Research On Backtracking Search Algorithm For Numerical Optimization

Optimization algorithm plays an important role in applied mathematics,and swarm intelligence algorithm has become an indispensable research direction of artificial intelligence and computer science.Backtracking search algorithm(BSA)is a novel population-based stochastic technique with a simple structure.BSA has a powerful exploration capability and achieves good results in solving multi-model problems.However,influenced by historical experience,the convergence speed of BSA slows down and prejudices exploitation on later evolution period.In the lights of the slow convergence on later stage of iteration,we propose best guided backtracking search algorithm and sequential quadratic programming enhanced backtracking search algorithm.First,we propose best guided backtracking search algorithm(BGBSA)which utilizes historical experience or information derived from the best individual at different iteration process to balance the exploration and exploitation capabilities.The iteration period is divided into early stage and later stage in BGBSA.At early stage,BGBSA takes full advantage of historical experience to maintain a powerful exploration capability.During later evolution process,the efficient information provided by the best individual is utilized to speed up the convergence.The performance of BGBSA is verified on CEC-2013,and the experiments results show that BGBSA can achieve faster convergence speed and solve benchmark problems more successfully than the compared algorithms.Second,we design a novel hybrid method named SQPBSA which combines BSA and sequential quadratic programming(SQP).BSA,as an exploration search engine,gives a good direction to the global optimal region,while SQP is used as a local search technique to exploit the optimal solution.SQP which is invoked in the early stage of algorithm is capable of calculating promising guiding direction as soon as possible to improve convergence speed and to favor exploitation.We use 28 benchmark functions to verify the performance of SQPBSA,and the results show improvement in effectiveness and efficiency of hybridization of BSA and SQP.