the Swarm IntelligenceAlgorithm is widely applied into science and engineering fields for solvingoptimization problems. However, serious performance degradation of theconventional intelligence algorithm happens when optimization problem increases itsscale over time. Currently, the swarm intelligence algorithm with the cooperativeco-evolutionary mechanism is one of the effective methods for large-scale continuousoptimization problems, which enables the abstract algorithm model to be built flexiblybased on the specific optimization problem, thus effectively improving solutionquality, efficiency and robustness.Particle Swarm Optimization algorithm (PSO) has long been regarded as a reliableand universal meta-heuristic optimization technique that has already provenremarkable performance in tackling all kinds of problems. However, the conventionalPSO algorithm can easily get defects such as local stagnation, premature convergenceand low accuracy. This article provides systematical discussion and research on thefundamental theory and algorithm key factors of PSO algorithm for improving itsperformance, incorporates the cooperative co-evolutionary theory into PSO algorithmfrom many different angles, brings out the computational model of particle swarmintelligence based on cooperative co-evolutionary mechanism and applies it to actualmanagement optimization problems, enlarges scale of research on large-scalecomplicated optimization problems and provides new resolution for large-scalecomplicated optimization problems. This article has provided a summary of creativework as below:(1) A hybrid self-adaptive learning parallel particle swarm optimization (HLPSO)was brought out to improve the universal application for particle swarm optimizationalgorithm. The HLPSO algorithm combined four different strategies: rapidconvergence, local optimization jumping out, exploration in depth and development inwidth, it was able to choose an appropriate strategy to solve optimization problemsbased on their complexity by incorporating the self-adaptive learning mechanism. Theemulation experiment has proved a great improvement of the algorithm onoptimization efficiency, robustness and universal application ability.(2) A multi-stage dynamic swarm intelligence algorithm (DMPSOABC) wasdesigned by incorporating the co-evolutionary mode and parallel evolutionmechanism into the particle swarm optimization algorithm and bee colony algorithmafter learning from the cooperative co-evolutionary strategy and the unique characterof swarm intelligence algorithm. The DMPSOABC algorithm utilized the advantagesof both dynamic swarm intelligence algorithm and strong development ability of bee colony algorithm. The algorithm process had been divided into three stages: Firstly,the DMPSOABC conducted rough search through local mode of swarm intelligencealgorithm for maintaining group diversities. Secondly, it conducted deep and widesearch through the co-evolutionary bee colony algorithm which gave strong feedback.Thirdly, it improved search optimization speed by used global mode of swarmintelligence algorithm to complete the global optimization. Through experiment onthe tests of functions optimization and flexible job shop scheduling problem, theDMPSOABC algorithm has proved merits such as quick convergence, strong globalsearch ability, stability and high efficiency in optimization.(3) An efficient multi-objective particle swarm optimization algorithm based onspace division with dynamic multi-swarms (ECMPSO) was proposed. Themulti-objective search space is divided into multiple divisions and the particles ineach division are guided by a new local and global guider to get close to the Paretofront. The time observer is applied to conduct real-time record of the contribution theguider has made for the particle to get close to the Pareto optimal solution set. Theguider will be changed periodically based on the contribution degree to keep groupdiversities. The ECMPSO algorithm created a new elite learning strategy whichincorporated differential evolution mechanism into Pareto optimal solution set andprevented its premature convergence. The emulation experiments on the internationalmulti-objective functions and environmental economic problem, which has provedthat ECMPSO was able to find a suitable optimal solution set quickly through full andcomplete research on the space.(4) The typical problem in management optimization, ie the vehicle routing withtime windows as research object, a multi-objective combinatorial optimization(HMPSO) was proposed, which adopted encoding methods based on set andprobability while incorporated the initialization method of the insertion heuristic andlocal search operator. The emulation experiments on the international standard test,optimization problem of the vehicle routing with time windows has proved thatHMPSO got higher precision reputation than other heuristic algorithms and offersutility value in improving distribution efficiency and reducing logistics costs.(5) A co-evolutionary two-stage multi-objective particle swarm optimizationalgorithm(BCMPSO) was proposed, by conducted research on a three level supplychain network related to factory supplier, distribution center and client, theoptimization solution mode for problems in the whole process in which themerchandise being delivered from factory suppliers through distribution center to theclient, such as site decision, inventory demand, route selection etc based onconsideration of carbon emissions with lowest supply cost and the least carbonemission as the objective goal. The research is analyzed and conclusion is made bycomparison of the results with and without consideration of carbon emission.