基于模拟退火的BP网络隐藏层节点估算算法

发布时间：2018-03-19 19:42

本文选题：BP网络　切入点：模拟退火算法(SA)　出处：《合肥工业大学学报(自然科学版)》2017年11期 　论文类型：期刊论文

【摘要】：单隐藏层BP神经网络在模式识别及数据挖掘等领域应用广泛,而隐藏层节点的数目受到众多因素的影响,因此节点数量的选取一直是一个复杂的问题。文章提出一种基于模拟退火算法(simulated annealing,SA)的单隐藏层BP神经网络隐藏层节点估算算法,基于经验确定隐藏节点数的下界,通过模拟退火不断增加隐藏节点个数直至算法结束,得到最优解。该方法与经验法和试凑法相比具有较强的理论依据,与遗传算法等方法相比不容易陷入局部最小值。实验证明,采用该方法估算隐藏层节点的准确率较高,速度也较快。
[Abstract]:Single hidden layer BP neural network is widely used in pattern recognition and data mining, but the number of hidden layer nodes is affected by many factors. Therefore, the selection of the number of nodes is always a complicated problem. This paper presents a single hidden layer BP neural network hidden layer node estimation algorithm based on simulated annealing algorithm (simulated annealing SAA), which determines the lower bound of hidden node number based on experience. Through simulated annealing, the number of hidden nodes is increased until the end of the algorithm, and the optimal solution is obtained. Compared with the empirical method and the trial method, this method has a strong theoretical basis. Compared with genetic algorithm, it is not easy to fall into the local minimum value. The experimental results show that the proposed method has higher accuracy and faster speed in estimating hidden layer nodes.
【作者单位】：合肥工业大学计算机与信息学院;
【分类号】：TP183

【相似文献】