【作者】 崔天意；

【导师】 张利军；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 舰艇装备软件的可靠程度对我军现役舰艇作战生存率有重大影响,并且对其服役期限的确定有重要作用。随着军队现代化水平的不断提高,军队武器自动化程度呈现了前所未有的进步和发展,现代化军队武器装备系统作为一个大型的软件密集型系统,软件的稳定、可靠直接关系着军队装备的成败。目前国内海上靶场针对舰载武备系统软件专项试验尚属起步阶段,软件可靠性测试尚未列入验收项目目录,靶场还没有建立起完善的软件可靠性测试和预测方法。为全面准确地对新一代舰艇武备系统考核,论文作者结合多年海上试验经验,参与制定了适应于新装备发展的舰艇装备软件可靠性测试和试验方法并且研究了其可靠性预测手段,为舰艇装备软件的验收工作提供方法和标准。论文在介绍软件可靠性和软件可靠性工程的基础上,分析了多种常用的软件可靠性模型,研究了传统可靠性预测技术。传统的可靠性预测方法大多是基于软件故障时间的,论文将软件可靠性预测分两个部分进行研究。第一个部分使用了不依赖于软件故障时间记录的静态预测方法。此方法适用于装备软件等现场测试危险性较高,数据难于记录,故障时间难以确定的软件的可靠性预测。在软件可靠性静态预测中,可以近似认为相似软件等量时间内的失效数据可以选用同一种可靠性模型预测,所以可以根据已知的软件测试结果,分析软件服役期故障数历史数据,使用BP神经网络预测同类软件的故障数和可靠度。仿真实验结果与实测数据对比表明BP神经网络静态预测方法是有效可行的。第二个部分是在依据软件故障间隔时间的软件可靠性动态预测中,分析了BP神经网络动态预测方法,并且将其与传统可靠性预测模型进行对比。仿真试验结果表明,BP神经网络软件可靠性预测模型简单易用,并且达到传统可靠性预测模型的估测能力。为了提高预测效率和精度,论文使用基于BP神经网络的聚类方法来选择最佳传统软件可靠性模型进行软件可靠性预测。仿真实验表明,BP神经网络模型选择方法达到目前广泛使用的高斯混合模型选择方法的效果,但其计算相对更简单、方便。现役舰艇装备软件在通过测试之后,各项技术指标可以达到预定要求,但不一定具有较高的作战效能。即软件技术测试合格的系统仍然不能实际应用。因为软件测试只注重软件本身的故障特性,而忽略了对于装备软件至关重要的软硬件配置拟合度和战术合理性等非技术性指标。论文中结合舰艇装备软件自身的特点,选取特定的可靠性预测参数和特定的可靠性测试结果,收集并整理出舰艇装备软件作战效能数据,建立了舰艇装备软件作战效能评估模型。最后使用并联分级构造方法建立了由静态预测、动态预测和作战效能预测三部分组成的舰艇装备软件可靠性的神经网络预测评估模型。更多还原

【Abstract】 Warship furnishment software reliability research affect warship’s life in battlefield, and can ensure its service time limit. The army weaponry automatic level make progress and development quickly than formerly. The modernistic weaponry is large dense software system. Its software stability and reliability make certain furnishment quality. The technical testing starts barely for warship furnishment at present in the navy testing base China. To comprehensively and accurately on a new generation of warships Wubei appraisal system, I for many years under the sea trial experience in the development of new equipment adapted to the development of ships and equipment software testing and test methods and on the reliability of forecasting tools, vessels and equipment inspection software provide methods and standards.Papers presented in the software reliability on the basis of analysis of a wide range of commonly used software reliability model, the classic reliability prediction techniques, and a brief introduction of the software reliability engineering. The reliability of traditional forecasting methods are mostly software-based fault of the time, papers will software reliability forecast in two parts for research. The first part is not dependent on the use of a software fault time record of static forecasting methods. This method can be applied to software and other equipment at a higher risk of testing, data difficult to record, it is difficult to determine fault time the reliability of the software forecast. On the software reliability static forecast, similar to software that can approximate equivalent of the time lapse data can be selected with the reliability of a model projections, the software can be the basis of known test results, analysis software failures of service period of historical data, the use of BP Neural networks predicted the failure of similar software and reliability. The simulation results contrast with the measured data show that BP neural network static prediction method is feasible and effective. The second part is based on software in the time between failures of the software reliability dynamic forecasts, BP neural network analysis of the dynamic forecasting methods, and its reliability forecasts with the traditional model of comparison. The simulation results showed that, BP neural network model predicted easy-to-use software reliability, and than most traditional model of a better estimate capacity. In order to improve the efficiency of forecast, the use of paper-based BP neural network clustering method to choose the best model of traditional software reliability software reliability forecasts. The simulation shows that, BP neural network performance for the current widespread use of the Gaussian mixture, but their relative terms more simple, convenient.Active-duty ships pass the test equipment software, the technical indicators can be achieved scheduled requirements, but not necessarily with high combat effectiveness. Software technology that tested system is still not practical application. Because software testing focused only on the fault of the software itself, and neglect the essential software and equipment for the hardware and software configuration of the fitting and tactical reasonable, and other non-technical indicators. In this paper, in combination with its vessels and equipment characteristics of the software, select the specific parameters of the reliability of specific reliability test, to collect and collate the ship equipment software combat effectiveness data sets, software and equipment to establish the ships operational effectiveness evaluation model. Parallel Application of the final classification structure established a network of BP ships and equipment reliability of the software neural network predictive assessment model framework.更多还原