利用并行神经网络进行航天软件质量评价  

Quality evaluation of aerospace software using parallel neural networks

作　　者：宋元章[1] 沈湘衡[1] 李洪雨[1] SONG Yuanzhang;SHEN Xiangheng;LI Hongyu(Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China)

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《哈尔滨工程大学学报》2020年第4期595-600,共6页Journal of Harbin Engineering University

基　　金：国家高技术研究发展计划项目(2011AA7031024G);国家自然科学基金项目(61133011).

摘　　要：针对航天软件质量评价准确率不高、受主观因素影响较大、可扩展性较弱等问题,本文结合神经网络和DS证据理论,提出了一种利用并行神经网络进行质量评价的方法。将选取的软件质量评价指标数据分别输入到多个相互独立的并行的神经网络中以获得多个初步评价结果,利用DS证据理论对各初步评价结果进行融合获得最终评价结果。以某航天相机系统为实验对象,测试本文方法的有效性,实验结果表明:评价准确率可以达到95.23%,训练时间为576.00 ms,评价处理时间为77.50 ms。本文方法评价准确率较高、训练时间和评价时间较短,满足对航天软件进行质量评价的要求。Aiming at the quality evaluation of aerospace software systems that are strongly influenced by subjective experts and poor scalability resulting in low evaluation accuracy,a quality evaluation method based on parallel neural networks is proposed that effectively combines a neural network with DS evidence theory.Selected software quality evaluation indexes are input into independently parallel neural networks to obtain preliminary evaluation results.The DS evidence theory is then used to combine the preliminary evaluation results for the final evaluation.The proposed method is validated in the quality evaluation of an aerospace camera system.The accuracy of the proposed method reached 95.23%,the training time was 576.00 ms,and the evaluation processing time was 77.50 ms.With advantages of higher accuracy,less processing time for training and evaluation,the method satisfies the quality evaluation requirements of an aerospace software system.

关 键 词：航天 软件 质量评价 并行神经网络 DS证据理论 决策准则 交叉验证法 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]