Article
Computer Science, Artificial Intelligence
Alex Wozniakowski, Jayne Thompson, Mile Gu, Felix C. Binder
Summary: The paper presents a reformulation of the standard formulation of gradient boosting algorithm, allowing it to improve nonconstant models and introducing a variant of multi-target stacking. Experimental results demonstrate that the approach outperforms state-of-the-art calibration models even with limited training examples, and significantly surpasses LightGBM and a data-driven reimplementation of the calibration model.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2021)
Article
Engineering, Civil
Jihong Qu, Kun Ren, Xiaoyu Shi
Summary: This study proposed a two-stage wrapper model for streamflow forecasting with high-dimensional candidate input variables, showing superior performance over commonly used models in terms of four error evaluation criteria. The findings suggest that the root mean square error is a more suitable criterion for evaluating the optimal objective function in the proposed model.
WATER RESOURCES MANAGEMENT
(2021)
Article
Computer Science, Information Systems
Yifei Zhang, Jue Wang, Lean Yu, Shouyang Wang
Summary: This study proposes a novel forecast combination method that reduces overfitting risk and improves forecast's generalization ability. Experimental results show that this approach outperforms individual models and other combination methods in forecasting gold, silver, and crude oil prices.
INFORMATION SCIENCES
(2022)
Article
Engineering, Civil
Mohammad Akbarian, Bahram Saghafian, Saeed Golian
Summary: This study evaluates 1 to 3-month runoff forecasts in 30 basins in Iran using the Copernicus Climate Change Service (C3S) data store. The results show that C3S runoff ensembles have the highest impact on forecast accuracy, followed by precipitation and temperature. The ANN, XGBoost, and RF models performed the best, while the MLR and SVR models performed the worst.
JOURNAL OF HYDROLOGY
(2023)
Article
Engineering, Civil
Jia Wang, Xu Wang, Soon Thiam Khu
Summary: This study proposes a hybrid decomposition-based multi-model and multi-parameter ensemble streamflow forecast method, which combines signal decomposition and artificial intelligence models to improve the accuracy and efficiency of streamflow prediction. The results demonstrate that this method effectively reduces forecast uncertainty and expands ensemble size, making it suitable for nonlinear and non-stationary hydrological series forecasting.
JOURNAL OF HYDROLOGY
(2023)
Article
Computer Science, Information Systems
Jose Fernando De Toledo, Hugo Valadares Siqueira, Lucas Henrique Biuk, Rodrigo Sacchi, Rodrigo Da Rosa Azambuja, Roberto Asano Junior, Patricia Teixeira Leite Asano
Summary: Hydroelectricity is a major source of electricity in Brazil and plays a crucial role in reducing carbon emissions. However, climate change poses challenges for the hydroelectric sector, and this study aims to improve streamflow predictions for better performance.
Article
Engineering, Civil
Xuehua Zhao, Hanfang Lv, Shujin Lv, Yuting Sang, Yizhao Wei, Xueping Zhu
Summary: This study proposes a new hydrological prediction model ICEEWT-IGWO-GRU, which combines empirical wavelet transform and improved complete ensemble empirical mode decomposition with adaptive noise, applies gated recurrent unit deep learning, and improved grey wolf optimizer to enhance accuracy and robustness of streamflow prediction. In comparison with other models, the results show that this model demonstrates superior performance in monthly streamflow forecasting.
JOURNAL OF HYDROLOGY
(2021)
Article
Engineering, Civil
Amir Mazrooei, A. Sankarasubramanian, Andrew W. Wood
Summary: This study introduces a novel VAR DA method which uses a basin-wide scaling factor to update the soil moisture conditions of the VIC model, resulting in improved hydrologic predictions.
JOURNAL OF HYDROLOGY
(2021)
Article
Mathematics
Dylan Norbert Gono, Herlina Napitupulu, Firdaniza
Summary: This study presents a forecasting method for silver prices using XGBoost with hyperparameter tuning. The best model predicts a decline in prices for the first two days, followed by an increase on the third day, another decrease on the fourth day, and a stable increase on the fifth and sixth days. Compared to other ensemble models, the proposed models exhibit the best performance.
Article
Engineering, Civil
Francesco Granata, Fabio Di Nunno, Giovanni de Marinis
Summary: Prediction of river flow rates is a challenging task due to the high uncertainty associated with basin characteristics, hydrological processes, and climatic factors. This study compares two different daily streamflow prediction models and finds that they have comparable forecasting capabilities. The stacked model based on the Random Forest and Multilayer Perceptron algorithms outperforms the bi-directional LSTM network model in predicting peak flow rates, but is less accurate in forecasting low flow rates. The prediction accuracy of both models decreases as the forecast horizon increases. The length of the time series and the presence of outliers in the data can also affect the accuracy of the prediction models.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Civil
Xingsheng Shu, Yong Peng, Wei Ding, Ziru Wang, Jian Wu
Summary: In this study, two innovative models, DirCNN and DRCNN, are proposed for multi-step-ahead monthly streamflow forecasting. Compared to traditional models, DirCNN and DRCNN outperform the comparison models and demonstrate good consistency in forecasting accuracy with an increase in lead time. The stacking order of candidate sequences has little effect on the forecasting accuracy of DirCNN and DRCNN.
WATER RESOURCES MANAGEMENT
(2022)
Article
Engineering, Civil
Jincheng Zhou, Dan Wang, Shahab S. Band, Changhyun Jun, Sayed M. Bateni, M. Moslehpour, Hao-Ting Pai, Chung-Chian Hsu, Rasoul Ameri
Summary: This study aimed to forecast the monthly river discharge time-series of two gauging hydrometric sites on the Missouri River using two machine learning models (XGB and KNN). XGB outperformed KNN in forecasting river flow. Wavelet analysis was incorporated to develop hybrid W-XGB and W-KNN approaches. Two novel hybrid models, XGB-LJA and W-XGB-LJA, were established through the hybridization of XGB with the Levy-Jaya optimization algorithm and simultaneous integration of wavelet analysis and LJA with XGB. The performance of the models was evaluated using RMSE, MAE, MBE, R, and NSE. The best discharge forecasts were obtained using the hybrid WXGB2-LJA and W-XGB4-LJA models.
WATER RESOURCES MANAGEMENT
(2023)
Article
Geosciences, Multidisciplinary
Johannes Laimighofer, Michael Melcher, Gregor Laaha
Summary: Accurate prediction of seasonal low flows is crucial for water management tasks. This study proposes an extreme gradient tree boosting model for predicting monthly low flows in ungauged catchments. The model focuses on the lowest values and uses an expectile loss function to enhance prediction accuracy. Tests and evaluations reveal that the model with expectile value 0.5 performs the best.
HYDROLOGY AND EARTH SYSTEM SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Laura Fdez-Diaz, Jose Ramon Quevedo, Elena Montanes
Summary: Hyperparameter Optimization (HPO) is aimed at improving predictive performance by tuning hyperparameters. Traditional methods choose the best performing hyperparameter configuration after multiple trials, but some works propose ensemble methods to take advantage of training with different configurations. These ensemble methods involve averaging or weighting model predictions, and more sophisticated strategies such as the Caruana method or stacking have been introduced in AutoML frameworks.
Article
Geosciences, Multidisciplinary
Leah A. Jackson-Blake, Francois Clayer, Elvira de Eyto, Andrew S. French, Maria Dolores Frias, Daniel Mercado-Bettin, Tadhg Moore, Laura Puertolas, Russell Poole, Karsten Rinke, Muhammed Shikhani, Leon van der Linden, Rafael Marce
Summary: This study examines the value of seasonal forecasting for decision-making in extratropical regions and highlights the need to reduce forecast uncertainty and develop practical experience before incorporating forecasts into operational decision-making.
HYDROLOGY AND EARTH SYSTEM SCIENCES
(2022)
Article
Engineering, Civil
Yufei Ma, Ping-an Zhong, Bin Xu, Feilin Zhu, Qingwen Lu, Han Wang
Summary: This study compares parallel dynamic programming (SPDP) with parallel particle swarm optimization (SPPSO) via cloud computing for the optimal operation of a large-scale reservoir system. The results show that SPDP outperforms SPPSO in terms of parallel performance and precision, with SPPSO having faster convergence speed but lower precision compared to SPDP. Overall, DP solves more accurately and efficiently than PSO via parallel cloud computing, ensuring global search capability of the algorithm. Cloud computing is highlighted as flexible, economical, safe, and with high practical value and application prospects.
JOURNAL OF HYDROLOGY
(2021)
Article
Engineering, Civil
Bin Xu, Xin Huang, Ran Mo, Ping-an Zhong, Qingwen Lu, Hanwen Zhang, Wei Si, Jianfeng Xiao, Yu Sun
Summary: This study proposed an integrated flood risk identification, analysis, and diagnosis model framework to enhance the accuracy of flood risk analysis and support reliable real-time flood control operation of a multireservoir system. The copula function accurately described and quantified the multidimensional temporal and spatial dependences of uncertainties, revealing positive-dominated dependences of forecast uncertainties. Neglecting spatial and high-order temporal dependences could underestimate flood risks in a reservoir system, highlighting the importance of considering all dependences in risk analysis and operation strategies.
JOURNAL OF HYDROLOGY
(2021)
Article
Engineering, Civil
Qingwen Lu, Ping-an Zhong, Bin Xu, Feilin Zhu, Xin Huang, Han Wang, Yufei Ma
Summary: The study proposed a risk-based method for floodwater utilization in a multi-reservoir system, which involves aggregation-decomposition approach and stochastic programming model to maximize hydropower generation. Joint operation and dynamic control of flood limited water levels significantly increase power generation and improve floodwater utilization rate compared to the original design model.
JOURNAL OF HYDROLOGY
(2021)
Article
Engineering, Civil
Yufei Ma, Ping-an Zhong, Bin Xu, Feilin Zhu, Jieyu Li, Han Wang, Qingwen Lu
Summary: A novel parallel dynamic programming algorithm based on Spark via cloud computing is proposed for optimal operation of reservoir system. The efficiency of cloud-based PDPoS is influenced by factors like the number of CPU cores, with high stability and good scalability. Cloud computing has rich resources and good portability of online operations, providing an attractive alternative for large-scale reservoir system operation.
WATER RESOURCES MANAGEMENT
(2021)
Article
Environmental Sciences
Jisi Fu, Ping-An Zhong, Bin Xu, Feilin Zhu, Juan Chen, Jieyu Li
Summary: Research on water resources allocation models constructed based on game theory and multi-objective optimization provides solutions to water-related conflicts. The results indicate that the AR-MOEA method outperforms the NSGA-II method in terms of diversity metric. The asymmetric game model can achieve the same water resources allocation scheme as the multi-objective optimization model under specific preference structures and directly obtain solutions through negotiation.
Article
Engineering, Civil
Xin Huang, Bin Xu, Ping-an Zhong, Hongyi Yao, Hao Yue, Feilin Zhu, Qingwen Lu, Yu Sun, Ran Mo, Zhen Li, Weifeng Liu
Summary: The study established a robust multiobjective operation and risk decision-making model for informing reservoir operations. The model utilizes copula function, robust multiobjective optimization, adaptive reference multiobjective evolutionary algorithm, and TOPSIS multicriteria decision-making method to achieve more widely distributed noninferior solutions than traditional models. The methodologies were verified by application to Xianghongdian reservoir in China and proved to be effective in reducing maximum release and improving system vulnerability.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Civil
Yufei Ma, Ping-an Zhong, Bin Xu, Feilin Zhu, Luhua Yang, Han Wang, Qingwen Lu
Summary: The ISO model is widely used for the mid-long term optimal operation of reservoir systems. In this study, a generative adversarial network named DC-WGAN was developed as a runoff generation model, showing strong performance in replicating the spatial-temporal correlation of ten-day runoff series. This proposed methodology has high practical value and application prospects.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Civil
Qingwen Lu, Ping-an Zhong, Bin Xu, Xin Huang, Feilin Zhu, Han Wang, Yufei Ma
Summary: This study analyzes the spatial correlation of reservoir initial water level errors and the spatiotemporal correlation of flood forecast errors using the copula function. It establishes a risk analysis model for multi-objective flood control operation of a complex reservoir system, considering multiple uncertainties. The study finds that neglecting the correlations between errors underestimates the flood control risk. There is a competitive tradeoff between upstream and downstream objectives, and choosing a compromise solution balances the risks.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Civil
Yu Zhang, Haifei Sha, Xiufeng Wu, Shiqiang Wu, Jiangyu Dai, Bin Xu, Lei Yu, Qianqian Yang
Summary: This paper proposes a contract volume risk decision model based on rainfall uncertainty and efficiency uncertainty of rainwater harvesting system (RHS) to provide decision support for determining contract volumes in rainwater resource forward transactions. The model quantifies the uncertainty of rainfall using a mathematical model and determines RHS efficiency by simulating the operation of RHS. The risk of rainwater resource supply is defined as the probability of not being able to fulfill the contract volume, and different decision-making processes are proposed for different negotiation scenarios.
WATER RESOURCES MANAGEMENT
(2022)
Article
Environmental Sciences
Bin Xu, Yu Sun, Xin Huang, Ping-an Zhong, Feilin Zhu, Jianyun Zhang, Xiaojun Wang, Guoqing Wang, Yufei Ma, Qingwen Lu, Han Wang, Le Guo
Summary: This study proposes a multiobjective robust optimization and decision-making framework to minimize the multiple risks of a cascade hydropower system through robust operation. The framework includes risk analysis, robust control, and decision-making models. The findings suggest that robust optimization can effectively reduce the risk values of ecological water shortfall, consumptive water shortfall, and energy shortfall compared to chance-constrained programming, but it also leads to tradeoffs in energy production efficiency and water usage efficiency.
WATER RESOURCES RESEARCH
(2022)
Article
Engineering, Civil
Zhong-kai Feng, Wen-jing Niu, Xin-yu Wan, Bin Xu, Fei-lin Zhu, Juan Chen
Summary: This study proposes a hybrid model combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and twin support vector machine (TSVM) for accurate hydrologic forecasting. Experimental results show that the hybrid model significantly outperforms conventional models in prediction accuracy.
JOURNAL OF HYDROLOGY
(2022)
Article
Engineering, Civil
Jieyu Li, Ping-an Zhong, Yuanjian Wang, Minzhi Yang, Jisi Fu, Weifeng Liu, Bin Xu
Summary: This paper evaluates the reliability of the multi-reservoir real-time flood control hybrid operation model in real-time flood control by conducting a risk analysis, using the case study of the Huaihe River basin to analyze the impact of model structure reduction on flood risk and related factors. The results indicate that reducing the model structure does not significantly increase flood risk, but considering the uncertainties in flood forecasting and model structure may lead to 65% increased risk.
JOURNAL OF HYDROLOGY
(2022)
Article
Environmental Sciences
Han Wang, Ping-an Zhong, Ervin Zsoter, Christel Prudhomme, Florian Pappenberger, Bin Xu
Summary: This paper aims to improve flood forecasting by comparing the simulation results of a global hydrological forecast system and a regional system, and testing their influence on input data. The results showed that the global system had poorer simulation results due to lower input data quality. However, the global system had higher forecast quality in terms of high flow and longer lead times. Quantile mapping was effective in eliminating errors in input data, model, and initialization.
Article
Engineering, Civil
Yubin Chen, Manlin Wang, Yu Zhang, Yan Lu, Bin Xu, Lei Yu
Summary: In this study, a coordination model of power generation and ecological flow for cascade hydropower system was established using a multi-objective optimization method. Three MOGAs (NSGA-II, NSGA-III, and RVEA) were selected to solve the model, and the results show that NSGA-III has certain advantages over NSGA-II and RVEA in terms of performance metrics. Therefore, NSGA-III is recommended as the solution algorithm for the established coordination model.
WATER RESOURCES MANAGEMENT
(2023)
Article
Engineering, Civil
Ran Mo, Bin Xu, Ping-an Zhong, Yuanheng Dong, Han Wang, Hao Yue, Jian Zhu, Huili Wang, Guoqing Wang, Jianyun Zhang
Summary: Accurate, reliable, and stable streamflow forecasts are crucial for water resources management. This study proposes a long-term probabilistic streamflow forecast model, which improves the inputs selection, structure designation, and error parameters calibration of the traditional forecast model. The proposed model incorporates predictor optimization, deterministic forecasting using the LSTM model, and time-varying error identification using the GARCH model. Case studies on two lakes in China demonstrate that the proposed model significantly improves the accuracy, reliability, and stability of streamflow forecasts compared to the benchmark.
JOURNAL OF HYDROLOGY
(2023)
