Fujian Economic Forecastingbased On Deep Belief Network

Making accurate predictions of macroeconomic trends is significant for analyzing the pros and cons of macro-control policies,evaluating the movement of economic system and making correct development strategy.In recent years,deep learning method is popular in machine learning area.So this paper introduces the Deep Belief Networks to the field of economic forecasting,deal with the lack of traditional artificial neural network in economic forecasting,such as the slow training speed,falling into local minimum,over-fitting,poor generalization ability,which make the neural network method glow new vitality in the field of economic forecasting.First,this paper describes the background and current situation of economic forecasting.Secondly,the paper summarizes characteristics,classification,methods,and basic steps in economic forecasting.Thirdly,the basic theory of the BP Neural Network and Deep Belief Network(DBN)are given.Finally,the two macroeconomic indexes,the national consumer price index and the national total imports,are regarded as the object of empirical study to test the economic forecasting ability of Deep Belief Network,BP Neural Network and Vector Autoregression.This paper compares the performance of DBN with the performance of BP neural network and Vector Autoregression from the aspect of the prediction accuracy and the speed of convergence of training study.Further,reach about whether the DBN works well,when using a lot of non-label samples to Pre-train the network and using a few of labeled samples for fine tuning.Related empirical results show that,DBN model and BP Neural Network can both deal with the non-linear problems such as economic forecasting modeling.DBN deep learning model can be effectively utilized to non-label samples,but BP Neural Network is only effective in label samples,and DBN has a faster convergence,higher precision and stronger generalization ability.All in all,DBN model is a more favorable economic forecast modeling tools,with good usability and applicability.