基于GA-BP神经网络的作物参考腾发量预测

Predicting Reference Evapor Transpiration Based on GA-BP Neural Network

采用黄金分割原理优化算法确定BP神经网络的隐含层节点数,进而确定BP神经网络的结构,并针对BP神经网络容易陷入局部极小值和全局搜索能力弱的缺点,引入遗传算法(GA)优化网络权值,建立GA-BP网络模型,预测作物参考腾发量ET0。以北京地区的相关资料为基础,选用6种输入因子组合方案,对该模型进行验证,结果表明该网络模型具有较好的预测能力;同时,对6种方案比较分析表明,方案4最优,该方案只需选用4项输入因子(日序数、平均气温、风速和日照时数),就能以较高的精度预测作物参考腾发量。

In this paper,determined the BP neural hidden network nodes on the basis of principles of golden section optimization algorithm,ulteriorly,determined the structure of BP neural network;to overcome BP neural network easily get into the local minimum and weak ability of global search by introducting genetic algorithm optimization network weights,established GA-BP network model to predict reference evapotranspiration.Choose six kinds of integrated schemes for input factors to validate application reliability of the network model based on relevant information for the Beijing area,the results show that the network model has better predictive capability;comparative analysis on six kinds of integrated schemes shows scheme4 is the optimal,it can accurately,predict reference evapotranspiration by inputing four factors（ordinal day,average temperature,wind speed and sunshine hours）.