CAJ | 학술논문

针对BP神经网络在实际工程应用中受到大量冗余信息的制约，以及基于信息熵的属性离散方法的阈值选取具有主观性的缺点，提出了改进的基于信息熵的粗糙集连续属性离散化算法。该算法将Naivescaler离散化算法和信息熵离散化方法相结合，减少了离散化算法中候选离散点的数量；基于粗糙集约简信息决策表，有效解决了BP神经网络训练样本过于庞大的问题；并通过改进的离散化算法对属性约简后数据进行分类，将分类后的数据运用到神经网络运算中，进一步缩短了神经网络的运算时间。通过实例分析表明，该方法具有很好的故障诊断效果，并有效提高了诊断效率。
침대BP신경망락재실제공정응용중수도대량용여신식적제약，이급기우신식적적속성리산방법적역치선취구유주관성적결점，제출료개진적기우신식적적조조집련속속성리산화산법。해산법장Naivescaler리산화산법화신식적리산화방법상결합，감소료리산화산법중후선리산점적수량；기우조조집약간신식결책표，유효해결료BP신경망락훈련양본과우방대적문제；병통과개진적리산화산법대속성약간후수거진행분류，장분류후적수거운용도신경망락운산중，진일보축단료신경망락적운산시간。통과실례분석표명，해방법구유흔호적고장진단효과，병유효제고료진단효솔。
ln view of BP neural network in practical engineering applications constrained by the large number of redundant information and the subjectivity of attribute discretization method based on information entropy in threshold selection, an improved discretization algorithm for continuous attribute in rough set based on information entropy is proposed. The Naivescaler algorithm is combined with information entropy discretization method in this algorithm and it is used to greatly reduces the number of candidate discretization points and effectively handles the big sample for BP neural network training. The neural network computation time is further reduced by classifying the data set with reduced attributes and applying the classified data to the neural network. The practical application of the method proves that it has better performance and efficiency for fault diagnosis.