Multi-step reservoir inflow prediction using a rolling window strategy and decomposed LSTM

Reservoir inflow prediction plays a significant role in the field of hydrological prediction. Accurate and reliable prediction of reservoir inflow is the key to flood control decision. In this paper, we combine the deep belief network (DBN) with the Long Short-Term Memory (LSTM) to present a hybrid model based on deep learning (HDL) for reservoir inflow prediction. We take a full consideration of the basin flow and rainfall factors, which significantly affect the inflow flow. According to the rainfall data, we divide the corresponding flow data into two cases: rain and no rain. The proposed approach consists of three parts: we apply the DBN to learn the characteristics of the flow data and get predicted values of the reservoir inflow in case of rain and in case of no rain respectively. Then, we use the basin rainfall data and adopt the LSTM to fit the differential between the predicted inflow value generated by DBN in case of rain and the real inflow value, and get the predicted differential. Finally, the outputs of these three parts are added to obtain the final predicted result. Experiments are evaluated by the historical flow and rainfall data of a reservoir in China, and the results have proved that our method is effective and has higher prediction accuracy.

Climate change has significant impacts on changing precipitation patterns causing the variation of the reservoir inflow. Nowadays, Indonesian hydrologist performs reservoir inflow prediction according to the technical guideline of Pd-T-25-2004-A. This technical guideline does not consider the climate variables directly, resulting in significant deviation to the observation results. This research intends to predict the reservoir inflow using the statistical downscaling (SD) of General Circulation Model (GCM) outputs. The GCM outputs are obtained from the National Center for Environmental Prediction/National Center for Atmospheric Research Reanalysis (NCEP/NCAR Reanalysis). A new proposed hybrid SD model named Wavelet Support Vector Machine (WSVM) was utilized. It is a combination of the Multiscale Principal Components Analysis (MSPCA) and nonlinear Support Vector Machine regression. The model was validated at Sutami Reservoir, Indonesia. Training and testing were carried out using data of 1991–2008 and 2008–2012, respectively. The results showed that MSPCA produced better extracting data than PCA. The WSVM generated better reservoir inflow prediction than the one of technical guideline. Moreover, this research also applied WSVM for future reservoir inflow prediction based on GCM ECHAM5 and scenario SRES A1B.

Prediction of reservoir inflow is an important aspect of water supply management in an urbanized region. In this regard, this study aims to improve the reservoir inflow prediction using the calibration of the MODIS (Moderate resolution Imaging Spectroradiometer) evapotranspiration (ET) data in addition to the streamflow in the SWAT (Soil and Water Assessment Tool) model. The results of this study show that constraining SWAT parameters to the ET combined with streamflow helps to improve the simulation of ET. Thereby, it enhances the representation of vertical fluxes in regional hydrology. The reservoir inflow was calibrated with streamflow alone with Nash-Sutcliffe Efficiency (NSE) of 0.63, whereas the inclusion of ET provides an NSE of 0.59. However, the simulation of ET has improved by 10%. The results of this study demonstrate that the inclusion of ET data helps to improve the simulation of hydrologic processes in the region.

Flooding occurs frequently compared to other natural disasters. Less developed countries are severely affected by floods. This research provides an integrated hydrometeorological system that forecasts hourly reservoir inflows using a full physically based rainfall–runoff and numerical weather models. This study develops a 5-day lead time reservoir inflow prediction using TIGGE ensemble datasets from ECMWF, UKMO, and NCEP for the Dharoi Dam in Gujarat, India. The ensemble data were post-processed using censored non-homogeneous Linear Regression and Bayesian model averaging approach. These post-processed data were used in a hydrological model to simulate hydrological processes and predict Dharoi Dam reservoir inflows. Results show that ECMWF with a BMA approach and HEC-HMS hydrological model can predict reservoir inflows in the Sabarmati River basin. The correlation result of an observed reservoir inflow is 0.91. This research can help regional water resource managers and government officials to plan and manage water resources.

Although there have been many successful applications of artificial neural networks (ANNs) to capture non linear relationship of inflow into a reservoir, there are cases when ANNs have not been able to predict flow extremes accurately. Applicability of non linear model based on ANN with a random component embedded is explored for Pawana reservoir in Upper Bhima River Basin, Maharashtra, India. Suitability of time lagged recurrent networks (TLRNs) with time delay, gamma and laguarre memory structures is investigated for predicting seasonal (June to October) reservoir inflow with a monthly time step. Trial and error procedure is adopted in selecting input nodes and hidden nodes. Input to the network includes preceding records of reservoir inflow (varying from one lag to four lags). Performance of back propagation through time (BPTT) algorithm is investigated for various inputs assuming different training and testing combinations. Due to large variation in the observed data values, transformation of the series is also tried and found that the network trained for 50:50 training and testing data sets predicted better values. The validation of the models was performed using comparison of principal statistics (mean, standard deviation, skewness and kurtosis), goodness-of-fit measures (MSE, MAE, MRE and R), time series plots and scatter plots. Encouraging results obtained in the present study indicated that the log-transformed, BPTT trained TLRN resulted in better and reliable forecast of extreme high and low inflows as compared to non-transformed series.

Real-time optimization of a large-scale reservoir operation in Thailand using adaptive inflow prediction with medium-range ensemble precipitation forecasts

The inflow forecasting is one of the most important technologies for modern hydropower station. Under the joint influence of soil, upstream inflow, and precipitation, the inflow is often characterized by time lag, nonlinearity, and uncertainty and then results in the difficulty of accurate multistep prediction of inflow. To address the coupling relationship between inflow and the related factors, this paper proposes a long short-term memory deep learning model based on the Bagging algorithm (Bagging-LSTM) to predict the inflows of future 3 h, 12 h, and 24 h, respectively. To validate the proposed model, the inflow and related weather data come from a hydropower station in southern China. Compared with the classical time series models, the results show that the proposed model outperforms them on different accuracy metrics, especially in the scenario of multistep prediction.

Reservoir inflow prediction is crucial for water resource management, yet existing approaches mainly focus on single-reservoir models that ignore spatial dependencies among interconnected reservoirs. We introduce AdaTrip as an adaptive, time-varying graph learning framework for multi-reservoir inflow forecasting. AdaTrip constructs dynamic graphs where reservoirs are nodes with directed edges reflecting hydrological connections, employing attention mechanisms to automatically identify crucial spatial and temporal dependencies. Evaluation on thirty reservoirs in the Upper Colorado River Basin demonstrates superiority over existing baselines, with improved performance for reservoirs with limited records through parameter sharing. Additionally, AdaTrip provides interpretable attention maps at edge and time-step levels, offering insights into hydrological controls to support operational decision-making. Our code is available at https://github.com/humphreyhuu/AdaTrip.

ABSTRACTA model fusion approach was developed based on five artificial neural networks (ANNs) and MODIS products. Static and dynamic ANNs – the multi-layer perceptron (MLP) with one and two hidden layers, general regression neural network (GRNN), radial basis function (RBF) and nonlinear autoregressive network with exogenous inputs (NARX) – were used to predict the monthly reservoir inflow in Mollasadra Dam, Fars Province, Iran. Leaf area index and snow cover from MODIS, and rainfall and runoff data were used to identify eight different combinations to train the models. Statistical error indices and the Borda count method were used to verify and rank the identified combinations. The best results for individual ANNs were combined with MODIS products in a fusion model. The results show that using MODIS products increased the accuracy of predictions, with the MLP with two hidden layers giving the best performance. Also, the fusion model was found to be superior to the best individual ANNs.

Regardless of multiple reservoirs built across the rivers to control the flow of water bodies, yet many calamities in low lying areas have occurred in recent times. One of the reasons for these is the legacy techniques being in dams for flow management. Since Machine Learning algorithms are making good progress in accurately predicting future probabilities based on past data by using the statistical methods as its basis these techniques can be applied to train the machine model on weather reports and Dam flow control and capacity data so as to provide efficient control over Dam water level management and create better alert systems in case of calamities. This paper presents an evaluation of a few machine learning algorithms like LOWESS, Logistic Regression, and deep learning techniques based on Recurrent neural networks to predict Reservoir Inflow. Also, it makes use of the ensembling/ bagging technique on the results of the aforementioned algorithms to improve the accuracy of the model.

Predicting reservoir inflow is important to reservoir management and scheduling; therefore, accurate prediction of reservoir inflow has been a vital task for researchers and water resources managers during last decades. Recently, Adaptive Neuro-Fuzzy Inference System (ANFIS) has been extensively used to find the relationship between inputs and outputs without considering the physical process of the phenomena. Therefore, the current study investigates and evaluates the capability and applicability of the ANFIS technique in modeling the reservoir inflow for predicting future values of monthly inflow to Darbandikhan reservoir in Kurdistan Region, Iraq. Five rain gage stations data from 1992-2009 were used to develop ANFIS models to predict monthly reservoir inflow from various combinations of antecedent values of inflow and rainfall within the basin in Iraq and Iran which are considered as input variables. Two different membership functions (MFs), triangular and generalized bell, with different numbers (2-5) for each input variable were investigated in the ANFIS models. The best fit models were selected using three performance evaluation criteria, namely; coefficient of determination ( R 2), coefficient of efficiency ( CE ) and normalized root mean squared error ( NRMSE ). In comparison of ANFIS model uses the triangular MF with that model uses the generalized bell MF, approximately, similar predict accuracies were obtained for both MFs except the latter needs fewer numbers of MFs. The results indicated that using a combination of inflow and rainfall as input variables is effective in improving predict accuracy and developing parsimonious models with fewer and readily available input variables. Moreover, among different architectures of ANFIS, structures, including three input variables, 1 and 12 time lags of inflow ( It-1 , It-12 ) and rainfall of Marivan station ( ) showed better performance for this application. For best models, the performance evaluation criteria, , and , for checking data set were obtained as 0.96, 0.95 and 0.2 respectively for model with four triangular MFs; and as 0.96, 0.96 and 0.2 respectively for model with three generalized bell MFs. Reservoir inflow modeling in this way will be more reliable than doing it using a time series model as a more effective parameter could be incorporated. By considering the results, ANFIS method is an effective tool that can be successfully applied for reservoir inflow modeling and have satisfactory performances in Darbandikhan reservoir monthly inflow prediction.

This study coupled the ensemble learning method with residual error (RE) correction to propose a more accurate hydrologic model for the time-series prediction of the reservoir inflow. To enhance the prediction capability of the model in mountain catchments, three deep learning (DL) models, namely the encoder–decoder gated recurrent units (ED-GRU), encoder–decoder long short-term memory network (ED-LSTM), and combined convolutional neural network with LSTM (CNN-LSTM), were deployed to train reservoir inflow prediction model for the lead times of 1–24 h. The prediction outputs from three DL models were then incorporated into the categorical gradient boosting regression (CGBR) model to resolve the highly non-linear relationship between model inputs and outputs. In the final procedure, the RE correction method was implemented by using the outcomes of the CGBR model to construct the proposed hybrid model. The proposed model was applied to simulate the hourly inflow in the Shihmen and Feitsui Reservoirs. The proposed model achieved improved performance by an average proportion of 66.2% compared to the three DL models. It is demonstrated that the proposed model is accurate in predicting the reservoir peak and total inflows and also performs well for storm events with multi-peak hydrographs.

Two-step daily reservoir inflow prediction using ARIMA-machine learning and ensemble models

Reservoir inflow prediction is a vital subject in the field of hydrology because it determines the flood event. The negative impact of the floods could be minimized greatly if the flood frequency is predicted accurately in advance. In the present study, a novel hybrid model, bootstrap quadratic response surface is developed to test daily streamflow prediction. The developed bootstrap quadratic response surface model is compared with multiple linear regression model, first-order response surface model, quadratic response surface model, wavelet first-order response surface model, wavelet quadratic response surface model, and bootstrap first-order response surface model. Time series data of monsoon season (1 July to 30 September) for the year 2010 of the Chenab river basin are analyzed. The studied models are tested by using performance indices: Nash–Sutcliffe coefficient of efficiency, mean absolute error, persistence index, and root mean square error. Results reveal that the proposed model, i.e., bootstrap quadratic response surface shows good performance and produces optimum results for daily reservoir inflow prediction than other models used in the study.

In water resources reservoir plays an important role as it functions for different purposes in different times such as providing effective water storage, water for hydropower generation, water for irrigation activities, mitigating the impacts of disastrous environmental effects, as well as meeting the water demand for the people in drought conditions. In order to minimize the impacts which are faced by the people in downstream areas and also for the safety and maintenance of the reservoir it requires fast and accurate prediction of decisions. These impacts are mainly caused due to the uncertain weather and climate conditions, other geographical characteristics and also due to the reservoir operations. So that implementing a system which consists of accurate modelling and prediction of water levels, floods existence and flood evacuation plan, and optimization of reservoir operations based on inflow prediction in order to minimize these impacts have been taken into consideration throughout the study. Since Artificial Neural Network (ANN) shows a significant improvement in the area of hydrological modelling in last decade, multilayer perceptron (MLP) method is applied to develop the flood model. Long Short-Term Memory (LSTM) models are developed for the modelling and prediction of water levels of downstream locations and also for the reservoir inflow prediction from the upstream basin based on the nonlinear time series analysis of hydrological data around the area.