Real‐time recurrent learning neural network for stream‐flow forecasting

Various types of neural networks have been proposed in previous papers for applications in hydrological events. However, most of these applied neural networks are classified as static neural networks, which are based on batch processes that update action only after the whole training data set has been presented. The time variate characteristics in hydrological processes have not been modelled well. In this paper, we present an alternative approach using an artificial neural network, termed real‐time recurrent learning (RTRL) for stream‐flow forecasting. To define the properties of the RTRL algorithm, we first compare the predictive ability of RTRL with least‐square estimated autoregressive integrated moving average models on several synthetic time‐series. Our results demonstrate that the RTRL network has a learning capacity with high efficiency and is an adequate model for time‐series prediction. We also investigated the RTRL network by using the rainfall–runoff data of the Da‐Chia River in Taiwan. The results show that RTRL can be applied with high accuracy to the study of real‐time stream‐flow forecasting networks. Copyright © 2002 John Wiley & Sons, Ltd.     

Estimation of river flow by artificial neural networks and identification of input vectors susceptible to producing unreliable flow estimates

Reliable river flow estimates are crucial for appropriate water resources planning and management. River flow forecasting can be conducted by conceptual or physical models, or data-driven black box models. Development of physically-based models requires an understanding of all the physical processes which impact a natural process and the interactions among them. Since identification of the relationships among these physical processes is very difficult, data-driven approaches have recently been utilized in hydrological modeling. Artificial neural networks are one of the widely used data-driven approaches for modeling hydrological processes. In this study, estimation of future monthly river flows for Guvenc River, Ankara is conducted using various artificial neural network models. Success of artificial neural network models relies on the availability of adequate data sets. A direct mapping from inputs to outputs without consideration of the complex relationships among the dependent and independent variables of the hydrological process is identified. In this study, past precipitation, river flow data, and the associated month are used to predict future river flows for Guvenc River. Impacts of various input patterns, number of training cycles, and initial values assigned to the weights of the connections are investigated. One of the major weaknesses of artificial neural networks is that they may fail to generate good estimates for extreme events, i.e. events that do not occur at all or often enough in the training data set. It is very important to be able to identify such unlikely events. A fuzzy c-means algorithm is used in this study to cluster the training and validation input vectors into regular and extreme events so that the user will have an idea about the risk of the artificial neural network model to generate unreliable results.     

基于长短时记忆神经网络的鄱阳湖水位预测

湖泊水位是维持其生态系统结构、功能和完整性的基础.鄱阳湖受流域"五河"和长江来水双重影响,水位变化复杂.为了准确预测鄱阳湖水位变化,采用长短时记忆神经网络方法(LSTM)构建了鄱阳湖水位预测模型.该模型以赣江、抚河、信江、饶河和修水"五河"入湖流量和长江干流流量作为输入条件,预测鄱阳湖湖区不同代表站(湖口、星子、都昌、吴城和康山)的水位过程.研究以1956—1980年的水文时间序列数据作为训练集,1981—2000年作为验证集,探讨了LSTM模型输入时间窗、隐藏神经元数目、初始学习率等模型参数对预测精度的影响,并确定了鄱阳湖水位预测模型的最优参数.结果表明,采用LSTM神经网络方法可基于流域"五河"和长江来水量历时数据合理预测鄱阳湖不同湖区的水位过程,五站水位预测的均方根误差为0.41～0.50 m,纳什效率系数和决定系数达0.96～0.98.为考察模型训练数据集对鄱阳湖水位预测结果的影响,进一步选取了随机5年(1956—1960年)的资料和5个典型水文年(1954年、1973年、1974年、1977年和1978年)的日均流量资料来训练模型.结果显示随机5年资料作为训练数据的预测精度要...     

Comparison of artificial neural network models for hydrologic predictions at multiple gauging stations in an agricultural watershed

This paper reports on an evaluation of the use of artificial neural network (ANN) models to forecast daily flows at multiple gauging stations in Eucha Watershed, an agricultural watershed located in north‐west Arkansas and north‐east Oklahoma. Two different neural network models, the multilayer perceptron (MLP) and the radial basis neural network (RBFNN), were developed and their abilities to predict stream flow at four gauging stations were compared. Different scenarios using various combinations of data sets such as rainfall and stream flow at various lags were developed and compared for their ability to make flow predictions at four gauging stations. The input vector selection for both models involved quantification of the statistical properties such as cross‐, auto‐ and partial autocorrelation of the data series that best represented the hydrologic response of the watershed. Measured data with 739 patterns of input–output vector were divided into two sets: 492 patterns for training, and the remaining 247 patterns for testing. The best performance based on the RMSE, R² and CE was achieved by the MLP model with current and antecedent precipitation and antecedent flow as model inputs. The MLP model testing resulted in R² values of 0·86, 0·86, 0·81, and 0·79 at the four gauging stations. Similarly, the testing R² values for the RBFNN model were 0·60, 0·57, 0·58, and 0·56 for the four gauging stations. Both models performed satisfactorily for flow predictions at multiple gauging stations, however, the MLP model outperformed the RBFNN model. The training time was in the range 1–2 min for MLP, and 5–10 s for RBFNN on a Pentium IV processor running at 2·8 GHz with 1 MB of RAM. The difference in model training time occurred because of the clustering methods used in the RBFNN model. The RBFNN uses a fuzzy min‐max network to perform the clustering to construct the neural network which takes considerably less time than the MLP model. Results show that ANN models are useful tools for forecasting the hydrologic response at multiple points of interest in agricultural watersheds. Copyright © 2008 John Wiley & Sons, Ltd.     

Simulating and predicting river discharge time series using a wavelet‐neural network hybrid modelling approach

Accurate simulation and prediction of the dynamic behaviour of a river discharge over any time interval is essential for good watershed management. It is difficult to capture the high‐frequency characteristics of a river discharge using traditional time series linear and nonlinear model approaches. Therefore, this study developed a wavelet‐neural network (WNN) hybrid modelling approach for the predication of river discharge using monthly time series data. A discrete wavelet multiresolution method was employed to decompose the time series data of river discharge into sub‐series with low (approximation) and high (details) frequency, and these sub‐series were then used as input data for the artificial neural network (ANN). WNN models with different wavelet decomposition levels were employed to predict river discharge 48 months ahead of time. Comparison of results from the WNN models with those of the ANN models alone indicated that WNN models performed a more accurate prediction. Copyright © 2011 John Wiley & Sons, Ltd.     

In situ and high frequency monitoring of suspended sediment properties using a spectrophotometric sensor

It is well known that sediment properties, including sediment‐associated chemical constituents and sediment physical properties, can exhibit significant variations within and between storm runoff events. However, the number of samples included in suspended sediment studies is often limited by time‐consuming and expensive laboratory procedures after stream water sampling. This restricts high frequency sampling campaigns to a limited number of events and reduces accuracy when aiming to estimate fluxes and loads of sediment‐associated chemical constituents. In this study, we address the potential of a portable ultraviolet–visible spectrophotometer (220–730 nm) to estimate suspended sediment properties in a resource efficient way. Several field deployable spectrophotometers are currently available for in‐stream measurements of environmental variables at high temporal resolution. These instruments have primarily been developed and used to quantify solute concentrations (e.g. dissolved organic carbon and NO₃‐N), total concentrations of dissolved and particulate forms (e.g. total organic carbon) and turbidity. Here we argue that light absorbance values can be calibrated to estimate sediment properties. We present light absorbance data collected at 15‐min intervals in the Weierbach catchment (NW Luxembourg, 0.45 km²) from December 2013 to January 2015. In this proof‐of‐concept study, we performed a local calibration using suspended sediment loss‐on‐ignition (LOI) measurements as an example of suspended sediment property. We assessed the performance of several regression models that relate light absorbance measurements with the percentage weight LOI. The MM‐robust regression method presented the lowest standard error of prediction (0.48%) and was selected for calibration (adjusted r² = 0.76 between observed and predicted values). The model was then used to predict LOI during a storm runoff event in December 2014. This study demonstrates that spectrophotometers can be used to estimate suspended sediment properties at high temporal resolution and for long‐time spans in a simple, non‐destructive and affordable manner. Copyright © 2016 John Wiley & Sons, Ltd.     

Using recurrent neural networks to reconstruct rainfall‐runoff processes

Many novel techniques for reconstructing rainfall‐runoff processes require hydrometeorologic and geomorphologic information for modelling. However, certain information is not always measurable. In this paper, we employ a special recurrent neural network to reconstruct the rainfall‐runoff process by using collected rainfall data. In addition, we propose an indirect system identification to overcome the drawback of a traditional, time‐consuming trial‐and‐error search. The indirect system identification is an efficient method to recognize the structure of a recurrent neural network. The unit hydrograph can be derived directly from the weights of the network due to a state‐space form embedded in the recurrent neural network. This improves the link between the weights of the network and the physical concepts that most neural networks fail to connect. The case studies of 41 events from 1966 to 1997 have been implemented in Taiwan's Wu‐Tu watershed, where the runoff path‐lines are short and steep. Two recurrent neural networks and one state‐space model are utilized to simulate the rainfall‐runoff processes for comparison. The results are validated by four criteria: coefficient of efficiency; peak discharge error; time to peak arrival error; total discharge volume error. The resulting data from the recurrent neural network reveal that the neural network proposed herein is appropriate for hydrological systems. Copyright © 2005 John Wiley & Sons, Ltd.     

Groundwater-level forecasting under climate change scenarios using an artificial neural network trained with particle swarm optimization

Abstract

Artificial neural networks (ANNs) have recently been used to predict the hydraulic head in well locations. In the present work, the particle swarm optimization (PSO) algorithm was used to train a feed-forward multi-layer ANN for the simulation of hydraulic head change at an observation well in the region of Agia, Chania, Greece. Three variants of the PSO algorithm were considered, the classic one with inertia weight improvement, PSO with time varying acceleration coefficients (PSO-TVAC) and global best PSO (GLBest-PSO). The best performance was achieved by GLBest-PSO when implemented using field data from the region of interest, providing improved training results compared to the back-propagation training algorithm. The trained ANN was subsequently used for mid-term prediction of the hydraulic head, as well as for the study of three climate change scenarios. Data time series were created using a stochastic weather generator, and the scenarios were examined for the period 2010–2020.

Editor Z.W. Kundzewicz; Associate editor L. See

Citation Tapoglou, E., Trichakis, I.C., Dokou, Z., Nikolos, I.K., and Karatzas, G.P., 2014. Groundwater-level forecasting under climate change scenarios using an artificial neural network trained with particle swarm optimization. Hydrological Sciences Journal, 59(6), 1225–1239. http://dx.doi.org/10.1080/02626667.2013.838005     

Neuro‐control of seismically excited steel structure through sensitivity evaluation scheme

The neuro‐controller training algorithm based on cost function is applied to a multi‐degree‐of‐freedom system; and a sensitivity evaluation algorithm replacing the emulator neural network is proposed. In conventional methods, the emulator neural network is used to evaluate the sensitivity of structural response to the control signal. To use the emulator, it should be trained to predict the dynamic response of the structure. Much of the time is usually spent on training of the emulator. In the proposed algorithm, however, it takes only one sampling time to obtain the sensitivity. Therefore, training time for the emulator is eliminated. As a result, only one neural network is used for the neuro‐control system. In the numerical example, the three‐storey building structure with linear and non‐linear stiffness is controlled by the trained neural network. The actuator dynamics and control time delay are considered in the simulation. Numerical examples show that the proposed control algorithm is valid in structural control. Copyright © 2001 John Wiley & Sons, Ltd.     

A two‐step‐ahead recurrent neural network for stream‐flow forecasting

In many engineering problems, such as flood warning systems, accurate multistep‐ahead prediction is critically important. The main purpose of this study was to derive an algorithm for two‐step‐ahead forecasting based on a real‐time recurrent learning (RTRL) neural network that has been demonstrated as best suited for real‐time application in various problems. To evaluate the properties of the developed two‐step‐ahead RTRL algorithm, we first compared its predictive ability with least‐square estimated autoregressive moving average with exogenous inputs (ARMAX) models on several synthetic time‐series. Our results demonstrate that the developed two‐step‐ahead RTRL network has efficient ability to learn and has comparable accuracy for time‐series prediction as the refitted ARMAX models. We then investigated the two‐step‐ahead RTRL network by using the rainfall–runoff data of the Da‐Chia River in Taiwan. The results show that the developed algorithm can be successfully applied with high accuracy for two‐step‐ahead real‐time stream‐flow forecasting. Copyright © 2003 John Wiley & Sons, Ltd.     

Short-term prediction of influent flow in wastewater treatment plant

Predicting influent flow is important in the management of a wastewater treatment plant (WWTP). Because influent flow includes municipal sewage and rainfall runoff, it exhibits nonlinear spatial and temporal behavior and therefore makes it difficult to model. In this paper, a neural network approach is used to predict influent flow in the WWTP. The model inputs include historical influent data collected at a local WWTP, rainfall data and radar reflectivity data collected by the local weather station. A static multi-layer perceptron neural network performs well for the current time prediction but a time lag occurs and increases with the time horizon. A dynamic neural network with an online corrector is proposed to solve the time lag problem and increase the prediction accuracy for longer time horizons. The computational results show that the proposed neural network accurately predicts the influent flow for time horizons up to 300 min.     

Prediction and modelling of rainfall–runoff during typhoon events using a physically-based and artificial neural network hybrid model

Abstract

The accurate prediction of hourly runoff discharge in a watershed during heavy rainfall events is of critical importance for flood control and management. This study predicts n-h-ahead runoff discharge in the Sandimen basin in southern Taiwan using a novel hybrid approach which combines a physically-based model (HEC-HMS) with an artificial neural network (ANN) model. Hourly runoff discharge data (1200 datasets) from seven heavy rainfall events were collected for the model calibration (training) and validation. Six statistical indicators (i.e. mean absolute error, root mean square error, coefficient of correlation, error of time to peak discharge, error of peak discharge and coefficient of efficiency) were employed to evaluate the performance. In comparison with the HEC-HMS model, the single ANN model, and the time series forecasting (ARMAX) model, the developed hybrid HEC-HMS–ANN model demonstrates improved accuracy in recursive n-h-ahead runoff discharge prediction, especially for peak flow discharge and time.     

River flow forecasting using wavelet and cross‐wavelet transform models

In this study, short‐term river flood forecasting models based on wavelet and cross‐wavelet constituent components were developed and evaluated for forecasting daily stream flows with lead times equal to 1, 3, and 7 days. These wavelet and cross‐wavelet models were compared with artificial neural network models and simple perseverance models. This was done using data from the Skrwa Prawa River watershed in Poland. Numerical analysis was performed on daily maximum stream flow data from the Parzen station and on meteorological data from the Plock weather station in Poland. Data from 1951 to 1979 was used to train the models while data from 1980 to 1983 was used to test the models. The study showed that forecasting models based on wavelet and cross‐wavelet constituent components can be used with great accuracy as a stand‐alone forecasting method for 1 and 3 days lead time river flood forecasting, assuming that there are no significant trends in the amplitude for the same Julian day year‐to‐year, and that there is a relatively stable phase shift between the flow and meteorological time series. It was also shown that forecasting models based on wavelet and cross‐wavelet constituent components for forecasting river floods are not accurate for longer lead time forecasting such as 7 days, with the artificial neural network models providing more accurate results. Copyright © 2008 John Wiley & Sons, Ltd.     

The influences of the Southern and North Atlantic Oscillations on climatic surface variables in Turkey

In this study, Turkish climatic variables (precipitation, stream flow and maximum and minimum temperatures) were first analysed in association with both the Southern Oscillation (SO) and the North Atlantic Oscillation (NAO). The relationships between Turkish maximum and minimum monthly temperatures and the extreme phases of the SO (El Niño and La Niña events) were examined. The results of this analysis showed that relationships between Turkish monthly maximum temperatures and El Niño and La Niña contain some complexity still to be identified, because both events produce a signal indicating a correspondence with cold anomalies in the aggregate composites. A relationship between turkish minimum temperatures and El Niño was detected in western Anatolia, whereas there was no significant and consistent signal associated with La Niña. Moreover a series of cross‐correlation analyses was carried out to demonstrate the teleconnections between the climatic variables and both the NAO and SO. The NAO during winter was found to influence precipitation and stream‐flow patterns. In contrast temperature patterns appeared to be less sensitive to the NAO. Furthermore, lag‐correlation results indicated a prediction potential for both precipitation and stream‐flow variables in connection with the NAO. Simultaneous and time‐lag correlations between the climatic variables and the SO index, in general, indicated weaker relationships in comparison with those for the NAO. These analyses also showed that the influences of the SO on Turkish temperature data are negligible. The outcomes were presented in conjunction with an explanation regarding physical mechanisms behind the implied teleconnections. Copyright © 2004 John Wiley & Sons, Ltd.     

Biofiltration of Hexane Vapor: Experimental and Neural Model Analysis

Biofiltration is a commonly practiced biological technique to remove volatile compounds from waste gas streams. From an industrial view‐point, biofilter (BF) operation should be flexible to handle temperatures and inlet load (IL) variations. A compost BF was operated at different temperatures (30–45°C) and at various inlet loading rates (ILR; 8–598 g m^?3 h^?1) under intermittent loading conditions. Complete removal of n‐hexane was observed at 30 and 35°C at ILRs up to 330 g m^?3 h^?1. Besides, 20–75% of the pollutant was removed at 40°C, corresponding to the different ILs applied to the BF. Increasing the temperature to 45°C decreased the removal efficiency (RE) significantly. A feed forward neural network was used to predict the RE of BF with temperature and ILR as the input variables. The experimental data was divided into training (2/3) and test datasets (1/3). The best structure of neural network was obtained by trial and error on the basis of the least differences between predicted and experimental values, as ascertained from their coefficient of regression (R²) values. The modeling results showed that a multilayer network with the topology 2?10?1 was able to predict BF performance effectively with R²‐value of 0.995 for the test data. The results from this study showed the predicting capability of ANNs which can be considered as an alternative for conventional knowledge‐based models.     

基于数据分布域变换与贝叶斯神经网络的渗透率预测及不确定性估计

渗透率是储层评价和油气藏开发的关键参数.传统测井方法与常规机器学习方法估算的渗透率都是固定值.但由于测井数据本身存在噪声, 渗透率的预测结果可能受到噪声的影响出现测量性的随机误差(即任意不确定性); 同时, 当测试数据与训练数据存在差异时, 机器学习模型在预测渗透率时可能出现模型参数的不确定性(即认知不确定性).为实现渗透率的准确预测并量化两种不确定性对结果的影响, 本文提出基于数据分布域变换和贝叶斯神经网络同时实现渗透率预测及其不确定性的估计.提出方法主要包括两个部分: 一部分是不同域数据分布的相互转换, 另一部分是基于贝叶斯理论的神经网络渗透率建模预测和不确定性估计.由于贝叶斯神经网络存在数据分布的假设, 当标签的概率分布与网络的分布保持一致时, 贝叶斯神经网络可以更好的学习到数据之间的关系.因此通过寻找一个函数将一个原始域的渗透率标签转换为目标域的与渗透率有关的变量(我们称为目标域渗透率), 使得该变量符合贝叶斯神经网络的分布假设.我们使用贝叶斯神经网络预测目标域渗透率以及任意不确定性和认知不确定性.随后, 通过分布域的逆变换, 我们将目标域渗透率还原回原始域渗透率.应用本文方法到某油田的18口井的测井数据中, 使用16口井的数据进行训练, 2口井进行测试.测试井的预测渗透率与真实渗透率基本一致.同时, 任意不确定性的预测结果提供了渗透率预测值受到的测井数据噪声影响的位置.认知不确定的预测结果说明数据量少的位置具有更高的认知不确定性.我们提出的这一流程不仅显示了在储层表征方面的巨大潜力, 同时可以降低测井解释时的风险.

     

Basin‐scale stream‐flow forecasting using the information of large‐scale atmospheric circulation phenomena

It is well recognized that the time series of hydrologic variables, such as rainfall and streamflow are significantly influenced by various large‐scale atmospheric circulation patterns. The influence of El Niño‐southern oscillation (ENSO) on hydrologic variables, through hydroclimatic teleconnection, is recognized throughout the world. Indian summer monsoon rainfall (ISMR) has been proved to be significantly influenced by ENSO. Recently, it was established that the relationship between ISMR and ENSO is modulated by the influence of atmospheric circulation patterns over the Indian Ocean region. The influences of Indian Ocean dipole (IOD) mode and equatorial Indian Ocean oscillation (EQUINOO) on ISMR have been established in recent research. Thus, for the Indian subcontinent, hydrologic time series are significantly influenced by ENSO along with EQUINOO. Though the influence of these large‐scale atmospheric circulations on large‐scale rainfall patterns was investigated, their influence on basin‐scale stream‐flow is yet to be investigated. In this paper, information of ENSO from the tropical Pacific Ocean and EQUINOO from the tropical Indian Ocean is used in terms of their corresponding indices for stream‐flow forecasting of the Mahanadi River in the state of Orissa, India. To model the complex non‐linear relationship between basin‐scale stream‐flow and such large‐scale atmospheric circulation information, artificial neural network (ANN) methodology has been opted for the present study. Efficient optimization of ANN architecture is obtained by using an evolutionary optimizer based on a genetic algorithm. This study proves that use of such large‐scale atmospheric circulation information potentially improves the performance of monthly basin‐scale stream‐flow prediction which, in turn, helps in better management of water resources. Copyright © 2007 John Wiley & Sons, Ltd.     

Diagnostic study and modeling of the annual positive water temperature onset

A data-driven model is designed using artificial neural networks (ANN) to predict the average onset for the annual water temperature cycle of North-American streams. The data base is composed of daily water temperature time series recorded at 48 hydrometric stations in Québec (Canada) and northern US, as well as the geographic and physiographic variables extracted from the 48 associated drainage basins. The impact of individual and combined drainage area characteristics on the stream annual temperature cycle starting date is investigated by testing different combinations of input variables. The best model allows to predict the average temperature onset for a site, given its geographical coordinates and vegetation and lake coverage characteristics, with a root mean square error (RMSE) of 5.6 days. The best ANN model was compared favourably with parametric approaches.     

River flow time series prediction with a range-dependent neural network

Abstract

Artificial neural networks provide a promising alternative to hydrological time series modelling. However, there are still many fundamental problems requiring further analyses, such as structure identification, parameter estimation, generalization, performance improvement, etc. Based on a proposed clustering algorithm for the training pairs, a new neural network, namely the range-dependent neural network (RDNN) has been developed for better accuracy in hydrological time series prediction. The applicability and potentials of the RDNN in daily streamflow and annual reservoir inflow prediction are examined using data from two watersheds in China. Empirical comparisons of the predictive accuracy, in terms of the model efficiency R² and absolute relative errors (ARE), between the RDNN, back-propagation (BP) networks and the threshold auto-regressive (TAR) model are made. The case studies demonstrated that the RDNN network performed significantly better than the BP network, especially for reproducing low-flow events.     

Identification of the best architecture of a multilayer perceptron in modeling daily total ozone concentration over Kolkata,India

Autoregressive neural network (AR-NN) models of various orders have been generated in this work for the daily total ozone (TO) time series over Kolkata (22.56°N, 88.5°E). Artificial neural network in the form of multilayer perceptron (MLP) is implemented in order to generate the AR-NN models of orders varying from 1 to 13. An extensive variable selection method through multiple linear regression (MLR) is implemented while developing the AR-NNs. The MLPs are characterized by sigmoid non-linearity. The optimum size of the hidden layer is identified in each model and prediction are produced by validating it over the test cases using the coefficient of determination (R ²) and Willmott’s index (WI). It is observed that AR-NN model of order 7 having 6 nodes in the hidden layer has maximum prediction capacity. It is further observed that any increase in the orders of AR-NN leads to less accurate prediction.