The Application Research Of Process Neural Network In GDP Forecast

As an important indicator measuring the economic development of a country, Gross Domestic Product (GDP) comprehensive reflects a country's economic growth rate, scale of economic development and economic trends and so on. It is not only the most important economic indicator that influences the economic life and social life, but also an important basis for a country to formulate the strategic targets of economic development and macro-economy policies. Therefore, it has great theoretical and practical significance to forecast GDP accurately. However, the traditional economic forecasting methods, such as time series method and multivariate regression method, can not describe the complex non-linear relationship between GDP and its influence factors, so the forecast accuracy is lower. And although the traditional neural network method has good non-linear characteristics and better forecast accuracy compared with above methods, it is difficult to express the time cumulative effect that exists in the GDP forecast problems, and affects the forecast results.Process neural network was used into GDP forecast against the problems that existed in the traditional forecasting methods. Process neural network is the expansion of traditional neural network in the time domain. It has the non-linear characteristics of traditional neural network. And the time aggregation'operation function added can consider the time cumulative effect of GDP time series into the forecasting, which could better solve the problems existed in the traditional forecasting methods.In the process of using process neural network into forecasting, three aspects work as follows was mainly completed:firstly, based on researching the related theories of process neural network, the algorithm was improved by introducing penalty factor and momentum factor for solving the problems that the speed of convergence slower and the algorithm easily fall into the local minimum those existed in the basic BP algorithm of process neural network. In the meantime, for reducing the network oscillation in the process of the training, adaptive learning rate was adopted to further improve the algorithm. Secondly, Heilongjiang Province was taken as an example. And by using the GDP relevant data of Heilongjiang Province from 1981 to 2009, based on the improved algorithm, and combined with the related theories of process neural network, the GDP forecasting model based on process neural network was established. And drew support from MATLAB 7.0 to completing the network training and testing, GDP was forecasted. Finally, GDP forecasting model was established by choosing the BP neural network which is the most widely used in traditional neural network and using the same training samples and testing samples. Then GDP was forecasted. And for validating the advantage of the process neural network forecasting model, the forecast results of both the process neural network model and the BP neural network model were compared and analyzed in different aspects. The result shows that process neural network used into GDP forecast can get higher accuracy and better forecast effect.