摘要
准确的含沙量多步预测对于区域水土流失治理、防洪减灾等具有重要意义。为提高含沙量多步预测精度,改进深度信念网络(DBN)的预测性能,基于小波包变换(WPT),分别提出人工兔优化(ARO)算法、鹰栖息优化(EPO)算法与DBN组合的月含沙量多步预测模型,通过云南省龙潭站月含沙量时序数据对模型进行验证。首先利用WPT对实例月含沙量时序数据进行3层分解处理,得到8个更具规律的子序列分量;其次介绍ARO、EPO算法原理,利用ARO、EPO对DBN隐藏层神经元数等超参数进行寻优,建立WPT-ARO-DBN、WPT-EPO-DBN预测模型,并构建WPT-PSO(粒子群算法)-DBN、WPT-DBN作对比分析模型;最后利用4种模型对各子序列分量进行预测,将预测值叠加得到最终月含沙量多步预测结果。结果表明:(1)WPT-ARO-DBN、WPT-EPO-DBN模型对实例超前1步—超前4步月含沙量具有满意的预测效果,对超前5步具有较好的预测结果,对超前6步、超前7步的预测效果一般,对超前8步的预测精度较差,已不能满足预测精度需求;(2)WPT-ARO-DBN、WPT-EPO-DBN模型的多步预测效果要优于WPT-PSO-DBN模型,远优于WPT-DBN模型,具有更高的预测精度、更好的泛化能力和更大的预测步长;(3)ARO、EPO能有效优化DBN超参数,提高DBN预测性能,优化效果优于PSO,WPT-ARO-DBN、WPT-EPO-DBN模型能充分发挥WPT、新型群体智能算法和DBN网络优势,提高月含沙量多步预测精度,且预测精度随着预测步数的增加而降低。
Accurate multi-step sediment concentration prediction is of significance for regional soil erosion control,flood control and disaster reduction.To improve the multi-step prediction accuracy of sediment concentration and the prediction performance of the deep belief network(DBN),this paper proposes a multi-step prediction model of monthly sediment concentration by combining the artificial rabbit optimization(ARO)algorithm,eagle habitat optimization(EPO)algorithm,and DBN based on wavelet packet transform(WPT).The model is validated using time series data of monthly sediment concentration from Longtan Station in Yunnan Province.Firstly,WPT is employed to decompose the time series data of the monthly sediment concentration of the case in three layers,and eight more regular subsequence components are obtained.Secondly,the principles of ARO and EPO algorithms are introduced,and hyperparameters such as the neuron number in the hidden layer of DBN are optimized by ARO and EPO.Meanwhile,WPT-ARO-DBN and WPT-EPO-DBN prediction models are built,and WPT-PSO(particle swarm optimization)-DBN and WPT-DBN are constructed for comparative analysis.Finally,four models are adopted to predict each subsequence component,and the predicted values are superimposed to obtain the multi-step prediction results of the final monthly sediment concentration.The results are as follows.①WPT-ARO-DBN and WPT-EPO-DBN models have satisfactory prediction effects on the monthly sediment concentration of the case from one step ahead to four steps ahead.This yields sound prediction results for five steps ahead.The prediction effect for six steps ahead and seven steps ahead is average,and the prediction accuracy for eight steps ahead is poor and cannot meet the prediction accuracy requirements.②The multi-step prediction performance of WPT-ARO-DBN and WPT-EPO-DBN models is superior to WPT-PSO-DBN models and far superior to WPT-DBN models,with higher prediction accuracy,better generalization ability,and larger prediction step size.③ARO and EPO can effectively optimize DBN hyperparameters,improve DBN prediction performance,and have better optimization effects than PSO.Additionally,WPT-ARO-DBN and WPT-EPO-DBN models can give full play to the advantages of WPT,new swarm intelligence algorithms and the DBN network and improve the multi-step prediction accuracy of monthly sediment concentration,and the prediction accuracy decreases with the increasing prediction steps.
作者
高雪梅
崔东文
GAO Xuemei;CUI Dongwen(Yunnan Wenshan Institute of Water Resources and Electric Power Survey and Design,Wenshan 663000,China;Yunnan Wenshan Water Bureau,Wenshan 663000,China)
出处
《人民珠江》
2024年第3期69-78,共10页
Pearl River
基金
国家重点研发计划项目(2019YFC0507500)
国家自然科学基金项目(41702278)
中国地质调查局地质调查项目(DD20221758、DD20190326)。
关键词
月含沙量预测
深度信念网络
人工兔优化算法
鹰栖息优化算法
小波包变换
组合模型
prediction of monthly sediment concentration
deep belief network
artificial rabbit optimization algorithm
eagle perching optimization algorithm
wavelet packet transform
combining model
