小波神经网络模型的改进及其应用  

Improvement of wavelet neural networks model and application

作　　者：杨娜[1,2] 付强[1] 王淑丽[2] 李荣东[2] 朱萍萍[2] 张少坤[1] 杨先野[1] 

机构地区：[1]东北农业大学水利与建筑学院,哈尔滨150030 [2]佳木斯市水利勘测设计研究院,佳木斯154003

出　　处：《系统工程理论与实践》2009年第1期168-173,共6页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(30400275);东北农业大学创新团队发展计划资助(190211)

摘　　要：将优化函数的连续型蚁群算法与小波神经网络耦合,用蚁群算法优化神经网络的权值和小波参数,找到蚁群算法中信息素更新的最佳衡量标准,且建立了基于蚁群优化的小波神经网络模型,旨在准确预测水稻需水量,为制定合理的灌溉制度、提高水利用率提供科学依据.通过对三江平原富锦市1985至2001年的井灌水稻区全生育期需水量预测检验,确定网络结构为6-12-1,训练最大次数20次时网络收敛,误差精度达到0.0024.研究结果表明,该模型不但计算简便,而且具有较强的逼近能力、较快的收敛速度和较好的预报精度,并且为网络模型的参数优化提供一种新方法,也为预测、预报的研究拓宽新思路.Coupling Continuous Ant Colony Algorithm based function optimization and Wavelet Neural Networks, and optimizing weights in Neural Networks and wavelet parameters by Continuous Ant Colony Algorithm, best standard renewing pheromone in Continuous Ant Colony Algorithm was found, and that Wavelet Neural Networks model based on Continuous Ant Colony Algorithm was established, which it was aimed at to forecast accurately Rice Evapotranspiration, make reasonable irrigation scheduling and improve water efficiency. Testing the effects of forecasting water requirement, for well-irrigation rice area of Fujin city in Sanjiang Plain in 1985-2001, network structure was 6-12-1, best training times were 20 until the network converged, and error was only 0.0024. The results indicated that the model is simple, approximation capability is higher, convergence speed is fast enough, and forecast precision is also better. Besides a new method had been proposed for optimizing network parameters and a new thought for it formed.

关 键 词：小波神经网络 蚁群算法 需水量 参数优化 

分 类 号：S152.7[农业科学—土壤学] TP183[农业科学—农业基础科学]