基于BP神经网络的中国省域知识溢出实证研究

Empirical study about knowledge spillover in Chinese province area based on BP neural network

按综合科技进步水平将全国31个省市分为五类,综合以往的文献资料并结合BP神经网络工作原理确定了6个影响和制约知识溢出因素的指标作为神经网络的输入变量,提出了基于BP神经网络的知识溢出分析模型;经测算后,取精度较好的两组检验数据,对其各因素进行敏感性分析。结论为影响科技进步水平较高的地区与北京间知识溢出的最重要的因素为创新能力、信任水平;而对科技进步水平较低的地区,影响与北京间的知识溢出最重要的因素为结构相似性水平、吸收能力。

In this paper, in line with the level of scientific and technological progress ranked by China Ministry of Science and Technology, we divide China＇ s 31 provinces and municipalities into five categories. Through analysis, we extract 6 key factors which con- straint and impact the knowledge spillover as neural network input variables by synthesizing the past documentations with BP neural network working principles . we propose a new knowledge overflow analysis model based on variable learning rate backpropagation （VLBP） neural network , train and test the model. Then we take sensitivity analysis for the factors of Jiangsu and Hebei provinces whose test results are relatively more accurate. The Conclusions are that： the most important factors that affect the knowledge spillover between Beijing and the places with higher level of scientific and technological progress are the Innovation Capability and the Trust whereas for the regions with relatively lower level of scientific and technological progress ,the Structural Similarity and Absorption Capability play more important role in inter- Beijing knowledge spillover.