摘要
针对传统BP神经网络算法在对预测问题中存在的网络具有易陷入局部极小、收敛速度慢的缺陷,引入附加动量法和自适应学习速率法改进BP神经网络预测模型.将改进后的预测方法应用于企业的市场需求预测问题,以某汽车制造企业过去12个月汽车销售量的实际数据为样本,分别采用基于时间序列和基于因素分析两种预测模型,对所提出的改进预测方法进行实证分析.结果表明:所提出的算法对销售量的预测精度较高,误差均小于8.8%,运算时间也有所降低,预测结果表明文中所提出的算法在处理网络易陷入局部极小、收敛速度慢的预测问题方面的有效性.
The paper adopts the additional momentum method and adaptive learning rate method to im- prove the traditional BP algorithm with the defect to easily fall into local minima and slow convergence The improved predicting model was applied in the forecasting of market demand. Based on the actual date of auto sales of an automobile manufacturing companies from the past i2 months, two prediction models on based on time series and factor analysis are adopted respectively, and improved prediction methods for empirical analysis. Research results show that the prediction accuracy of the proposed al- gorithm is higher, the error is less than 8.8%, and the computing time is decreased. Forecasting re- sults prove the presented algorithm is effective to solve the prediction problems of network easily into the local minimum and slow convergence speed.
出处
《武汉理工大学学报(交通科学与工程版)》
2012年第6期1252-1255,共4页
Journal of Wuhan University of Technology(Transportation Science & Engineering)
关键词
附加动量法
需求预测
改进的BP神经网络
时间序列
additional momentum method
demand prediction
improved BP neural network
time serial