【摘要】 时间序列预报是神经网络的一个应用领域,多数研究集中在神经网络直接预报方面。本文从故障预报角度研究了神经网络组合预报模型,由神经网络给出常规预报方法的最佳组合。首先从函数逼近角度阐述这种模型的理论依据,在此基础上给出了模型的评价指标和神经网络的有效训练算法,最后给出在空间推进系统上的应用实例。更多还原

【Abstract】 Time series prediction is an application domain of neural network. Most studies on it are focused on direct prediction. In point of fault prediction, neural network based combination prediction is studied in the paper in which the best combining of traditional prediction methods is given by neural network. Firstly the theoretical foundation of the model is stated in sense of function approximation. Secondly, the appraisal criterion of the model and an efficienty training algorithm for neural network are given. In the end, an application example in propulsion system of Spacecraft is introduced.更多还原