关键词: 麻雀搜索算法, 低差异序列, 高斯变异, 逐维反向学习策略, 函数优化

Abstract: Sparrow search algorithm (SSA) is easy to fall into the local optimal point in the search process, resulting in a single population diversity in the late iteration stage, so multi-strategy sparrow search algorithm (MSSSA) is proposed in this paper. In practice, random number sequences are widely used to generate the initial population in the search space, but the distribution uniformity of the initial population by this method is not good, and the exploratory nature of the initial population to the solution space cannot be guaranteed.
Firstly, the low-discrepancy sequence and chaotic mapping are used to generate a uniform initial population to ensure the uniformity of the initial population, so as to accelerate the convergence rate of the algorithm and improve the ability of the algorithm to jump out of the local optimal value. The producer of the SSA, as the main means of exploring the search space of the algorithm, does not balance local search and global search well. In order to solve this problem, the Euclidean distance and nonlinear inertia weight information between the current sparrow position and the upper and lower limits of the search are fully used to balance the accuracy and breadth of the producer’s search. In order to improve the ability to jump out of the local optimum, the random unit vector with uniform distribution of the d-dimensional hypersphere is introduced, which fully improves the random walking nature of the population.The scroungers strategy of the SSA makes most of the followers only follow the current optimal sparrow, sacrificing a certain global search ability, and is easy to fall into the local optimal value.
For preventing the loss of the current optimal search information, the elite strategy is introduced to preserve the optimal individual and improve the convergence speed and accuracy. When the former optimal individual is too close to the coordinate origin, the Gaussian variation range is limited, and it plays a very limited role in the process of jumping out of the local optimum. Although the reverse search strategy of one-dimensional oppositional learning reduces the interference between dimensions and expands the search area of the algorithm, it is difficult to improve the optimization accuracy of the algorithm in the late stage of population iteration. Therefore, the Gaussian mutation method and the one-dimensional oppositional learning are comprehensively improved, and the improved Gaussian mutation and improved one-dimensional oppositional learning are used to perturbate and accurately search at the position of the optimal solution, so as to improve the performance of the algorithm.
Under the same experimental conditions, the proposed algorithm (MSSSA) is compared with 9 algorithms on 24 benchmark functions, 50-dimensional and 100-dimensional CEC2017 test functions, and the excellent performance of the proposed algorithm in optimization ability is proved by the Wilcoxon signed ranks test. In order to verify the effectiveness of the comprehensive strategy, model ablation experiments are carried out on the algorithm, and the comprehensive effectiveness of the proposed improved strategy is proved. The comparison with the practical engineering application of 9 algorithms in the classical pressure vessel design problem also shows that the MSSSA has better optimization ability, effectively saves production costs, and has certain practical application value.

Key words: sparrow search algorithm, low-discrepancy sequences, Gaussian mutation, one-dimensional oppositional learning, function optimization