Chaotic neural network algorithm with competitive learning for global optimization

Neural network algorithm (NNA) is one of the newest proposed metaheuristic algorithms. NNA has strong global search ability due to the unique structure of artificial neural networks. Further, NNA is an algorithm without any effort for fine tuning initial parameters. Thus, it is very easy for NNA to solve different types of optimization problems. However, when used for solving complex optimization problems, slow convergence and premature convergence are its drawbacks. To overcome the two drawbacks, this work presents an improved NNA, namely chaotic neural network algorithm with competitive learning (CCLNNA), for global optimization. In CCLNNA, population is first divided into excellent subpopulation and common subpopulation according to the built competitive mechanism. Then, to balance exploration and exploitation of CCLNNA, excellent subpopulation is optimized by the designed transfer operator while common subpopulation is updated by the combination of the designed bias operator and transfer operator. Besides, chaos theory is introduced to increase the chance of CCLNNA to escape from the local optimum. To investigate the effectiveness of the improved strategies, CCLNNA is first used to solve the well-known CEC 2014 test suite with 30 benchmark functions. Then it is employed for solving three constrained real-world engineering design problems. Experimental results reveal that the improved strategies introduced to NNA can significantly improve the optimization performance of NNA and CCLNNA is a very powerful algorithm in solving complex optimization problems with multimodal properties by comparing with the other competitive algorithms. (C) 2021 Elsevier B.V. All rights reserved.

Automated summary

NNA is a metaheuristic algorithm with strong global search ability, but its drawbacks include slow convergence and premature convergence when solving complex optimization problems. To overcome these issues, an improved algorithm CCLNNA is introduced, which utilizes competitive learning and chaos theory to enhance optimization performance. Experimental results demonstrate the superiority of CCLNNA in solving complex optimization problems with multimodal properties.