An adaptive hybrid differential evolution algorithm for continuous optimization and classification problems

Differential evolution (DE) is recognized as a simplistic yet robust evolutionary algorithm: it has been utilized to tackle different challenging optimization problems in various science and engineering disciplines. DE has some disadvantages, such as premature convergence and slow convergence rate, leading to the worst DE execution arrangement. Two DE variations named adaptive parameter selection-based DE (APSDE) and chaotic map hybridization based on DE and PSO (CMHDE-PSO) have been proposed to tackle the issues mentioned above. The proposed variants contain three unique advantages for APSDE: (1) new population initialization scheme to keep up the decent variety of the population diversity; (2) controlled mutation factor technique followed by adaptive decreasing parameter selection procedure; (3) novel mutation strategy with a specific weighted pattern to determine the mutant vector for the mutation operation. Similarly, for CMHDE-PSO (1) novel distribution called Torus for the selection of initial population located in the search space; (2) new parameter adoption technique based on chaotic circle maps defined by chaos theory; (3) average pattern means of two different mutation strategies; (4) and lastly the hybridization of proposed improved DE with PSO to supports DE escaping local minima. Both APSDE and CMHDE-PSO are compared with several standard non-DE old-fashioned optimization algorithms and various advanced DE variants. We accomplish definite experiments behind the powerful searching technique by applying the APSDE and CMHDE-PSO-based mutation and parameter selection strategy for the function optimization and weight optimization of feed-forward neural networks (FFNN) on real-world data classification problems. For data classification performance evaluation, 10 data sets are utilized from the repository of UCI machine learning. Experimental results showed that APSDE and CMHDE-PSO extensively beat different EAs in all test functions and obtained higher accuracy with the recent state-of-the-art algorithms for weight optimization.

Automated summary

Differential evolution (DE) is a simple yet robust evolutionary algorithm used for optimization problems. Two variations, APSDE and CMHDE-PSO, have been proposed to improve DE's convergence and execution performance. Experimental results show that both APSDE and CMHDE-PSO outperform other algorithms in function optimization and weight optimization for neural networks.