【摘要】 人工神经网络（ＡｒｔｉｆｉｃｉａｌＮｅｕｒａｌＮｅｔｗｏｒｋ，ＡＮＮ）对于处理非线性体系建立模型有其独到之处，尤其近些年来将人工神经网络用于药物定量构效关系（ＱｕａｎｔｉｔａｔｉｖｅＳｔｒｕｃｔｕｒｅ－ＡｃｔｉｖｉｔｙＲｅｌａｔｉｏｎｓｈｉｐｓ，ＱＳＡＲ）的研究已取得了令人欣慰的结果，这就使得越来越多的药物学家和化学家相信并采用这一方法。但是值得关注的是在应用ＡＮＮ方法分析和预报中存在着过拟合（Ｏｖｅｒｆｉｔｔｉｎｇ）的现象，这将影响模型的预报性能。为此本文提出了一种消除过拟合现象的方法，这种方法保证所得出模型在一定范围内有较可靠的预报能力（ＰｒｅｄｉｃｔＡｂｉｌｉｔｙ）。更多还原

【Abstract】 Artificial Neural Network ( ANN ) is a powerful tool to construct model for processing nonlinear system. Especially in recent years the application of ANN in Quantitative Structiire-Activity Relationships (QSAR) has obtained good results, in which the chemists and pharmaceutists are interested.However it is notable that the overfitting phenomena could be caused in analysing and forcasting. This is a serious problem, in order to solve it, a convergence criterion and a method to optimize the net structure are provided. This approach can ensure the model have a reliable predict ability.更多还原