摘要
针对BP神经网络应用中存在的训练时间长、收敛速度慢的问题,对常用的BP神经网络方法进行了改进,增加了数据前处理和后处理的过程。前处理过程是对BP神经网络的输入变量采用主成分分析法进行预处理,确定主要的影响因素,解决了神经网络训练时输入变量过多造成的效率下降问题;后处理过程是对训练结果采用回归和相关性分析的方法进行评价,验证了训练结果的精度。对农业商品总产值的预测结果表明,改进的BP神经网络方法能够提高神经网络的训练效率,并且达到了较高的预测精度。该方法适用于解决多因素预测的问题。
An improved method was proposed in order to accelerate the convergence speed and shorten the training time of BP neural network. The principal component analysis was used as the preprocessing to select principal components from the input variables, which numbers are too large to deal with. The regression and correlation analysis were used as the post?鄄processing to analyze the result and test the precision of training. The predicting result of agricultural commodity total production value showed that the training efficiency could be improved and the structure of network could be simplified by the method. Moreover, the precision of predicting was very high. The method can be applied to resolve the question of many factors prediction.
出处
《科技通报》
2005年第1期6-9,18,共5页
Bulletin of Science and Technology
基金
国家自然科学基金资助项目(项目批准号:30270773)
高等学校优秀青年教师教学科研奖励计划资助项目
浙江省自然科学基金资助项目(项目批准号:301270)
浙江省自然科学基金人才基金资助项目
关键词
系统工程
BP神经网络
主成分分析法
回归分析
农业商品总产值
system engineering
BP neural network
principal component analysis
regression analysis
Agricultural commodity total production value
