【摘要】 遗传算法和Broyden类的准牛顿法是近年出现的两种航天器热模型修正新技术,目前处于尝试用于工程实践的试探研究阶段。文章对两种方法的算法原理进行了初步调研,并分别分析温度和热模型不确定参数两方面的修正效果。单从温度修正结果来衡量,两种方法均能取得较好效果。但不确定参数的修正效果不佳,两种方法均无法保证不确定参数的精度、甚至只能获得丧失了物理真实性的参数解。在此基础上进一步分析了航天器热模型修正问题的定解性,指出一般情况下实际航天器热模型修正属于欠定解或过定解问题,无法得到精确的参数的反解值。据此提出工程上应成组使用修正获得的参数,并应保持同一参数在修正模型和预示模型中的一致性。最后,初步展望了利用人工神经网络深度学习进行航天器热模型修正的可能性。更多还原

【Abstract】 Genetic algorithm and quasi-Newton algorithm of the Broyden class are two novel thermal model correlation technologies used in spacecraft design, which are currently in the primary phase of engineering application. In this paper, these two methods are reviewed, focusing on their validity about both temperature and uncertain model parameters. Temperature deviation between model and test result can be successfully minimized by either of the two methods. But neither of the two methods is able to result in exact values of uncertain thermal parameters, and values obtained through correlation lose their physical significance. The determinability of thermal model correlation is then further discussed. In practice, spacecraft thermal model correlations are generally underdetermined or overdetermined problems, whose exact values of parameters are unavailable. The investigation indicates that complete set of correlated parameters should be used in further predication, and values used in predictive model should be kept identical with those from correlation. The possibility of using deep-learning of artificial neural network to correlate thermal model is proposed finally.更多还原