Abstract: Firstly some qualitative analysis methods such as literature research and network investigation are applied to find out all the possible factors influencing scientific and technological(S&T) outputs, and considering data availability, collect all related data to S&T productivity and their influencing factors for the period 1996-2008. Then based on the collected data, a bivariate correlation analysis method is utilized to analyse the mutual relations between S&T outputs and their influencing factors, and with the multiple linear regression method selecting the high-influencing factors to construct a model analyzing influencing factors and prediction for S&T outputs. Lastly based on the results of bivariate correlation analysis, a currently prevalent BP neural network prediction method is used to do a prediction study on S&T outputs, and compare the predictive performance with that of multiple linear regression method.

摘要： 笔者首先通过文献研究和网络调查等定性分析方法梳理出科技产出能力的所有可能的影响因素，并在数据可获得性的前提下，以1996-2008年为时间维，采集科技产出能力及其影响因素的相关数据，然后对科技产出能力及其影响因素之间的相互关系进行二元相关分析，并利用多元线性回归分析方法从所有相关因素中筛选出影响程度较高的因素，构建科技产出能力的影响因素分析与预测模型。最后基于二元相关分析的结果，选择相关程度较高的因素，利用目前流行的BP神经网络预测方法对科技产出能力进行预测研究，并与多元回归分析预测模型的预测性能进行比较。

关键词: 科技产出 , 影响因素分析 , 多元线性回归分析 , BP神经网络 , 二元相关分析 , 预测 , PCT专利申请 , SCI论文产出 , Scientific and technological outputs, Impact factors analysis, Multiple linear regression, BP neural network, Bivariate correlation analysis, Prediction, PCT patent applications, SCI papers productivity