深度信念网络的等效模型及权值扩展算法研究

发布时间：2018-03-07 11:32

本文选题：深度信念网络　切入点：等效模型　出处：《电测与仪表》2017年23期 　论文类型：期刊论文

【摘要】：针对深度信念网络(DBN)中小样本情况下,训练模型泛化性较差,分类识别率不够理想,系统性能有待提高等问题,研究了DBN的等效模型,分析了小样本情况下识别率差的问题;并提出一种区间化权值扩展方法,扩大了样本和权值的匹配空间,使判决更有利于正确分类,提高了小样本情况下的图像分类准确性;用检测与估值理论给出了算法能提高系统检测性能的依据,并在不同的数据库上进行了实验测试,进一步证明了小样本情况下图像分类准确率的提高。最后,将该方法应用到了小样本绝缘子故障识别中。
[Abstract]:In view of the problems of poor generalization of training model, poor classification recognition rate, and system performance to be improved, the equivalent model of DBN is studied, and the problem of poor recognition rate under small sample is analyzed. An interval weight extension method is proposed, which expands the matching space between samples and weights, makes the decision more favorable to the correct classification, and improves the accuracy of image classification in the case of small samples. Based on the theory of detection and estimation, the basis of the algorithm to improve the detection performance of the system is given, and the experimental results are carried out on different databases, which further prove the improvement of the accuracy of image classification in the case of small samples. The method is applied to fault identification of small sample insulators.
【作者单位】：华北电力大学电气与电子工程学院;
【分类号】：TP181;TP391.41

【相似文献】