Xinting Yu

Bio: Xinting Yu is an academic researcher from Zhejiang University. The author has contributed to research in topics: Water resources & Water supply. The author has an hindex of 1, co-authored 2 publications receiving 1 citations.

AI-based techniques for multi-step streamflow forecasts: application for multi-objective reservoir operation optimization and performance assessment

Abstract: . Streamflow forecasts are traditionally effective in mitigating water scarcity and flood defense. This study developed an artificial intelligence (AI)-based management methodology that integrated multi-step streamflow forecasts and multi-objective reservoir operation optimization for water resource allocation. Following the methodology, we aimed to assess forecast quality and forecast-informed reservoir operation performance together due to the influence of inflow forecast uncertainty. Varying combinations of climate and hydrological variables were input into three AI-based models, namely a long short-term memory (LSTM), a gated recurrent unit (GRU), and a least-squares support vector machine (LSSVM), to forecast short-term streamflow. Based on three deterministic forecasts, the stochastic inflow scenarios were further developed using Bayesian model averaging (BMA) for quantifying uncertainty. The forecasting scheme was further coupled with a multi-reservoir optimization model, and the multi-objective programming was solved using the parameterized multi-objective robust decision-making (MORDM) approach. The AI-based management framework was applied and demonstrated over a multi-reservoir system (25 reservoirs) in the Zhoushan Islands, China. Three main conclusions were drawn from this study: (1) GRU and LSTM performed equally well on streamflow forecasts, and GRU might be the preferred method over LSTM, given that it had simpler structures and less modeling time; (2) higher forecast performance could lead to improved reservoir operation, while uncertain forecasts were more valuable than deterministic forecasts, regarding two performance metrics, i.e., water supply reliability and operating costs; (3) the relationship between the forecast horizon and reservoir operation was complex and depended on the operating configurations (forecast quality and uncertainty) and performance measures. This study reinforces the potential of an AI-based stochastic streamflow forecasting scheme to seek robust strategies under uncertainty.

Complex water resource system optimal configuration method based on partition-classification theory

Abstract: The invention discloses a complex water resource system optimal configuration method based on a partitioning-grading theory, and the method comprises the following steps: obtaining the basic information data of a water resource system, and carrying out the generalization of the complex water resource system; then partitioning the whole water resource system according to the water supply relationship among a reservoir, a pump station, a riverway, a water conveying pipeline and a water plant, and establishing a digital matrix of the hydraulic relationship of different partitions; and finally, constructing and solving a multi-stage water resource system optimal configuration model according to the type of the water supply source, and determining a reasonable optimal configuration scheme set.According to the invention, a complex water resource system can be simplified, and a new way is provided for efficient utilization of water resources.

Multi-objective operation of cascade reservoirs based on short-term ensemble streamflow prediction

Abstract: • Establish numerical weather prediction-based ensemble streamflow prediction (ESP) models of cascade reservoirs. • Develop a forecast-informed operating rule based on Gaussian Radial Bias Function. • Investigate the application value of ensemble forecast in multi-objective operation of Hanjiang cascade reservoirs. Ensemble streamflow predictions (ESPs) play an essential role in reservoir operation in terms of flood control, water supply and hydropower generation, etc. However, the working mechanism of dam-regulated ESP is rarely explored even though a large fraction of major waterways is affected by artificial hydraulic infrastructure, let alone its application value in multi-objective multi-reservoir operating system. Here this study develops three-layered hydrologically-relevant long and short-term memory (LSTM) models to implement 1–3 days ahead ESP of cascade reservoirs with the help of numerical weather prediction (NWP), and designs operating policies based on Gaussian radial basis function which can explicitly incorporate ESP information into reservoir operating decision. The framework of our proposed ESP-based operating way is demonstrated on a cascade reservoir system in the upper Hanjiang River basin, China and assessed against two benchmark control ways (i.e., no-forecast and deterministic-forecast operations). Results show that (1) LSTM models have reasonable accuracy in short-term NWP-based ESP of cascade reservoirs, especially for high-flow regimes; (2) compared with the no-forecast operation, our ESP-based operation could harvest additional approximately 36 million kW·h of hydropower generation per year with a slight improvement in water supply over 11 years of operation during the flood season. These findings highlight the application potential of our numerically efficient and skillful NWP-based ESP scheme for multi-objective cascade reservoir operation to address water-energy shortage issues.

Improving long-term streamflow prediction in a poorly gauged basin using geo-spatiotemporal mesoscale data and attention-based deep learning: A comparative study

Abstract: Precise long-term streamflow prediction has always been important in the hydrology field, and has provided essential information for efficient water-resource management and disaster prevention. Attention to this field has increased recently owing to water and climate crises. Despite the remarkable improvements in existing data-driven models, they still have weaknesses, particularly for multistep predictions in poorly gauged basins. The purpose of this study is to improve the multistep ahead prediction ability by considering mesoscale hydroclimate data as booster predictors and employing attention-based deep learning. Therefore, a meta-algorithm is proposed for analyzing time-series data in complex geo-spatiotemporal environments to assess the superiority of prediction with booster predictors through direct and direct-recursive hybrid strategies. In the subsequent stage, a novel, integrated neural network architecture is demonstrated that couples a deep convolutional neural network (CNN) with a deep attention network. In particular, the former network performs automatic feature engineering, and the latter focuses on lengthy sequence complexities. Four state-of-the-art combinations for 12-month ahead prediction are introduced, including pairs of TimeDistributed-CNN (TD-CNN) and 3D-CNN along with a Long- and Short-term Time-series network (LSTNet) or a Transformer network. Moreover, a base architecture was employed for model comparison that contains the ConvLSTM2D layer that is compatible with multidimensional time-series data. For this, all models were applied to the Karkheh River basin in the northeast of the Persian Gulf, where monthly historical records of streamflow are available from 1955 to 2021. The results revealed that the application of hydroclimate sea surface temperature data and mean surface level pressure along with local data increased the prediction accuracy. Moreover, the proposed integrated networks delivered more accurate long-term streamflow predictions than the base models through various evaluation criteria, including r, R2, mean absolute error, root-mean-square error, Kling–Gupta efficiency, Willmott's Index, Legates–McCabe's, and the Akaike information criterion. In summary, the 3D-CNN–Transformer achieved the best performance, followed by the TD-CNN–Transformer, TD-CNN–LSTNet, and 3D-CNN–LSTNet with R2 values equal to 0.952, 0.930, 0.900, and 0.837, respectively. This study demonstrates that the application of hydroclimate data with proposed integrated networks are particularly useful for poorly gauged basin. Thus, the proposed models can potentially improve multistep ahead streamflow prediction compared to univariate and equation-based models.

Monitoring the extreme flood events in the Yangtze River basin based on GRACE and GRACE-FO satellite data

Abstract: Abstract. Gravity Recovery and Climate Experiment (GRACE) and its successor GRACE Follow-on (GRACE-FO) satellite provide terrestrial water storage anomaly (TWSA) estimates globally that can be used to monitor flood in various regions at monthly intervals. However, the coarse temporal resolution of GRACE and GRACE-FO satellite data has been limiting their applications at finer temporal scales. In this study, TWSA estimates have been reconstructed and then temporally downscaled into daily values based on three different learning-based models, namely a multi-layer perceptron (MLP) model, a long-short term memory (LSTM) model and a multiple linear regression (MLR) model. Furthermore, a new index incorporating temporally downscaled TWSA estimates combined with daily average precipitation anomalies is proposed to monitor the severe flood events at sub-monthly timescales for the Yangtze River basin (YRB), China. The results indicated that (1) the MLP model shows the best performance in reconstructing the monthly TWSA with root mean square error (RMSE) = 10.9 mm per month and Nash–Sutcliffe efficiency (NSE) = 0.89 during the validation period; (2) the MLP model can be useful in temporally downscaling monthly TWSA estimates into daily values; (3) the proposed normalized daily flood potential index (NDFPI) facilitates robust and reliable characterization of severe flood events at sub-monthly timescales; (4) the flood events can be monitored by the proposed NDFPI earlier than traditional streamflow observations with respect to the YRB and its individual subbasins. All these findings can provide new opportunities for applying GRACE and GRACE-FO satellite data to investigations of sub-monthly signals and have important implications for flood hazard prevention and mitigation in the study region.