Artificial neural network regression as a local search heuristic for ensemble strategies in differential evolution
Artificial neural network regression as a local search heuristic for ensemble strategies in differential evolution
复制标题
DOI：
10.1007/s11071-015-2537-8
复制
发表时间：
2016-04-01
影响因子：
5.6
通讯作者：
Strnad, Damjan
中科院分区：
文献类型：
--
作者：
Fister, Iztok;Suganthan, Ponnuthurai Nagaratnam;Strnad, Damjan
大自然经常成为开发新算法来解决具有挑战性的现实世界问题的灵感。数学建模促进了人工神经网络 (ANN) 的发展，事实证明，人工神经网络对于解决分类和回归等问题特别有用。此外，受达尔文自然进化论启发的进化算法（EA）已成功应用于解决优化、建模和模拟问题。差分进化 (DE) 是一种特别著名的 EA，它拥有多种生成后代解决方案的策略，但事先并不知道最佳策略。在本文中，ANN 回归被用作 DE 算法中的局部搜索启发式算法，尝试预测最佳策略或尝试从 DE 策略集合中生成更好的后代。这种局部搜索启发法根据控制参数应用于解决方案群体，该控制参数在计算的时间复杂度和解决方案的质量之间进行调节。由 30 个基准函数组成的 CEC 2014 测试套件上的实验揭示了开发这一想法的全部潜力。
Nature frequently serves as an inspiration for developing new algorithms to solve challenging real-world problems. Mathematical modeling has led to the development of artificial neural networks (ANNs), which have proven especially useful for solving problems such as classification and regression. Moreover, evolutionary algorithms (EAs), inspired by Darwin's natural evolution, have been successfully applied to solve optimization, modeling, and simulation problems. Differential evolution (DE) is a particularly well-known EA that possesses a multitude of strategies for generating an offspring solution, where the best strategy is not known in advance. In this paper, the ANN regression is applied as a local search heuristic within the DE algorithm that tries predicting the best strategy or attempting to generate a better offspring from an ensemble of DE strategies. This local search heuristic is applied to the population of solutions according to a control parameter that regulates between the time complexity of calculation and the quality of the solution. The experiments on a CEC 2014 test suite consisting of 30 benchmark functions reveal the full potential in developing this idea.