摘要
针对基于单一神经网络的软件可靠性模型预测精度低和可信性差的问题,提出一种基于加权信息熵(WIE)的Real BP-AdaBoost算法。首先,用BP神经网络个体代替Real AdaBoost算法的基分类器,构建Real BP-AdaBoost算法。然后,对Real BP-AdaBoost算法的加权方式进行改进,以基分类器对训练样本的整体分类权值与基分类器对测试样本的个体分类权值的乘积作为最终的加权系数,得到WIE Real BP-AdaBoost算法。最后,通过2组软件实际失效数据对WIE Real BP-AdaBoost算法的有效性进行验证,并与SVM、BP网络、Elman网络和Real BP-AdaBoost算法进行比较研究。实验结果显示,WIE Real BP-AdaBoost算法对2组数据预测的均方误差分别为0.442 87和0.284 71,均低于4个对比模型的均方误差,说明了WIE Real BP-AdaBoost算法模型具有更高的预测精度和可信性。
Aimed at the problems that the prediction accuracy is low and the dependability of software reliability model is weak based on single neural network, a Real BP-AdaBoost based on weighted information entropy （WIE） is proposed. First, the Real BP-AdaBoost is established by taking BP neural network as the base classifier of Real AdaBoost. Then, the weighted method of base classifier of Real BP-AdaBoost is improved by utilizing the product of the overall weights of the classifier for training samples and individual weight of the classifier to test samples as the final weight, and the WIE Real BP-AdaBoost is produced. Finally, the proposed algorithm is compared with SVM, BP neural network, Elman neural network and Real BP-AdaBoost with respect to two real software failure data. The mean square error of WIE Real BP-Ada- Boost of the forecasted two sets data is 0. 442 87 and 0. 284 71 respectively, both are below the mean square error of the four comparison models. The result shows that WIE Real BP-AdaBoost is higher in prediction accuracy and reliable in dependability.
出处
《空军工程大学学报（自然科学版）》
CSCD
北大核心
2018年第1期91-96,共6页
Journal of Air Force Engineering University(Natural Science Edition)
基金
国家自然科学基金(61402517)
