基于PSO算法的DBN模型参数优化

Weight parameter optimization of depth belief network model based on genetic and particle swarm optimization

针对深度置信网络模型每层神经元个数难以确立的问题,提出利用粒子群寻优确立DBN网络每层节点数,利用Kmeans聚类来决定是否需要增加隐藏层的方法来确立DBN的网络结构。该算法根据粒子群寻优算法以最小化所有样本重构误差的平方和为目标函数来确定DBN每层神经元个数,以确定DBN的初步结构,为了验证DBN结构的有效性,利用DBN提取的数据特征来进行聚类测试,进一步根据聚类结果来修正DBN,以获得DBN的最佳结构,以红酒数据集进行分类实验,实验结果表明,与传统未经改进的深度置信网络进行对比,发现该方法确立的深度置信网络分类效果更优。

To solve the problem that it is difficult to establish the number of neurons in each layer of the deep confidence network model,this paper proposes to use particle swarm optimization to establish the number of nodes in each layer of DBN network,and K-means clustering to determine whether to add hidden layer to establish the network structure of DBN.According to the particle swarm optimization algorithm,the number of neurons in each layer of DBN is determined to minimize the square sum of all reconstruction errors of samples as the objective function,so as to determine the initial structure of DBN.In order to verify the effectiveness of DBN structure,the data features extracted by DBN are used for clustering test,and the DBN is further modified according to the clustering results to obtain the best structure of DBN.In this paper,red wine data is used The experimental results show that the depth confidence network established by this method is better than the traditional depth confidence network.