基于神经网络集成的经济预测模型

Economic forecasting model based on neural network ensemble

针对单个BP神经网络用于经济预测存在的不足,提出了一种新的更有效的经济预测模型——神经网络集成。神经网络集成通过训练多个神经网络并将各网络输出进行合成,能够显著提高网络的泛化能力。以广东省江门市的经济数据为例,采用Bagging算法训练了五个BP神经网络,构建了一个神经网络集成的GDP预测模型,并运用MATLAB7.0语言程序实现。预测结果令人满意,优于单个神经网络预测方法。实证表明,神经网络集成用于经济预测是有效可行的,同时在一定程度上克服了单个神经网络的缺陷。

In view of the weaknesses of simplex BP neural network for economic forecasting, a new and more effective economic forecasting model called neural network ensemble （NNE） is developed in this paper. NNE can improve the generalization ability through training multiple neural networks and combining their results. According to the economic data of Jiangmen, Guangdong, five neural networks have been trained by adopting Bagging to build a NNE, which is realized by MATLAB 7.0 and employed to forecast GDP. Theforecasting results are satisfactory, proving that NNE is superior to simplex neural network. Meanwhile, NNE turns out-to be valid and feasible for economic forecasting and can overcome the shortcomings of simplex BP neural network to some degree.