In the optimization of cascade reservoir operation for power generation,traditional methods such as dynamic programming suffer from bottlenecks like low computational efficiency,limiting their practical engineering applications.Although single-agent deep reinforcement learning(DRL)can achieve continuous control in an end-to-end manner,it struggles to effectively capture the dynamic couplings among cascade reservoirs,potentially leading to unstable policy gradients and training oscillations.Moreover,treating the cascade reservoir system as a single entity limits scalability.To address these issues,this paper proposes a multi-agent deep reinforcement learning(MADRL)method for cascade reservoir operation based on the"Centralized Training and Decentralized Execution"framework.Each reservoir is treated as an independent agent,and a collaborative network integrating local decision-making with global information sharing is constructed to enhance training stability.A global reward function is designed,combin-ing power generation benefit with constraint penalty terms.Furthermore,the Copula-Gibbs joint distribution is intro-duced to generate runoff scenarios,improving adaptability to inflow uncertainty.Finally,hyperparameter combina-tions,including network architecture,discount factor,and learning rate,are tuned through sensitivity analysis and grid search.Engineering application results show that,under identical hardware and data conditions,the proposed method can achieve faster convergence.While strictly adhering to end-of-period water level and operational safety constraints,it achieves online inference times of 6.5 to 8.5 ms in 4 and 6 reservoirs cascaded system,which is approximately two orders of magnitude faster than discrete differential dynamic programming(DDDP).Under dry,normal,and wet typical inflow scenarios,the annual power generation deviation is controlled within 0.65%,and the end-of-period water level control accuracy is satisfactory.In conclusion,the multi-agent collaborative mechanism effectively addresses challenges posed by high-dimensional decision-making and system coupling,enhancing the environmental adaptability and engineering practicality of operational strategies.This provides reliable methodologi-cal support for efficient operation of large-scale cascade reservoirs.
The hydraulic friction parameters of pipelines are among the most crucial hydraulic parameters for the operation scheduling and digital-intelligent construction of long-distance water diversion projects.For a long time,issues such as a wide variation range of recommended values and a lack of field calibration based on measured data have persisted,affecting pipe design and accurate regulation.Based on the prototype tests conducted on DN2800 and DN3200 pressure water pipelines of the Yinchuojiliao Water Diversion Project,this paper systematically studies the values and variation patterns of the Manning roughness coefficient,Hazen-Williams coefficient and equivalent rough-ness.The results indicate that the pipeline roughness coefficient decreases with increasing flow velocity and increases with increasing pipe diameter,showing a positive correlation with flow velocity to the power of-0.1.When a constant roughness coefficient is used,the calculation error for head loss can range from 10%to 20%.The measured Hazen-Williams coefficient ranges from 145 to 150,which is greater than the recommended values in most specifications.Assuming a constant Hazen-Williams coefficient can lead to a calculation error of 5%to 10%in head loss.The equivalent roughness of PCCP pipelines with cement mortar-lining should be taken as 0.05 mm.When the equivalent roughness is reduced from 0.05 to 0.01 mm for the velocity of 1.5 m/s,the roughness coefficients of DN2800 and DN3200 pipelines decrease from 0.1071 to 0.010 34 and from 0.010 83 to 0.010 46,respectively,representing reductions of 3.45%and 3.42%.This indicates that the roughness coefficient of 0.009 or lower cannot be achieved.
The patrol inspection of water conservancy projects is a core management task for ensuring the safe and stable operation of critical infrastructures.Traditional patrol inspections suffer from high reliance on manual labor,low accuracy in hazard identification,and insufficient dynamic decision-making capabilities.This paper proposes a multi-agent collaborative intelligent decision-making framework that integrates Large Language Models(LLMs)and Graph Retrieval-Augmented Generation(GraphRAG)technologies.Adopting a modular architecture encompassing perception,memory,communication,planning,and action,the framework achieves full automation of the patrol inspection processes.A multi-modal dataset was constructed using multi-source patrol inspection data from the past three years.Domain-adaptive fine-tuning significantly improved the F1 scores of multi-modal large language models in equipment recognition and defect detection by 7.2%and 6.9%,respectively.Furthermore,a dynamic knowledge graph system based on Graph RAG was developed to bridge domain-specific knowledge gaps through knowledge infu-sion techniques,while simultaneously employing an entity-relation reasoning mechanisms to effectively mitigate model hallucinations.Experimental results demonstrate that patrol inspection reports generated by this method,upon dual evaluation by both domain experts and operational maintenance personnel,accurately reflect the professional expertise and technical depth required in operations and maintenance of water conservancy projects.This research provides a novel,interpretable,and reliable technical paradigm for intelligent operation and maintenance of water infrastructure,holding significant engineering application value for advancing digital transformation within the indus-try.
Report review serves as a core link in project quality control.Traditional manual review methods face chal-lenges such as inefficiency and inconsistent standard implementation,while existing general-purpose review systems struggle to adapt to the multiple challenges posed by the complex structural characteristics,multidimensional review tasks,and domain knowledge dependencies inherent in water resources science and technology reports.To address this,this study proposed an agent-based formal review system architecture for such reports.Through agent-driven dynamic task planning and collaborative review mechanisms,adaptive parsing of complex report structures and multi-task orchestration were realized.A water resources domain knowledge enhancement mechanism was established based on LoRA fine-tuning and Retrieval-Augmented Generation(RAG)technology.A terminology knowledge base con-taining 27,005 specialized water resources terms and a computational relationship database were constructed,form-ing a command fine-tuning dataset with 10,358 samples.The system's review capabilities in tasks involving special-ized terminology,computational logic,and contractual consistency was enhanced.Benchmarked against human review results,the agent system achieved an average F1 score above 80%across eight types of review tasks.Experi-mental results demonstrate that this system enables fully intelligent processing for the formal review of water resources science and technology reports,significantly improving efficiency while ensuring review accuracy.The intelligent review system developed in this study provides standardized tool support for quality management of water resources project documentation and offers a reference technical pathway for the intelligent review of texts in specialized domains.
In recent years,extreme flash flood events in mountainous watersheds under the combined influence of cli-mate change and human activities have occurred more frequently.Characterized by sudden onset and high destructive potential,these events pose significant challenges to China's disaster prevention and mitigation efforts.This study employs the spatiotemporally-variable-source distributed hydrological model(SKY-HydroSAT),independently developed by the China Institute of Water Resources and Hydropower Research,to establish a distributed hydrologi-cal model considering reservoir regulation in the upper catchment of the Guilin City in the Lijiang River Basin.The model demonstrates good applicability,with absolute peak flow errors of 3.7%and 5.4%during the calibration and validation periods,respectively,and Nash-Sutcliffe efficiency(NSE)values of 0.87 and 0.83.Taking the"6·19"extreme flash flood event in 2024 as a case study,with the accurate modelling results(peak flow error 10.5%,NSE 0.94),the research investigated the flood dynamics under reservoir influence.The analysis reveals that this event was triggered by a bimodal heavy rainfall pattern,with a three-day average rainfall of 442 mm across the watershed above Guilin and a maximum cumulative rainfall of 777 mm at Maolingjiao Station.Simulation results indicate the formation of combined infiltration-excess and saturation-excess runoff mechanisms,along with subsurface stormflow in major runoff-producing areas,leading to rapid runoff generation and posing severe challenges to midstream reservoirs and downstream flood safety.The joint operation of 4 upstream reservoirs upstream of Guilin played a crucial role in flood control and disaster mitigation.Without reservoir regulation,the peak flow at Guilin Station would have exceeded the 100-year return period level.After regulation by the upstream reservoirs,the peak flow at Guilin Stationwas reduced to 6380 m³/s(approximately equivalent to a 30-year return period flood),significantly mitigating downstream disas-ter losses.This study validates the applicability of the distributed hydrological model in the Guilin region,demonstrat-ing its capability to accurately characterize watershed runoff mechanisms and quantify the benefits of regulation by hydraulic projects.The proposed approach holds potential for extension to other mountainous watersheds in China and can provide scientific support for regional flood management decision-making.
To address the key problems in 3D finite element mesh generation for rockfill dams with complex dam foundation topography—such as difficulties in fitting the boundaries between different material zones and filling stages,controlling the meshes of the transition zone between the dam body and foundation,and accurately represent-ing the excavation boundaries—an adaptive 3D finite element mesh generation method for rockfill dams is proposed.The method employs element transformation to adaptively fit complex boundaries of material zones and filling stages.A transition factor for the meshes of the dam body and foundation is constructed using fractal theory to control the rate of element size variation and adaptively generate the foundation transition zone.A rigorous terrain cutting surface fit-ting algorithm for 3D meshes is developed based on a chained dimension-reduction scheme of"body-face-edge-vertex",enabling the generation of a 3D dam foundation mesh that accurately reflects the excavation surface.Statisti-cal analyses of mesh quality indicators—including aspect ratio,parallel deviation,maximum angle,and Jacobian ratio—for the 3D meshes of five dams generated using this method demonstrate that the method is featured by control-lable overall mesh quality meeting the shape quality requirements for subsequent finite element analysis,and signifi-cantly improved efficiency of 3D modeling.This method enables adaptive generation of rockfill-dam finite element meshes from 2D maximum cross sections to 3D finite element models,providing an efficient and reliable preprocess-ing approach for refined numerical simulation of rockfill dam body-foundation systems with complex dam foundation topography.
Fish passage facilities have been the subject of ongoing debate in China for decades regarding their neces-sity and effectiveness since the construction of the Gezhouba dam.The aim of the present study was to critically assess theories that support and question the efficacy of fish passage facilities.Through comparative case analysis and theo-retical examination,the present study moves beyond the superficial"fish rescue debate"to explore underlying eco-logical theoretical roots and practical challenges.Three typical cases—the Gezhouba dam,Poyang Lake Water Con-trol Project,and new Three Gorges shipping channel—reveals conflict between"human-led restoration"and"preser-vation of natural processes."Based on this,this paper introduces the theory of heavily disturbed river ecology,ana-lyzes the scientific validity and effectiveness of current fish protection measures for water conservancy and hydro-power projects such as ecological flow,fish passage facilities,and proliferation and release,and demonstrates the significance of prioritizing habitat protection and restoration.A systematic strategic framework for fish conservation is developed,proposing coordinated efforts across five dimensions:integrated planning of fish conservation measures,prioritizing the restoration of natural flow regimes,systemic advancement of habitat restoration,targeted implementa-tion of stock enhancement,and scientific design of fish passage facilities.Ultimately,this framework aims to shift from project-level compensation to holistic watershed ecosystem restoration,providing scientific guidance for promot-ing the ecological construction and high-quality development of water projects in China.
Following the impoundment of the Longyangxia Reservoir in 1986,the mean flood peak discharge recorded at the Guide hydrological station decreased by 46.62%,indicating significant non-stationarity.However,methods for quantifying the reservoir regulation effect are still immature,leading to difficulties in ensuring the accu-racy of design floods,thereby seriously compromising both flood control safety and the multiple-purpose benefits of water resources projects.This study,taking the Longyangxia Reservoir as the research object,proposes the Available Flood Control Reservoir Index(ACRI)as a quantitative indicator of reservoir regulation effects.By integrating the Copula theory with the mechanism-based reconstruction method(Me-RS),a nonstationary flood frequency analysis was conducted,incorporating historical flood investigation data.The results show that the source-tracing reconstruc-tion method based on ACRI can accurately represent the reservoir's regulation effect on downstream flood peak regimes and effectively reconstruct the flood peak series into a stationary state.The design flood peaks derived using the Pearson Type Ⅲ distribution deviate by less than 3%from the original design values.Incorporating historical flood investigation data further improves the goodness-of-fit of the probability distribution,reducing the RMSE by 12.87%and increasing the 10,000-year design flood peak by 10.23%compared to the estimate that excludes histori-cal flood.Owing to flow regulation by the Longyangxia Reservoir,the design flood peak discharges at Guide for return periods of 50 to 10,000 years show an average reduction of 38.30%relative to the original design values.The nonsta-tionary flood frequency analysis method proposed in this study—accounting for reservoir regulation effects and inte-grating historical flood investigation data—can more accurately characterize the probability distribution of flood peaks,enhance the reliability of design flood estimation,and support flood control and water resources management at the river basin scale.
Ice hazards are prevalent in rivers,canals,and reservoirs in cold regions.As an important parameter influencing water heat loss and ice formation,ice concentration is used for evaluating ice hazards.Its efficient moni-toring and accurate identification are crucial for preventing ice floods.Compared with natural rivers,water convey-ance projects exhibit less variation in boundary and hydrodynamic conditions,as well as superior water quality.This makes it challenging to distinguish between ice and water,leading to greater errors in image-based ice concentration recognition methods.To address this challenge,an intelligent ice concentration recognition algorithm based on a deformable convolutional neural network is proposed.This algorithm incorporates deformable convolutional layers,which can adaptively adjust the sampling positions of convolutional kernels to achieve more accurate capture of com-plex ice and water features in low-contrast scenarios.A floating ice dataset containing 330 images from the Middle Route of South-to-North Water Diversion Project was constructed,and a five-fold cross-validation method was used to optimize the algorithm parameters.Experimental results of 16 typical floating ice images show that the average accuracy(ACC)of ice concentration identification reaches 0.96,while the mean intersection over union(IoU)achieves 0.91.Compared with commonly used ice concentration recognition algorithms such as the Otsu and SVM,the proposed algorithm improves the average ACC by 16％and 10％,and the average IoU by 19％and 9％,respec-tively.The findings of this study provide an alternative method for ice concentration recognition in water conveyance channels.
To address the structural rigidity and peak flow forecast deviations in existing machine learning models for flood forecasting under complex rainfall patterns and delayed response conditions,this study proposes a machine learning model for flood forecasting integrated with a process-enhanced mechanism(ML-P-EF).This method intro-duces three types of structural features:Runoff Process Vectorization,Dynamic Lag Encoding,and Event-Driven Features,which process the input data for flood forecasting from the aspects of process structure,time-lag response,and event attributes,respectively.Using four typical watersheds in the middle reaches of the Yellow River as valida-tion cases,and based on three fundamental model structures(Long Short-Term Memory(LSTM),Artificial Neural Network(ANN),and Transformer),we constructed three process-enhanced models and three full-structure-enhanced models to conduct flood forecasting with lead times of lh,3h,and 6h.The results demonstrate that under the 6h lead time condition,the ML-P-EF model improved the average Nash-Sutcliffe Efficiency(NSE)from the basic model ML's 0.212 and the process-enhanced model ML-P's 0.749 to 0.937,reduced the Root Mean Square Error(RMSE)by approximately 62.77％,and decreased the peak flow error by an average of 56.74％.Taking the Daning Station as an example,the NSE of the LSTM-structured model increased from 0.165(basic model)to 0.775(process-enhanced model)and further to 0.982(full-structure-enhanced model),while the RMSE decreased from 46.11 m3/s to 23.96 m3/s and then to 6.75 m3/s,and the peak flow error changed from-68.41％to-41.48％and finally to+3.32％.The ML-P-EF full-structure-enhanced model developed in this study significantly outperforms the basic model in peak response,temporal fitting,and error control,particularly demonstrating stronger generalization capability and stability under the 6h lead time condition.The research findings provide a new structural awareness modeling approach for watershed flood forecasting.
To maximize the revenue of a hydropower station in the uncertain day-ahead market,this paper proposes a bi-level model to formulate the stepped bidding curve for each hydropower unit,ensuring compliance with opera-tional constraints.The model features a nested structure consisting of an outer model for station-level scheduling and an inner model for bidding optimization.The outer model utilizes dynamic programming to determine the short-term optimal scheduling for the hydropower station,considering hydraulic constraints and integrating the expected revenue feedback from the inner model.The inner model first addresses electricity price uncertainty using multiple electricity price scenarios.Subsequently,based on an optimal unit commitment,it develops a unit bidding strategy to generate these curves,taking into account unit-specific constraints and bilateral contract obligations.A genetic algorithm is employed to optimize the bidding curves,with the station's expected revenue serving as the fitness value.The results of a case study demonstrate that this model increases the expected revenue of a hydropower station by 2.4％compared with the current conservative approaches,providing a valuable decision-support tool for market participation.
Air entrained into the pump intake system by pump suction can generate a locally coherent air-core vortex connecting the free surface to the pipe intake.The air-core vortex is one primary adverse factor affecting the opera-tional efficiency,safety and stability of pumping stations.This paper employs large-eddy simulation coupled with the CLSVOF interface-tracking technique to numerically investigate the air-core vortex phenomena at vertical hydraulic intakes.The interaction between vortex structures and turbulent flow fields during the air-core vortex formation was analyzed using the turbulent kinetic energy transport equation and proper orthogonal decomposition(POD)method.The results indicate that the evolution of the air-core vortex in vertical hydraulic intakes can be categorized into three distinct stages:V-shaped surface swirl,surface dimple,and coherent air-core vortex.The turbulent kinetic energy distribution exhibits a strong correlation with vortex dynamics.Analyses of the turbulent kinetic energy transport equa-tion reveal that the interaction between the streamwise and spanwise components of fluctuation velocities and their corresponding velocity gradients drives local turbulent kinetic energy production,while turbulent kinetic energy diffu-sion provides an essential energy transport pathway for the evolution of random vortices into an air-core vortex.The POD analyses of the fluctuation velocity and vorticity fields reveal that the first two modes of the fluctuation fields exhibit a pair of vortices with opposite rotating directions,which causes the strength of the air-core vortex to increase on one side while decreasing on the other,thereby inducing a lateral displacement of the vortex core.The instability characteristics and mechanisms of the air-core vortex revealed in this paper enrich the theoretical understanding of such vortices and provide a basis for vortex control measures and the optimal design of pumping station intake systems.
The evaluation of the effect of rural drinking water safety policy is an important issue in the field of rural revitalization,water security and public health.In order to accurately identify the health effects of water supply proj-ects and related policy implementation,this paper takes Xihaigu region of Ningxia as a case study,focusing on the implementation of rural-urban drinking water safety projects in central and southern Ningxia and the benefit of the application of"Internet+rural-urban water supply"Pengyang model.The difference-in-differences(DID)model was used for quantitative evaluation,and the experimental group,control group and analysis period are carefully selected.Based on the county-level panel data of Ningxia from 2013 to 2019,the average mortality rate of the region was selected as the explanatory variable.A model framework including multi-dimensional control variables such as population,economy,environment,education and medical care was constructed,and the robustness of the results of the benchmark regression scheme was verified by dynamic effect test,placebo test and other methods.The empirical research shows that the rural-urban water supply project and related policy intervention brought about a significant decrease in the mortality rate of residents in the beneficiary counties(districts)by 0.75％o on average compared with other counties(districts)in Xihaigu,and the policy effect can positively influence outcomes for 4-5 years.The contri-bution of rural-urban water supply policy intervention to the reduction of mortality in Xihaigu region is only slightly smaller than the effects of per capita local public financial expenditure and the proportion of Han population,and greater than the effects of other variables such as the number of practicing physicians per thousand population and the primary school graduation rate.It was identified as a key driver significantly affecting residents' health.The evalua-tion results provide an empirical basis for the analysis of the effects of similar water supply policies,and provide strong support for scientific decision-making on the high-quality development of rural water supply.
The main channel storage capacity is directly related to the flood control safety and sediment transport capacity of the Lower Yellow River.Affected by channel sedimentation changes,the specific variation patterns of the main channel storage capacity in the Lower Yellow River over the past 50 years remains unclear.This study is based on annual post-flood cross-sectional data of 91 major sections from Tiexie to Lijin in the lower Yellow River from 1970 to 2020.Using the frustum method,the main channel storage capacity of each river section was calculated,revealing its spatiotemporal evolution patterns.A delayed response simulation method for main channel storage capac-ity was established,and the impact of the Xiaolangdi Reservoir on the main channel storage capacity was evaluated.The results show:(1)The main channel storage capacity and storage capacity per unit length exhibited a trend of"slight increase(1970-1985)-significant decrease(1985-2000)-significant increase(2000-2020)."Spatially,the wandering reach had the largest main channel storage capacity with notable fluctuations.However,the proportional distribution among all sections remained relatively stable,with average proportions of 55％,19％,and 26％for the wandering,transitional,and meandering sections,respectively.After the operation of the Xiaolangdi Reservoir,the main channel storage capacity increased significantly,from 1.039 billion m3 in 2000 to 2.747 billion m3 in 2020,with an average annual growth rate of 5％.(2)The delayed response model,based on annual average flow and sedi-ment concentration,effectively simulated the evolution of main channel storage capacity(R2=0.8480-0.9325,MAPE=3.72％-10.02％).The time to reach quasi-equilibrium for the main channel storage capacity in the lower Yellow River ranged from 7.6 to 13.2 years.(3)With the Xiaolangdi Reservoir in operation,the main channel stor-age capacity increased by 1.588 billion m3 compared with that in 1999,whereas without Xiaolangdi,it would have increased by only 0.509 billion m3,marking a difference of 1.079 billion m3.This indicates that the main channel expansion is the combined effect of soil and water conservation and sediment trapping by the Xiaolangdi Reservoir.Soil and water conservation contributes approximately 30％to the increase in the main channel storage capacity of the lower reaches,while the Xiaolangdi Reservoir accounts for about 70％.The research results help to reveal the mecha-nism behind changes in main channel storage capacity and provide support for maintaining the storage capacity and achieving the long-term stability of the Yellow River.
Due to the pressure exerted by the self-weight of support slurry,the cut-off wall concrete would cure under high curing pressure conditions.However,there is still a lack of experimental verification on how high pressure environments affect the mechanical performance and microstructure of concrete.This research investigated the evolu-tion of mechanical performance and micro-structure of concrete hardened under high curing pressure via a series of multi-scale tests.The results showed that under the high curing pressure,the initial loading stage of the specimens was shortened,and the elastic phase of the stress-strain curve became steeper.The compressive strength and elastic modulus increased by 35.60％and 44.51％,respectively,indicating an overall enhancement in mechanical perfor-mance.In micro-scale,sample cured in pressure exhibited a denser microstructure with highly developed solid skel-eton than that cured in ambient condition,which has inhibited the growth of crack.The pore size distribution within the pressurized specimens performed a refinement trend from larger to smaller pores,indicating an effective densifica-tion of the pore structure.The content of bound water and the degree of hydration were significantly increased under curing pressure,which facilitated the dissolution of cement grains and promoted the precipitation of high-density hydration products.Furthermore,high curing pressure enhanced the polymerization of silicate chains within the hydration products,strengthened the bonding between interlayer water and hydration products,and promotes the development of a highly connected silicate tetrahedral network structure.
To overcome the limitations of existing research in effectively addressing the high-dimensional and nonlin-ear hydraulic processes in open-channel sections of water transfer projects under water diversion disturbances during the icing period,and the high reliance on manual experience in actual operations,this study takes the Wangnou-ruwugou section of the Jiaodong Water Transfer Project as a case study to investigate real-time intelligent hydraulic regulation of gate-pump groups in open channels.Through one-dimensional hydrodynamic simulations of gate-pump groups,step disturbances were applied to the diversion flows of canal pools to reveal the coupling mechanism between hydraulic processes in open channels and the operational responses of gate-pump groups under various diver-sion disturbances,and thus determine safety thresholds for water diversion disturbances during the icing period.On this basis,a real-time intelligent regulation model for gate-pump groups in open channels was developed by coupling a hydraulic model with a deep reinforcement learning algorithm,which excels at handling nonlinear,high-dimensional problems and requires minimal modeling data.The robustness of the proposed model was validated under various operational conditions.Application results demonstrate that the derived scheduling strategy effectively raises and stabilizes the water level at control sections to the target ice-period regulation level while ensuring operational safety and reducing the gate adjustment frequency.
To address the limitations of traditional hydrodynamic analysis methods in large-scale water diversion proj-ects—specifically long computation cycles,high computational costs,and the inability to capture dynamic character-istics in real-time—this paper proposes a digital twin flow field state prediction method driven by a Reduced Order Model(ROM)within the Simulink environment.This study focuses on axial-flow pump units in low-head pump sta-tions.By constructing physical field ROMs and establishing a virtual-real dynamic interaction mechanism,the pro-posed method achieves millisecond-level prediction of flow field states under complex operating conditions.Further-more,a physical model test rig was constructed to conduct systematic verification.The results indicate that,as evalu-ated by the RRMSE and LOOCV methods,the average relative errors of the static pressure,velocity,and total pres-sure models established via Singular Value Decomposition(SVD)are controlled within 5％,while the average rela-tive error for turbulence intensity is 8.5％.These findings demonstrate the model's generalization capability for unknown operating conditions.Concurrently,while maintaining simulation accuracy,the average simulation time per operating point is approximately 0.1 seconds.This significant improvement in computational efficiency provides robust technical support for efficient modeling and real-time decision-making in smart water conservancy systems.
