Study On Optimized Grey Neural Network Prediction Models

Gray neural network model has been a relatively wide range of applications with the industry needs for small sample of data information processing.Although the neural network and gray system theory are applied widely in information processing, the results of prediction data are unsatisfactory.Of course,established gray neural network models of the integration of these two methods have good data processing and prediction,and it can make up the lack of only using anyone of them.Gray neural network model is simple,if to optimize the learning algorithm and the model structure and performance,we are able to achieve better results.At present,many scholars do considerable researches about optimizing gray neural network and made out some achievements.Optimized gray neural network model can deal with uncertain information and technical data to improve forecast accuracy,it has become a very important topic.At first,the paper states the characters of small sample data,and then analyzes the complexity and the particularity of estimating small samples,two methods are proposed by comparison and validate,and the thinking,which are based on grey prediction model and neural network technology,is adopted.In the paper,a SGRBF static model is established on the basis of RBF(Radial Basis Function) and Grey Model(0,N).The model can deal with the prediction problem very well,because it makes use of the RBF's good ability of in approaching nonlinear function,and the accuracy of Grey Model(0,N) in making a prediction of small sample data.A DGRBF dynamic model is also established in the paper,which can select the best initialization conditions and dynamic identifying parameters and is fit for dynamic and long-term data prediction.When combining grey system with RBF neural network local optimization and convergence problems are still existed,so genetic algorithm is introduced to assist the modeling of grey neural network in this paper.Firstly,genetic algorithm is employed to solve the parameters of improved GM(1,1) with Lagrange's Mean Value Theorem, two new dynamic prediction models integrating genetic algorithm and grey RBF,one is a grey RBF compensation prediction GA-GRBF model,the other is inlaid grey neural network GRBF model.The new models with preferable structure and parameters are applied to simulation and analysis of time-displacement data of wind response.The comparative experiment results show that this model is capable of predicting a small sample of data accurately,easily and conveniently.