Research On Past Evaluation Method Of Provincial Natural Science Foundation Based On The Evidential Reasoning Algorithm

Past Evaluation of Provincial Natural Science Foundation(PNSF for short) is to judge the quality of projects' implementation. Objective and exact evaluation not only promote the development of science and technology, but also supply reference information for the next foundation application. There are both qualitative and quantitative evaluation approaches, and the most popular approach is peer-reviewing in the world. However this approach lies with peer-reviewers' knowledge and subjective attitude, the evaluation results may be subjective. And it will cost much on human resources, material resources and money. PNSF mainly support fundamental projects with practicality. However, evaluating these projects is a decision problem which involves multiple attributes of both quantitative and qualitative nature. There are some limitations when using AHP or peer-reviewing to deal with such decision problem. In order to handle quantitative and qualitative information, an approach named Evidential Reasoning(ER) based on an evaluation analysis model, the Dempster-Shafer theory of evidence and computer technology was developed. This approach has been applied to many decision problems and shown its probabilities in dealing with multiple attributes decision problems.In the first place, the paper analyzes the status of science and technology evaluation in both home and abroad, and points out its faults. Then the characteristics of PNSF and Evidential Reasoning approach are investigated. Secondly, by using quantitative and qualitative information collected in survey, the original evaluation indexes system is improved according to these characteristics. An evaluation model of PNSF has been developed with Evidential Reasoning approach. At the last, a numerical example of PNSF of Anhui province is examined using the ER approach. And BP-Neural Networks is used to train historical data to get the relative weights of attributes in the evaluation model. Then the evaluation results are calculated by Intelligent Decision System based on ER algorithm. The evaluation results are compared with the specialists' evaluation results and BP-Neural Networks' evaluation results, and illustrate that ER approach has good applicability in such evaluation. It uses quantitative with qualitative information well and makes the evaluation results more accurate and rational.