Research And Implement Intelligent Forecast System Based On ANN

Artificial neural network (ANN) is relatively crude electronic models based on the neural structure of the brain, and composed of a large number of highly interconnected processing elements (neurons). Although artificial neural network can not exactly show the really function of the brain it can abstract, predigest and simulate the living neural network to some extend, and has a considerable advantage on the forecast of economic system. Back-Propagated network (BP Network) is a kind of neural network;it can be used in the forecast of economic system, and has acquired brilliant achievement. But the classical BP algorithm has its intrinsic drawbacks;it also has some limits in the application of the forecast. This thesis is based on the theory of BP network;improve this algorithm against its limits, and implement an intelligent Communal web-based forecast system.Theoretically, in this paper, we use the Levenberg-Marquardt algorithm to optimize the modifying formula of weights and threshold so as to speed up the training step and avoid local convergence;we also Optimize the hidden layer units to search the Best number of cells;Principal component analysis (PCA) and confidence interval algorithm are adopted to optimize the training data and the structure of the network for the high accuracy and generalization capability improvement.On the design and implementation, we use Matlab to implement the ANN so as to guarantee the efficiency of the network;the Web-based system is developed by Asp.net to improve the facility;we also adopt the COM technique to realize the data transmission between Matlab based ANN and web system, where the Matlab based ANN is considered to be a web service, achieving the intergradations. In the data management, we use XML to collect and store the sample data and forecast model. In our prototype, these ideas are not only feasible, but also have better expandability.In practice of our prototype, take the forecast analysis about the ability of Chinese continuable development as an example, we use the true data set to test the system in the following aspects: sample management, foundation and management of the forecast model, neural network forecast and result analysis and so on;On the other hand, we adopt complex function fitting to examine the generalization capability of the forecast system and compare with different forecasting systems in different point of views. The result indicates that the ANN intelligent web-based forecast system is more accurate and efficient than the classical ones and regression forecast algorithms, furthermore, this system is more manageable, it has wonderful foreground on the forecast practice and research in different domains.