基于改进贝叶斯正则化BP神经网络的测试性评估技术研究

摘要

Evaluation of testability is important for testability, which can show the accuracy of design plan, find the disadvantages and check parameters. For lack of efficient evaluation methods, a new means is given. On the basis of system of testability evaluation, this method takes advantage of BP neural network to build basic model, brings training method of BR to improve ability of BP, and provide the AMC-QPSO to optimize initial weights and thresholds of BP. According to the result of experiment, this method can be used for design for testability of equipment, which is accurate and applicable.%测试性评估是整个测试性设计工作的关键环节,能够反映设计方案的正确性、识别设计缺陷以及检验设计要求；针对当前有效评估手段的缺乏,本文提出了一种新的测试性评估方法；在建立测试性评估指标体系的基础上,利用BP网络解决非线性问题的优势,建立了评估模型,同时采用贝叶斯正则化(BR)训练方法,提高网络的泛化能力,并提出带自适应变异的质心量子粒子群(AMCQPSO)算法,以优化BP网络初始权值和阈值,以提高评估模型结果的准确度；经实验验证,评估方法具有较高的准确性和实用性,能够更好地为装备测试性设计服务.