Research On Improved Sparrow Search Algorithm And Its Application

The meta-heuristic algorithm is a kind of famous optimization algorithm that can optimize problems without gradient information and is very easy to expand.Since the proposition of the Genetic Algorithm in 1967,thousands of meta-heuristic algorithms have been proposed.The proposed methods imitate the behavior of animals or phenomena of physics to build models to do optimizations.Although the existing methods contain many search structures and search strategies,there is still room to propose new meta-heuristic algorithms with better exploration and exploitation abilities.Sparrow Search Algorithm(SSA)is a novel meta-heuristic algorithm proposed in2020.Compared with other meta-heuristic algorithms,SSA has a special search structure and strategy for iterations.The search results provided by SSA are better and more robust.However,SSA has some shortcomings,such as tending to search around the origin and doing iterations based on the fitness values.Furthermore,SSA is designed for(constrained or unconstrained)continuous problems and is unsuitable for distinct problems.Aim to overcome these deficiencies,this paper does some work:(1)Using the distance information to build some new formulas and replace the formulas that tend to search around the origin in SSA.After this work,the algorithm does not tend to search a specific area.Adding the concept of ‘volume’ in the algorithm to build formulas with great ability to move out from the local optima.The improved sparrow algorithm(SSSA)is tested on 8 test functions,and the effects of different parameter settings on the search effect of SSSA are summarized.In the CEC2017 test suit(29 test functions in total),SSSA is compared with some frontier algorithms.According to the results,SSSA can be significantly better than other algorithms in at least 23 functions based on the Wilcoxon test in high-dimensional problems.The test on UAV path planning shows that SSSA can jump out of local optimization and find the shortest and smoothest in all compared algorithms.Furthermore,to reduce the time complexity of SSSA,a simplified version,kSSSA,is given in this paper.kSSSA has the same time complexity as SSA.On high-dimensional problems,kSSSA can be significantly better than other algorithms on at least 21 functions of the CEC2017 test set based on the Wilcoxon test.(2)To expand the SSSA into a binary algorithm and solve the binary optimization problem,such as the feature selection problem,this paper proposes three binary versions of SSSA.The three binary methods are designed based on heuristic,S-shaped transfer function,and V-shaped transfer function.Feature selection experiments on 9 UCI datasets show that the heuristic-based binary version(BSSSSA)has better search performance.On18 datasets from the UCI repository,heuristic binary SSSA is compared with four other similar algorithms.The results show that the heuristic binary SSSA is better than other methods and has application value in dealing with feature selection problems.