Cumulative learning-based competitive swarm optimizer for large-scale optimization

Competitive swarm optimizer (CSO) has shown advantages for solving large-scale optimization. However, some major problems, such as low solution accuracy and slow exploration speed, are still not effectively solved. To alleviate these problems, this paper proposes an enhanced version of CSO (shorted for CLBCSO), which uses the cumulative learning mechanism to provide promising evolutionary direction and strengthen the exploitation ability of losers. Moreover, a multi-directional learning strategy is introduced to guide the losers to explore in different directions, which can significantly improve the exploration performance of the population. CEC2014 benchmark functions, time series prediction problems and classification problem are employed to evaluate the effectiveness of CLBCSO algorithm. Experimental validation shows that the average excellent rate of CLBCSO in solving 30 CEC2014 benchmark functions with 50 variables and 100 variables is 77.08% and 79.58%, respectively. This confirms that the proposed CLBCSO algorithm is competitive compared with three CSO optimizers and five popular optimization algorithms.