This study demonstrates application of Deep Deterministic Policy Gradient (DDPG)-based algorithm to provide comprehensive and flexible plans for reservoir operation planning of the multiple reservoir system in the Chao Phraya River Basin (CPYRB), Thailand aiming to mitigate flood and drought risks in the region. The multi-agent-based Deep Reinforcement Learning (DRL) model is accordingly constructed considering 7-D predicted inflow, reservoir water released from adjacent reservoir, downstream flow condition, and changes in reservoir water storage, as state variables. The desired goal is to increase water storage levels in all reservoirs by 10–15% to ensure higher potential in supplying water for crop cultivation over the dry seasons and preventing flood occurrences during wet season. Simulation results from 2009 to 2022 indicate that DRL–DDPG-based algorithm can perform well in solving sequential decision problems for optimal operation of multiple reservoir system to achieve the desired water storage goal. It can offer realistic simulation results of seasonal and annual release schemes and reservoir release ratios among reservoirs in the system compared to actual operation and Fmincon and ANFIS optimizations. Importantly, DRL model demonstrates a significant advantage in view of increasing the long-term water storage levels in all reservoirs as targeted in the modelling process while maintaining the similar and consistent release schemes in the reservoir system. For the multipurpose multiple reservoir system operation, adjusting the dynamic desired goals within multi-agent-based RL model is advisable to attain the specific desired outcomes and address various water scenarios.
Elevated nutrient concentrations are of concern in many agricultural streams. This study observed seasonal variation of inorganic nitrogen concentrations in small drainage streams that receive discharge predominantly from three different types of agricultural practices, swine farms, aquaculture, and paddy fields, in the Tha Chin River Basin in central Thailand. The drainage streams had high inorganic nitrogen concentrations, low dissolved oxygen concentrations (50 m, and the ammonium retention declined by approximately 50%. These results suggest that small agricultural streams experience a high concentration of inorganic nitrogen, similar or even more than larger streams, and that in-channel vegetation plays important role in nitrogen retention. The management of in-channel vegetation will have considerable effects on nitrogen transport.
The modern constraint programming (CP) was adopted to minimize water scarcity and excessive water which are the critical issues in reservoir operation of Bhumibol Dam (BB) to solve consecutive droughts in the Chao Phraya River Basin (CPYRB), Thailand. The situations of the severe droughts have been frequently occurred in a broad area of CPYRB due to an extremely low rainfall leading to unusually low water levels and inflow in major reservoirs. Therefore, multi-objective optimization was conducted to characterize the actual operation and physical reservoir system of BB Dam. Two different CP models with seasonal and yearly constraints were manipulated using MiniZinc programming language and the constraint solver IPOPT to find the optimal daily release scheme from 2000 to 2018 of BB Dam. The potential of downstream flow conditions was also considered to partially supply downstream water demand and store savable water in a reservoir for subsequent use during possible future droughts. The results reveal that CP models can diminish some extent of yearly reservoir release, while daily long-term release scheme conforms well with the actual operation particularly during dry and wet seasons in specific drought years. These mean that amount of reservoir water of approximately 47.12–103.83 MCM/year can be saved and stored in reservoir for subsequent use in CPYRB when CP models are deployed. This results in a reservoir storage increase of 7.10–7.94% at the end of the wet season for seasonal and yearly CP models, respectively. When potential side flow is considered, the increase climbs up to 10.49%. This envisages the higher possibility of supplying reservoir water for agricultural water needs over the dry season in the Greater Chao Phraya Irrigation Scheme. As the potential hydraulic head is subject to increased reservoir water storage, therefore, the increase in hydropower production is definitely found of ranging from + 6.10% to + 13.79% by these two sorts of CP models. In addition, handling with huge volumes of flood and drought by CP models is well managed as the specific release constraints during refilled and drawdown periods are used to define the constraint satisfaction problem.
3 July 2024 in Modeling Earth Systems and Environment
Strategic reservoir operation, a primary water management measures, plays a significant role in mitigating floods and droughts. Since the reservoir operation involves making complicated decisions on uncertain hydrological variables driven by climate variability, therefore, constructive tool for decision making like fuzzy logic is essential to optimize reservoir management and ensure water security. This study demonstrated fuzzy logic application to multiple reservoir operation in tropical region like Thailand. A Fuzzy Rule–Based Model (FRBM) exploiting FL approach was developed to control the upstream reservoir operation in the Upper Mun River Basin (UMRB) using the data from 2008 to 2021. Implementing FRBM for UMRB was conducted by identifying two key variables; available water storage and 7–day ahead predicted inflow, as fuzzy inputs. The fuzzy output of the system is the release fraction determined by three operational condition modules; flood, neutral, and drought. For flood module, fuzzy release is primarily determined by the predicted inflow. However, the determination of reservoir release for drought and neutral modules is influenced by the targeted water demand. The results of base case illustrate the capability of FRBM in increasing reservoir storages at the start of dry season by 123.56 MCM/yr in UMRB due to the new daily release schemes generated. This allows supplying water closer to the theoretical agricultural needs and gross irrigation water requirement potentially reducing the risk of water shortfall during consecutive dry years. Whereas, the maximum fuzzy release is constrained corresponding to safe channel capacity of tributaries and Upper Mun river, therefore, downstream flooding is accordingly prevented.
This study demonstrated the predictability of Machine Learning (ML)– and Deep Learning (DL)–based univariate and multivariate predictions of reservoir inflows of Bhumibol (BB) and Sirikit (SK), two major dams in the Chao Phraya River Basin. XGBoost, tree–based ensemble–, and LSTM, deep neural network–based algorithms were selected for development of daily and monthly prediction models. For univariate prediction, the inflows of the BB and SK dams were predicted separately using two individual models. In contrast, for multivariate prediction, a single model was developed to simultaneously predict the inflows of both the BB and SK dams facilitating the integrated decision–making processes. Across all prediction scenarios, ML– and DL–based models demonstrated superior performances in predicting daily reservoir inflows for BB and SK dams compared to monthly predictions, achieving NSE values of 0.86 and 0.77, respectively. Since modeling with LSTM algorithm can effectively handle larger datasets, this enables single multivariate prediction model to predict closer results to those individual univariate models performed by XGBoost and LSTM for BB and SK prediction. XGBoost models mostly outperformed LSTM when tested on the datasets for both daily and monthly univariate predictions. Among all prediction scenarios, underprediction of low reservoir inflows and overprediction of high reservoir inflows by both univariate and multivariate models were consistently existed. Therefore, extracting specific and informative insights from the results of each model type, forecasting horizon, and algorithms used can significantly enhance decision–making support for both real–time reservoir operation and long–term reservoir management planning.
4 May 2026 in Modeling Earth Systems and Environment
This study demonstrated the predictability of Machine Learning (ML)– and Deep Learning (DL)–based univariate and multivariate predictions of reservoir inflows of Bhumibol (BB) and Sirikit (SK), two major dams in the Chao Phraya River Basin. XGBoost, tree–based ensemble–, and LSTM, deep neural network–based algorithms were selected for development of daily and monthly prediction models. For univariate prediction, the inflows of the BB and SK dams were predicted separately using two individual models. In contrast, for multivariate prediction, a single model was developed to simultaneously predict the inflows of both the BB and SK dams facilitating the integrated decision–making processes. Across all prediction scenarios, ML– and DL–based models demonstrated superior performances in predicting daily reservoir inflows for BB and SK dams compared to monthly predictions, achieving NSE values of 0.86 and 0.77, respectively. Since modeling with LSTM algorithm can effectively handle larger datasets, this enables single multivariate prediction model to predict closer results to those individual univariate models performed by XGBoost and LSTM for BB and SK prediction. XGBoost models mostly outperformed LSTM when tested on the datasets for both daily and monthly univariate predictions. Among all prediction scenarios, underprediction of low reservoir inflows and overprediction of high reservoir inflows by both univariate and multivariate models had consistently existed. Therefore, extracting specific and informative insights from the results of each model type, forecasting horizon, and algorithms used can significantly enhance decision–making support for both real–time reservoir operation and long–term reservoir management planning.