Research On Comprehensive Evaluation Of Science And Technology Strength In Fujian Province

The comprehensive strength of a country can be represented by Scientific and techno-logical strength,and the evaluation and research system of the strength of science and technology is becoming more and more perfect.However,the strength of science and technology embodies a complex system composed of many factors.The evaluation of science strength needs to be adapted to local conditions and changes with time.Through scientific and objective a.nalysis of the scientific and technological strength of Fujian Province,this paper researches the development trend of the scientific and technological strength of Fujian Province and the development situation of every cities,so as to achieve the 13th "Five-Year Plan" for Fujian's scientific and technological development as soon as possible.In this paper,we use multi-dimensional time series analysis method to evaluate the comprehensive scientific and technological strength of Fujian Province from four aspects:scientific and technological input,scientific and technological output,educational resources and economic benefits,which conclude 24 evaluation indicators of Fujian Province from 2002 to 2017.Firstly,the data are preprocessed to fill the missing data and eliminate the dimension;secondly,the weight of each index is determined by using entropy method and analytic hierarchy process respectively;thirdly,we Calculate the comprehensive evaluation value of scientific and technological strength of various cities in Fujian Province,and study the similarities and differences between cities;finally,under the framework of BP neural network,through the direct multi-step prediction algorithm with forgetting factor combined with time series difference method.The target variables are forecasted and feasible suggestions are put forward for the future development of science and technology in Fujian Province.In direct multi-step prediction,the expected output value can not be obtained in time,which shows its limitations.The sequential difference algorithm using the difference of actual output to drive the training of the network perfectly avoids this problem.At the same time,the characteristics of the current data can't be fully reflected by historical data,a learning algorithm with forgetting factor is introduced in our model.The network structure is also studied while training the network to ensure the accuracy of the prediction results.