【摘要】 为改善前向神经网络的性能,将禁忌搜索作为前向神经网络的训练算法,采用了一种集中性与多样性的自适应搜索策略,以提高禁忌搜索的有效性.并以正弦函数和sinc函数的逼近为例,验证了算法的可行性和有效性.更多还原

【Abstract】 To improve the performance of FNN, this paper attempted to employ a new global optimizing algorithm——Tabu Search (TS) in FNN for its training, and a novel adaptive search strategy of intensification and diversification was proposed to improve efficiency of TS. Taking function approximation as samples, many simulating experiments were carried out. The result shows: TS is much better than BP algorithm in function approximation, especially in the functions that their nonlinear degrees are very higher.更多还原