基于禁忌搜索的神经网络集成

Neural network ensembles based on tabu search

给出了一种利用禁忌搜索来实现神经网络集成AdaBoosting算法的方法。以Ada- Boosting算法中的权值向量为优化对象,对其若干个元素进行扰动,并通过设定其为禁忌对象,以禁忌搜索控制寻优方向,以逼近误差为适值函数,在满足逼近误差或达到进化代数时结束进化。由于禁忌搜索可以避免迂回搜索,从而提高进化效率,使得算法易收敛。通过实例验证了该方法的可行性。

A method of implementation AdaBoosting algorithm with tabu search is presented. The weight vector of AdaBoosting algorithm is the optimization object, Which disturbs some of its elements and it is the tabu object. The approach error is fitness function. Tabu search controls the optimization direction, finishes the search when the approach error is contended or achieves the steps of evolution. Because tabu search can avoid outflanking search, it consequently improves the efficiency of evolution and the algorithm is easy for convergence. Example is used to verify the feasibility.