GA-BP神经网络的GPS可降水量预测  

Prediction of GPS perceptible water vapor based on GA-BP neural network

作　　者：谢劭峰 赵云 李国弘 周志浩 XIE Shaofeng;ZHAO Yun;LI Guohong;ZHOU Zhihao(College of Geomatics and Geoinformation,Guilin University of Technology,Guilin,Guangxi 541006,China;Guangxi Key Laboratory of Spatial Information and Geomatics,Guilin,Guangxi 541006,China)

机构地区：[1]桂林理工大学测绘地理信息学院,广西桂林541006 [2]广西空间信息与测绘重点实验室,广西桂林541006

出　　处：《测绘科学》2020年第3期33-38,共6页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41864002,41704027);广西自然科学基金项目(2018GXNSFAA281182);广西中青年教师基础能力提升项目(2017KY0267)。

摘　　要：针对传统BP神经网络模型存在的学习速度慢、易陷入局部极值以及网络结构参数取值的不确定性等问题,该文研究了一种基于遗传算法与BP神经网络相结合的GPS可降水量预测的新方法。该方法利用遗传算法对BP神经网络的初始权值和阈值进行优化,并对该模型进行训练,以提高预测模型的性能。实验结果证明了遗传BP神经网络模型用于GPS可降水量预测的可行性,其预测结果的均方根误差为0.16 mm、平均绝对百分误差为0.23%。相对于BP神经网络和小波神经网络模型,均方根误差分别降低了0.37和0.19 mm,平均绝对百分误差分别降低了0.62%和0.33%。同时遗传BP神经网络模型亦显示了很好的非线性拟合能力,能更好地预测GPS可降水量,对实际工作具有较强的参考价值。Aiming at the slow learning speed,easy to fall into local extremum and uncertainty of network structure parameters existing in the traditional BP neural network model,a new method of GPS perceptible water vapor(PWV) prediction based on genetic algorithm(GA) and BP neural network was studied in this paper.The GA was used to optimize the initial weight and threshold value of the BP neural network,and then the prediction model was trained in order to improve its performance.The experimental results testify that it is feasible to apply GA-BP neural network model to the prediction of GPS PWV.The root mean square error(RMSE) of the predicted result is 0.16 mm,and the mean absolute percent error(MAPE) is 0.23%.Compared with the BP neural network and wavelet neural network,the RMSE reduces by 0.37 and 0.19 mm,respectively;the MAPE reduces by 0.62% and 0.33%,respectively.At the same time,GA-BP neural network model also shows a very good nonlinear fitting ability,which can better predict GPS PWV and has a strong reference value for practical work.

关 键 词：BP神经网络 遗传算法 GPS可降水量 预测 

分 类 号：P457.6[天文地球—大气科学及气象学] P228.9