基于支持向量机(SVM)的祁连山典型小流域日降水-径流模拟研究

及时准确的日径流预测在流域水资源的合理规划、利用及管理中具有十分重要的作用。本文以支持向量机(SVM)模型为基础,以祁连山典型小流域-排露沟流域为研究区域,建立了流域日降水-径流模型,对流域未来1~7 d的日径流量进行了模拟预测。为检验SVM模型的有效性,模拟结果与人工神经网络(ANN)模型预测结果进行了对比。结果表明:SVM和ANN均表现出了很高的精度;但相比于传统的ANN模型,SVM模型的预测精度显著提高。表明SVM模型在半干旱山区小流域径流预测中有更好的适用性,可以用于流域中长期日径流预测,是资料有限的条件下中长期日径流预测的有效工具。     

基于QR-RBL的短期径流预测研究

【目的】针对当前径流预测中存在的预测精确度低、稳定性差、延时高，机理认识不深刻等问题，提出一种新的基于正交三角分解的宽度学习模型(QR-RBL)。【方法】该模型利用正交三角矩阵分解重新定义宽度学习输出层权重矩阵求解方案，可有效提高宽度学习模型的预测效率。同时，为提高宽度学习的泛化能力，将正则化方法引入QR-BL,进一步完成QR-RBL预测模型。最后，在QR-RBL的基础上，构建基于QR-RBL的径流预测方法。该方法首先基于径流自相关思想，动态化、智能化选择预测序列，然后通过设置滑动窗口，获取预测步长，最后通过QR-RBL进行未来预测。【结果】以黄河流域部分水文站试验数据为基础，结果表明，基于QR-RBL的径流预测模型相比于传统宽度学习模型，其效率提高0.68倍。相比于传统神经网络模型(ANN)预测精准度提高0.79倍，可信度提高1.1倍。【结论】综合以上分析，QR-RBL算法在径流预测方面有效的提高了模型的精准度，可信度和效率，为径流预报，灾害监管提供了一种新方法和新思路。     

基于经验模态分解和支持向量机的日径流预测研究

准确的径流预测是水资源开发利用的重要依据，但预测难度大。为提高日径流预测精度，以榕江流域南河东桥园站日径流为例，建立了经验模态分解(EMD)和支持向量机(SVM)耦合的日径流预测模型。首先，利用经验模态分解将日径流系列分解为若干子过程，再采用支持向量机深度学习模型分别对每一个子过程进行预测，最后将每个预测结果相加得到原日径流数据的预测结果。研究表明：EMD-SVM组合模型相对于SVM、BP、LSTM单模型具有更好的预测性能。     

SSA分解预校正模型在辽宁西部水库年径流预测中的应用研究

本文采用SSA分解预测模型结合灰色预测模型对辽宁西部某水库年径流进行预测。研究结果表明:SSA分解预测模型可降低年径流低频振荡的显著性,可对灰色预测模型预测的年径流系列进行预先校正,提供模型预测精度。相比于校正前的灰色模型,校正后的模拟预测年径流系列和水库实测年径流系列相对误差减少15.2%,年径流相关系数提高0.21,研究成果对于水库年径流预测方法提供参考价值。     

基于集对分析的径流组合预测模型

采用人工智能BP模型、小波BP模型及GA-BP模型对径流进行预测,然后将径流的实测值系列A和上述3种模型的预测值Bi建立集对H(A,Bi),利用集对分析的同、异、反特性进行联系度计算,据此确定径流预测模型的相对隶属度,并对隶属度进行归一化处理,得到上述3种模型的权重,再依据此权重建立相应的径流组合预测模型。应用1950~1975年小浪底水库的资料,对径流组合预测模型进行模拟,结果显示其预测精度明显高于单个模型的预测精度。     

Seq2Seq 模型的短期水位预测

为有效预测未来一定时间内的连续水位,提出了基于序列到序列(Seq2Seq) 的短期水位预测模型,并使用一个长短期记忆神经网络(LSTM)作为编码层,将历史水位序列编码为一个上下文向量,使用另一个LSTM 作为解码层,将上下文向量解码来预测目标水位序列。以流溪河为研究对象,针对不同预测长度分别建立水位预测模型,并与LSTM 模型和人工神经网络(ANN)模型进行了对比。结果表明:Seq2Seq 模型对连续6 h、12 h 和24 h 水位预测的纳什效率系数最高分别为0.93、0.90和0.85;当预测长度为6 h 时,LSTM 和Seq2Seq 模型预测结果相似,ANN 模型精度较低;当预测长度为12 h 和24 h 时,Seq2Seq 模型相比LSTM 模型和ANN 模型预测效果更好,收敛速度更快。     

基于自相关分析的BP网络年径流预测模型及其应用

在目前利用BP神经网络进行径流预测的方法中,网络输入与输出的确定方法比较模糊,缺乏理论依据.鉴于此,提出通过对数据系列进行自相关性分析来确定网络结构的方法,此方法建立的BP网络预测模型应用于某电站的年径流预测.结果表明,该模型预测精度高,能够满足电站对径流预测的需要.     

混沌理论与BPNN耦合的径流中长期预测模型

受诸多因素的影响,径流时间序列具有非线性和混沌特征。单一的BPNN模型可以进行径流的中长期预测,但存在对径流影响因素量化不够的缺点;单一的混沌模型可以量化径流的影响因素,但只能实现短期预测。为此建立了混沌理论与BPNN耦合的径流中长期预测模型。针对黑河上游莺落峡水文站1944-2017年的径流序列,利用混沌理论计算了径流序列的延迟时间τ、嵌入维数m和最大Lyapunov指数λ_(max),并进行了径流序列的相空间重构,以此确定BPNN的输入层神经元个数、取值和预测的周期时长;利用BPNN对1944年1月-2012年12月的径流量数据进行训练,建立了混沌-BPNN和混沌-BPNN等维递补两种预测模型;以2013年1月-2017年12月(5 a)的径流量进行模型验证。结果表明:混沌-BPNN等维递补模型的预测精度达到了91.84%,预测效果较好。混沌理论与BPNN耦合的径流预测模型将两种方法的优势互补,尤其是混沌-BPNN等维递补模型,在补充新信息的同时剔除因系统发展而使特征意义降低的老数据,减小了BPNN训练的时间跨度,提高了预测精度,为径流的中长期预测提供了新的有效方法。     

基于EMD-FBI-ELM模型的径流预测研究

为提高径流预测精度,提出基于经验模态分解(EMD)和法务侦查(FBI)算法、极限学习机(ELM)相融合的径流预测方法。首先采用EMD将径流序列数据分解成多个更具规律的分量序列,基于自相关函数法(AFM)、虚假最邻近法(FNN)对每个分量序列进行相空间重构;其次利用FBI算法优化ELM输入层权值和隐含层偏值,建立EMD-FBI-ELM径流预测模型,并构建EMD-FBI-SVM、FBI-ELM、FBI-SVM作对比预测模型;最后通过云南省姑老河水文站年径流预测实例对EMD-FBI-ELM、EMD-FBI-SVM、FBI-ELM、FBI-SVM模型进行验证分析。结果表明：EMD-FBI-ELM模型对实例年径流预测的平均相对误差为3.97%,平均相对误差较EMD-FBI-SVM、FBI-ELM、FBI-SVM模型的预测结果分别降低了53.9%、81.7%、86.5%,具有较好的预测效果。EMD-FBI-ELM模型用于径流预测是可行的,模型及优化方法可为相关预测研究提供参考。     

基于遗传规划的径流预测新方法

应用遗传规划方法进行中长期径流预测,将预测模型视为遗传规划中的个体加以处理,依据生物界“优胜劣汰”的原则,运用复制、交叉和变异等遗传操作算子;根据历史样本数据自动生成最佳的径流预测模型,包括模型的函数形式以及模型参数;最后运用得到的预测模型对某水文站的年径流进行预测。仿真结果表明,基于遗传规划的径流预测模型可以明显提高径流预测精度,为解决中长期径流预测问题提供了一种行之有效的新方法。     

Artificial neural network model for synthetic streamflow generation

Time series of streamflow plays an important role in planning, design and management of water resources system. In the event of non availability of a long series of historical streamflow record, generation of the data series is of utmost importance. Although a number of models exist, they may not always produce satisfactory result in respect of statistics of the historical data. In such event, artificial neural network (ANN) model can be a potential alternative to the conventional models. Streamflow series, which is a stochastic phenomenon, can be suitably modeled by ANN for its strong capability to perform non-linear mapping. An ANN model developed for generating synthetic streamflow series of the Pagladia River, a major north bank tributary of the river Brahmaputra, is presented in this paper along with its comparison with other existing models. The comparison carried out in respect of five different statistics of the historical data and synthetically generated data has shown that among the different models, viz., autoregressive moving average (ARMA) model, Thomas-Fiering model and ANN model, the ANN based model has performed better in generating synthetic streamflow series for the Pagladia River.     

Forecasting Daily Streamflow for Basins with Different Physical Characteristics through Data-Driven Methods

Modelling streamflow is essential for activities, such as flood control, drought mitigation, and water resources utilization and management. Artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), and support vector machines (SVM) are techniques that are frequently used in hydrology to specifically model streamflow. This study compares the accuracy of ANN, ANFIS, and SVM in forecasting the daily streamflow with the traditional approach known as autoregressive (AR) model for basins with different physical characteristics. The accuracies of the models are compared for three basins, that is, 1801, 1805, and 1822, at the Seyhan River Basin in Turkey. The comparison was performed by using coefficient of efficiency, index of agreement, and root-mean-square error. Results indicate that ANN and ANFIS are more accurate than AR and SVM for all the basins. ANN and ANFIS perform similarly, while ANN outperformed ANFIS. Among the models used, the ANN demonstrates the highest performance in forecasting the peak flood values. This study also finds that physical characteristics, such as small area, high slope, and high elevation variation, and streamflow variance deteriorate the accuracy of the methods.     

基于改进BP算法的地下水动态预测模型

运用学习率自适应动量BP算法建立了吉林西部地下水埋深人工神经网络模拟预测模型。首先利用自回归分析方法确定网络输入输出样本,而后应用“试错法”确定隐含层节点数,最终建立了6∶10∶1的ANN地下水动态模拟预报模型,最后应用VB语言依据改进BP算法编制计算程序进行模拟计算。通过对模型检验可知该模型模拟和预测精度均较高,完全可应用于地下水位动态预报。2002年以后的预报结果表明该地区地下水位持续下降,应及时加以控制。     

Monthly Streamflow Forecasting Using Convolutional Neural Network

Monthly streamflow forecasting is vital for managing water resources. Recently, numerous studies have explored and evidenced the potential of artificial intelligence (AI) models in hydrological forecasting. In this study, the feasibility of the convolutional neural network (CNN), a deep learning method, is explored for monthly streamflow forecasting. CNN can automatically extract critical features from numerous inputs with its convolution–pooling mechanism, which is a distinct advantage compared with other AI models. Hydrological and large-scale atmospheric circulation variables, including rainfall, streamflow, and atmospheric circulation factors are used to establish models and forecast streamflow for Huanren Reservoir and Xiangjiaba Hydropower Station, China. The artificial neural network (ANN) and extreme learning machine (ELM) with inputs identified based on cross-correlation and mutual information analyses are established for comparative analyses. The performances of these models are assessed with several statistical metrics and graphical evaluation methods. The results show that CNN outperforms ANN and ELM in all statistical measures. Moreover, CNN shows better stability in forecasting accuracy.

     

Influence of Time Discretization and Input Parameter on the ANN Based Synthetic Streamflow Generation

The capability of ANN to generate synthetic series of river discharge averaged over different time steps with limited data has been investigated in the present study. While an ANN model with certain input parameters can generate a monthly averaged streamflow series efficiently; it fails to generate a series of smaller time steps with the same accuracy. The scope of improving efficiency of ANN in generating synthetic streamflow by using different combinations of input data has been analyzed. The developed models have been assessed through their application in the river Subansiri in India. Efficiency of the ANN models has been evaluated by comparing ANN generated series with the historical series and the series generated by Thomas-Fiering model on the basis of three statistical parameters-periodical mean, periodical standard deviation and skewness of the series. The results reveal that the periodical mean of the series generated by both Thomas–Fiering and ANN models are in good agreement with that of the historical series. However, periodical standard deviation and skewness coefficient of the series generated by Thomas–Fiering model is inferior to that of the series generated by ANN.     

Event-based Sediment Yield Modeling using Artificial Neural Network

In the present study, a back propagation feedforward artificial neural network (ANN) model was developed for the computation of event-based temporal variation of sediment yield from the watersheds. The training of the network was performed by using the gradient descent algorithm with automated Bayesian regularization, and different ANN structures were tried with different input patterns. The model was developed from the storm event data (i.e. rainfall intensity, runoff and sediment flow) registered over the two small watersheds and the responses were computed in terms of runoff hydrographs and sedimentographs. Selection of input variables was made by using the autocorrelation and cross-correlation analysis of the data as well as by using the concept of travel time of the watershed. Finally, the best fit ANN model with suitable combination of input variables was selected using the statistical criteria such as root mean square error (RMSE), correlation coefficient (CC) and Nash efficiency (CE), and used for the computation of runoff hydrographs and sedimentographs. Further, the relative performance of the ANN model was also evaluated by comparing the results obtained from the linear transfer function model. The error criteria viz. Nash efficiency (CE), error in peak sediment flow rate (EPS), error in time to peak (ETP) and error in total sediment yield (ESY) for the storm events were estimated for the performance evaluation of the models. Based on these criteria, ANN based model results better agreement than the linear transfer function model for the computation of runoff hydrographs and sedimentographs for both the watersheds.     

基于Matlab神经网络的流域年径流量预测

阐述了运用人工神经网络模型对流域年径流量径流序列做出预报,表明人工神经网络模型在水文预报中具有一定的优势。通过BP神经网络算法得到了适合该神经网络模型的训练算法。以渔峡口站年径流量实测序列为研究对象,在数值试验的基础上建立了年径流序列预报的人工神经网络预报模型结构,提高了该模型的预报准确性。     

Modeling of Sediment Yield Prediction Using M5 Model Tree Algorithm and Wavelet Regression

The forecast of the sediment yield generated within a watershed is an important input in the water resources planning and management. The methods for the estimation of sediment yield based on the properties of flow and sediment have limitations attributed to the simplification of important parameters and boundary conditions. Under such circumstances, soft computing approaches have proven to be an efficient tool in modelling the sediment yield. The focus of present study is to deal with the development of decision tree based M5 Model Tree and wavelet regression models for modeling sediment yield in Nagwa watershed in India. A comparison is also performed with the artificial neural network (ANN) model for streamflow forecasting. The root mean square errors (RMSE), Nash-Sutcliff efficiency index (N-S Index), and correlation coefficient (R) statistics are used for the statistical criteria. A comparative evaluation of the performance of M5 Model Tree and wavelet regression versus ANN clearly shows that M5 Model Tree and wavelet regression can prove more useful than ANN models in estimation of sediment yield. Further, M5 model tree offers explicit expressions for use by design engineers.     

Daily Mean Streamflow Prediction in Perennial and Non-Perennial Rivers Using Four Data Driven Techniques

This study examines and compares the performance of four new attractive artificial intelligence techniques including artificial neural network (ANN), hybrid wavelet-artificial neural network (WANN), Genetic expression programming (GEP), and hybrid wavelet-genetic expression programming (WGEP) for daily mean streamflow prediction of perennial and non-perennial rivers located in semi-arid region of Zagros mountains in Iran. For this purpose, data of daily mean streamflow of the Behesht-Abad (perennial) and Joneghan (non-perennial) rivers as well as precipitation information of 17 meteorological stations for the period 1999–2008 were used. Coefficient of determination (R²) and root mean square error (RMSE) were used for evaluating the applicability of developed models. This study showed that although the GEP model was the most accurate in predicting peak flows, but in overall among the four mentioned models in both perennial and non-perennial rivers, WANN had the best performance. Among input patterns, flow based and coupled precipitation-flow based patterns with negligible difference to each other were determined to be the best patterns. Also this study confirmed that combining wavelet method with ANN and GEP and developing WANN and WGEP methods results in improving the performance of ANN and GEP models.     

基于相空间重构理论的神经网络在大坝渗流预测中的应用

简述了混沌预测方法和原理,提出运用神经网络来模拟混沌系统的动力学模型。通过实例重构出大坝渗流观测数据相空间,将其作为神经网络的输入,对渗流观测数据进行了预测。计算结果表明,基于相空间重构理论的神经网络模型具有一定的准确性和有效性,对于处理非线性问题是一种有益的探索。