基于VMD和IGWO组合模型的可降水量预测研究  

Research on Precipitable Water Vapor Prediction Based on VMD and IGWO Combined Model

作　　者：李冰艳 张文煜 杨成林 寇梦刚 张昕宇 李明俊 LI Bingyan;ZHANG Wenyu;YANG Chenglin;KOU Menggang;ZHANG Xinyu;LI Mingjun(School of Earth Science and Technology,School of Computer and Artificial Intelligence,Zhengzhou University,Zhengzhou 450001,China;School of innovative engineering,Macao University of science and technology,Macao 999078,China)

机构地区：[1]郑州大学地球科学与技术学院,计算机与人工智能学院,河南郑州450001 [2]澳门科技大学创新工程学院,中国澳门999078

出　　处：《水文》2025年第1期68-75,共8页Journal of China Hydrology

基　　金：中部区域积层混合云人工增雨(雪)研究试验(商丘)(ZQC-H22256);第二次青藏高原综合科学考察研究项目(2019QZKK0104)。

摘　　要：水资源短缺是世界面临的重大问题,人工增雨能提高降水转化率从而增加降水量,缓解水资源短缺问题,但是如何选择作业时机是提高降水转化率的难点。随着探测技术的发展,能够观测获取的表征水汽的主要参量是可降水量(PWV)。由于环境和观测的影响,PWV序列通常具有非线性和非平稳性的特征,这些特性为PWV的精准预测带来挑战,本研究构建一种集数据分解和多模型预测于一体的组合模型,并针对郑州站微波辐射计观测的PWV数据进行多步预测。该模型采用变分模态分解(VMD)技术对PWV序列进行分解和去噪,采用反向传播神经网络(BPNN)、长短期记忆网络(LSTM)、双向门控循环单元(BiGRU)和时间卷积网络(TCN)模型分别预测分解的数据,最后采用改进的灰狼优化算法(IGWO)确定模型的最佳权重,通过加权组合得到最终预测值。结果表明,即使在5步预测中,与VMD-BiGRU、VMD-BP、VMD-LSTM和VMD-TCN相比,新构建的组合模型均方根误差最少也分别降低了50.0%、67.6%、57.9%和17.2%,验证了模型具有较好的稳定性和良好的泛化能力,能为降水预测、判断人工降水时机提供支撑。Water shortage is a major problem faced by the world.Artificial rain enhancement can improve the precipitation conver-sion rate to increase precipitation and alleviate the problem of water shortage,but how to choose the operation time is a difficult of artificial rain enhancement.With the development of detection technology,the main parameter that can be observed and ob-tained to characterize water vapor is the precipitable water vapor(PWV).Due to the influence of environment and observation,PWV series usually has the characteristics of nonlinearity and nonstationarity.These characteristics bring challenges to the accu-rate prediction of PWV.This study constructs a combined model integrating data decomposition and multi model prediction,and carries out multi-step prediction experiment for the PWV data observed by microwave radiometer in Zhengzhou station.The model uses the variational mode decomposition(VMD)technology to decompose and denoise the PWV sequence,and uses the back propagation neural network(BPNN),long short term memory network(LSTM),bidirectional gated recurrent unit(BiGRU)and tempo-ral convolutional network(TCN)models to predict the decomposed data respectively.Finally,the improved grey wolf optimization algorithm(IGWO)is used to determine the optimal weight of the model,and the final prediction value is obtained through the weighted combination.The results show that even in the 5-step prediction,compared with VMD-BiGRU,VMD-BP,VMD-LSTM and VMD-TCN,the root-mean-square error of the newly constructed combined model is at least reduced by 50.0%,67.6%,57.9%and 17.2%,respectively,which proves that the model has good stability and good generalization ability.It can provide sup-port for precipitation forecasting and judging the time of artificial precipitation.

关 键 词：可降水量预测 变分模态分解 智能优化 组合模型 

分 类 号：P339[天文地球—水文科学] P457.6[水利工程—水文学及水资源]