基于深度信念网络的集成分类器在气体识别中的应用  

Application of Ensemble Classifier Based on Deep Belief Network in Gas Identification

作　　者：王春香[1] 李丽宏[1] 张帝 

机构地区：[1]太原理工大学信息工程学院,太原030024 [2]国家电网山东省电力公司莘县供电公司,山东聊城252000

出　　处：《计算机工程》2016年第10期318-321,共4页Computer Engineering

摘　　要：为降低信号漂移对于气体识别的影响,提出一种基于深度信念网络(DBN)的集成分类器模型。利用不同时段的数据集训练单个DBN分类器,将得到的分类器对数据集进行分类,通过使分类误差最小得出每个单一分类器的最优集成权重,采用粒子群优化寻找最优权重并对所有分类器进行集成得到最终的气体识别结果。使用由4种气体传感器组成的传感器阵列对该方法和均匀加权DBN、最优支持向量机方法进行性能对比。实验结果表明,该方法能在较长时间里保持较高的分类准确率,在一定程度上抑制了信号漂移对分类结果的影响。In order to reduce the influence of signal drift on gas identification, an ensemble classifier model based on Deep Belief Network（DBN） is proposed. A single DBN classifier is trained by different periods of datasets and the obtained classifier is used to classify the datasets. By minimizing the classification error, the optimal integrated weights of each single classifier are obtained. Particle Swarm Optimization （PSO） is used to find the optimal weights and integrate all the classifiers to get the final gas identification results. The performance of the proposed method is compared with that of the uniform weighted DBN and the optimal Support Vector Machine（SVM） by using a sensor array composed of 4 gas sensors. Experimental results show that, this method can maintain higher classification accuracy in a long time. To a certain extent, the effect of signal drift on the classification results is suppressed.

关 键 词：气体传感器阵列 漂移补偿 深度信念网络 集成分类器 气体识别 网络训练 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]