自适应BP神经网络在日最高气温预报中的应用  

The Application of Self-adaptive BP Neural Network on Daily Maximum Temperature Prediction

作　　者：万飞笑 罗泽[1] 阎保平[1] 郝美玉[1] 

机构地区：[1]中国科学院计算机网络信息中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《科研信息化技术与应用》2015年第3期68-78,共11页E-science Technology & Application

基　　金：国家自然科学基金(61361126011);国家发展和改革委员会项目(GC-HG100452);中国科学院计算机网络信息中心青年基金项目(CNIC_QN_1401)

摘　　要：气象数据规模飞速增长,BP神经网络由于其强大的非线性系统拟合能力,在气象数据尤其是气温相关数据的分析和预测中得到广泛应用。考虑到气温数据的海量规模,采用传统BP神经网络的预报面临误差收敛速度慢、容易陷入局部极小等问题,从而严重影响预测的效率和精准度。本文以BP算法为基础提出了一种改进的学习率自适应算法建立BP网络预报模型,并针对气象数据的多维性和周期性提出了优化方案。实验结果表明该方法可以高效的完成最高气温的预测,并且相对传统算法拥有更高的预报精度。With the rapid growth of meteorological data volume, BP neural network that works well in fitting nonlinear systems has been widely used for analysis and prediction of meteorological data, especially in temperaturedata. Because of the large volume of temperature data, when used for predicting, the traditional BP neural network has the problems, such as slow convergence and easily falling into the local minimum, which are seriously affected the efficiency and precision of the prediction. Based on BP algorithm, this paper presented an improved self-adaptive learning rate algorithm to build BP neural network prediction model, and an optimization scheme according to the multidimensionality and periodicity of meteorological data. The experiment results show that the model can efficiently predict the highest temperature, and has higher prediction accuracy than the traditional algorithms.

关 键 词：日最高气温预报 BP神经网络 学习率自适应 逐步回归分析 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]