Research On Swam Intelligence Algorithm Based On Spark

Swarm intelligence algorithms can effectively tackle optimization problems that are difficult to solve by using traditional optimization algorithms.However,with the huge increase in the time and space cost for solving optimization problems,the use of swarm intelligence algorithms in a single-machine environment suffer from the limitation of overly long computation time.Based on Spark,which is the most popular open-source distributed computing framework,this thesis studies specifically using swarm intelligence algorithms to solve combinatorial optimization problems.This thesis aims at the high time and space complexity of the optimization problem,and uses permutation flowshop scheduling problem as a case study.Based on the different characteristics of typical swarm intelligent algorithms,including particle swarm optimization,ant colony optimization,bat algorithm,firefly algorithm and grey wolf optimizer,we proposed and implemented Spark-based distributed parallel acceleration schemes for swarm intelligence algorithm population iteration and parameter tuning,respectively.In the proposed distributed population iterative schemes,we first initialize the swarm and generates the initial solution,then perform the distributed iterative evolution procedure,and finally obtain and return the optimal solution.In addition,in order to improve solution quality,we rely on Spark platform to perform distributed parameter tuning.The tuning strategy first generates different parameter combinations according to a given parameter list,then execute swarm intelligence algorithms with different parameter combinations in a distributed and parallel manner,and finally determine the optimal parameter combination by comparing the solutions of all algorithms.Furthermore,to justify the applicability of the proposed approach in solving realistic problems,we apply the distributed particle swarm optimization algorithm to cope with the cloud computing resource allocation problem,which is a constrained combinatorial optimization problem.This thesis uses benchmark data for experiments to evaluate the performance of the proposed distributed algorithms.Experimental results show that the distributed algorithms significantly enhance the computational efficiency,in terms of up to 7×speedup with 8 computing nodes,while guaranteeing similar solution quality,and can obtain better solutions through parameter tuning.To conclude,the Spark-based distributed swarm intelligence algorithms proposed in this thesis can significantly improve the computational efficiency,and therefore provide a feasible solution to solving complex optimization problems.