Analysis Of EVA Driving Factors Based On Neural Network

Large scale special equipment manufacturing enterprises had high manufacturing level,high Industry Association level and the leading technology.Large scale special equipment manufacturing enterprises had become one of the most important forces in the national economic construction.Under the background of "intelligent manufacturing 2025" strategy,large manufacturing enterprises play an important role in leading the Chinese manufacturing industry towards information and intelligence.Economic value added(EVA)as a key indicator of total factor productivity,can fully reflect many aspects of enterprise management.After 2014,the state required large central enterprises to draw the ''value tree'' of the EVA,from which can identify the key factors of EVA.This made the researches on the economic value added up to a new height.It is obvious that find the influence factors of economic value added at different levels was more practical.Usually,researches on the influencing factors of the economic value added were mostly based on the multiple linear regression analysis.This method was characterized by fast calculation and easy operation.But large-scale special equipment manufacturing enterprises were complex production system.There was complicated nonlinear relationship between the factors of economic value added.At the same time,in order to avoid the problem of multicollinearity,multiple linear regression analysis method was difficult to include more variables.In order to find a better method,this paper firstly used the single hidden layer BP neural network and then constructed a double hidden layer BP network to analyze the influence factors of EVA of large scale special equipment manufacturing enterprises.The analysis results showed that the partial sensitivity of BP neural network was unsteadiness.No matter the BP neural network had one or two hidden layer.It could only get the overall trend of sensitivities of the influence factors of EVA.This paper used the generalized regression neural network(GRNN)which approximation ability and the learning rate were better than the BP network,and combining with the information entropy theory to solve the above problem.The results showed that the sensitivity of the GRNN network with second order Renyi entropy was similar to that of the double hidden layer BP network,but the sensitivity stability of the GRNN with second order Renyi entropy was better than that of the two hidden layer BP network.And the generalization of the results obtained from the GRNN network was better than the multiple linear regression analysis.GRNN network correctly captured the complex nonlinear relationship between the various factors.And the number of variables included in the GRNN was also superior to multiple linear regression analysis model.