基于RBF神经网络的科研绩效评价建模研究  

Scientific research performance evaluation modeling based on RBF network

作　　者：迟睿[1] 苏翔[1] 滕瑜[1] 

机构地区：[1]江苏科技大学经济管理学院,镇江212003

出　　处：《江苏科技大学学报（自然科学版）》2017年第4期525-530,共6页Journal of Jiangsu University of Science and Technology:Natural Science Edition

基　　金：国家自然科学基金青年项目(71303096);2016年校研究生教育教学改革研究与实践项目(YJG2016Y_14)

摘　　要：客观、公正、准确的科研绩效评价是调动和提高高校及科研机构科研人员工作积极性和科技创新能力的重要措施.文中提出了一种基于RBF神经网络的科研绩效精细评价模型,以归一化后的科研指标数据乘以相应权系数作为网络输入,利用优、良、中、及格和不及格5级评价作为输出,采用粒子群优化算法通过交叉验证对RBF网络结构参数进行了优化.通过RBF网络结构和输入输出特性分析发现,训练后的RBF网络权值与5级评价结果高度相关,并较5级评价结果更能精细区别科研绩效差异.该权值可直接用来进行科研绩效精细评价.文中推广了RBF网络在科研绩效评价中的应用,并为进行类似评价或评估工作提供了一种新思路.Objective,fair and accurate scientific research performance evaluation is an important measure to enhance personal work enthusiasm of scientific research in universities and scientific research institutions. In this paper,a model based on RBF neural network is employed to fine evaluate scientific research performance. The weighted normalized statistical data is taken as network input,and five rating levels,such as excellent,good,medium,pass and fail is the network output. The particle swarm optimization（ PSO） was performed to optimize the parameter of RBF network based on cross validation. After analyzing the structure and the input-output characteristics of RBF network,it is found that weights of trained RBF network and five rating levels are highly correlated,and network weights provide more detailed information than five rating levels in personal scientific research evaluation. Trained weights of RBF network can be directly used for the fine evaluation of scientific research performance. In this paper,the RBF network is generalized in the application of the performance evaluation of scientific research,which presents a new method for similar evaluation or assessment works.

关 键 词：绩效评价 粒子群优化 参数优化 神经网络权值 

分 类 号：G316[文化科学]