Evaluation of new sparrow search algorithms with sequential fusion of improvement strategies

Sparrow search algorithm (SSA) is a novel swarm intelligent algorithm inspired by foraging and anti-predation behaviors of the sparrow population. However, the population diversity of the basic SSA decreases in iterations and it tends to fall into the local optimum. In this paper, we propose five improved sparrow search algorithms (ISSAs 1–5) by sequentially integrating the five strategies of improved sine mapping, elite opposition-based learning, sine cosine algorithm, Lévy flight, and Gaussian mutation to enhance SSA performance. In other words, the nth ISSA adopts the first n above improvement strategies. An experiment based on 23 classical benchmark functions (F1-F23) is conducted to test the performance of proposed ISSAs. Based on a comprehensive comparison with existing algorithms, it is found that ISSA 5 which fuses all five improvement strategies has the best performance, and the ISSA performance increases as more improvement strategies are integrated. For F1-F13, ISSA 5 outperforms all other algorithms for 69.23% (9/13), 76.92% (10/13), 61.54% (8/13), 76.92% (10/13), 69.23% (9/13), 61.54% (8/13) of the functions at 10, 30, 50, 100, 500, 1000 dimensions respectively. For fix-dimension functions F14-F23, ISSA 5 is top-ranked for 70.00% (7/10) of the functions. Also, it is found that the sine cosine algorithm and Lévy flight strategies have greater impact on the improvement of SSA among the five strategies. Moreover, further analysis by Friedman test and Nemenyi post-hoc test reveals the superior performance of the proposed ISSAs in a statistical sense against other competing algorithms. As such, the results indeed postively demonstrate the efficiency, robustness, convergence and practical applicability of the proposed ISSAs.