An alternative approach to the operation of multinational reservoir systems: Application to the Amistad &; Falcon system (Lower Rio Grande/Río Bravo)

An optimization approach for the operation of international multi-reservoir systems is presented. The approach uses Stochastic Dynamic Programming (SDP) algorithms – both steady-state and real-time – to develop two models. In the first model, the reservoirs and flows of the system are aggregated to yield an equivalent reservoir, and the obtained operating policies are disaggregated using a non-linear optimization procedure for each reservoir and for each nation's water balance. In the second model a multi-reservoir approach is applied, disaggregating the releases for each country's water share in each reservoir. The non-linear disaggregation algorithm uses SDP-derived operating policies as boundary conditions for a local time-step optimization. Finally, the performance of the different approaches and methods is compared. These models are applied to the Amistad-Falcon International Reservoir System as part of a binational dynamic modeling effort to develop a decision support system tool for a better management of the water resources in the Lower Rio Grande Basin, currently enduring a severe drought.     

Water Supply Reservoir Operation by Combined Genetic Algorithm – Linear Programming (GA-LP) Approach

Multi-reservoir operation planning is a complex task involving many variables, objectives, and decisions. This paper applies a hybrid method using genetic algorithm (GA) and linear programming (LP) developed by the authors to determine operational decisions for a reservoir system over the optimization period. This method identifies part of the decision variables called cost reduction factors (CRFs) by GA and operational variables by LP. CRFs are introduced into the formulation to discourage reservoir depletion in the initial stages of the planning period. These factors are useful parameters that can be employed to determine operational decisions such as optimal releases and imports, in response to future inflow predictions. A part of the Roadford Water Supply System, UK, is used to demonstrate the performance of the GA-LP method in comparison to the RELAX algorithm. The proposed approach obtains comparable results ensuring non zero final storages in the larger reservoirs of the Roadford Hydrosystem. It shows potential for generating operating policy in the form of hegging rules without a priori imposition of their form.     

逐次逼近随机动态规划及库群优化调度

随机动态规划（SDP）在水库水电站长期优化调度中有着较为广泛的应用,它的最大缺陷就是用于库群优化调度时的“维数灾”问题,逐次逼近胡机动态规划可有效克服这一问题。该方法采用逐次迭代逼近的思想,每次仅对一个水库采用SDP求解,并假设其它水库的蓄水过程已确定为多年平均蓄水过程,对某水库进行SDP求解后,通过多年历史径流过程的模拟调度可对各水库的多年平均蓄水过程进行更新,最后以福建省闽江流域水电系统为例进行了实例应用。     

Optimization of Multi-Reservoir Operating Rules for a Water Supply System

To obtain the optimal releases of the multi-reservoir system, two sets of joint operating rules (JOR-I and JOR-II) are presented based on the aggregation-disaggregation approach and multi-reservoir approach respectively. In JOR-I, all reservoirs are aggregated to an equivalent reservoir, the operating rules of which, the release rule of the system is optimized following operating rule curves coupled with hedging rules. Then the system release is disaggregated into each reservoir according to water supply priorities and the dynamic demand partition approach. In JOR-II, a two-stage demand partition approach is applied to allocate the different demand priorities to determine the release from each reservoir. To assess the reliability and effectiveness of the joint operating rules, the proposed rules are applied to a multi-reservoir system in Liaoning province of China. Results demonstrate that JOR-I is suitable for high-dimensional multi-reservoir operation problems with large-scale inflow data, while JOR-II is suitable for low-dimensional multi-reservoir operation problems with small-scale inflow data, and JOR-II performs better than JOR-I but requires more computation time. The research provides guidelines for the management of multi-reservoir system.     

Multivariate Bayesian Regression Approach to Forecast Releases from a System of Multiple Reservoirs

This research presents a model that simultaneously forecasts required water releases 1 and 2 days ahead from two reservoirs that are in series. In practice, multiple reservoir system operation is a difficult process that involves many decisions for real-time water resources management. The operator of the reservoirs has to release water from more than one reservoir taking into consideration different water requirements (irrigation, environmental issues, hydropower, recreation, etc.) in a timely manner. A model that forecasts the required real-time releases in advance from a multiple reservoir system could be an important tool to allow the operator of the reservoir system to make better-informed decisions for releases needed downstream. The model is developed in the form of a multivariate relevance vector machine (MVRVM) that is based on a sparse Bayesian regression model approach. With this Bayesian approach, a predictive confidence interval is obtained from the model that captures the uncertainty of both the model and the data. The model is applied to the multiple reservoir system located in the Lower Sevier River Basin near Delta, Utah. The results show that the model learns the input–output patterns with high accuracy. Computing multiple-time-ahead predictions in real-time would require a model which guarantees not only good prediction accuracy but also robustness with respect to future changes in the nature of the inputs data. A bootstrap analysis is used to guarantee good generalization ability and robustness of the MVRVM. Test results demonstrate good performance of predictions and statistics that indicate robust model generalization abilities. The MVRVM is compared in terms of performance and robustness with another multiple output model such as Artificial Neural Network (ANN).     

Reliability Improved Stochastic Dynamic Programming for Reservoir Operation Optimization

Stochastic Dynamic Programming (SDP) is widely used in reservoir operation problems. Besides its advantages, a few drawbacks have leaded many studies to improve its structure. Handling the infeasible conditions and curse of dimensionality are two major challenges in this method. The main goal of this paper is proposing a new method to avoid infeasible conditions and enhance the solution efficiency with new discretization procedure. For this purpose, an optimization module is incorporated into regular SDP structure, so that, near optimal values of state variables are determined based on the available constraints. The new method (RISDP) employs reliability concept to maximize the reservoir releases to satisfy the downstream demands. Applying the proposed technique improves the reservoir operating policies compared to regular SDP policies with the same assumptions of discretization. Simulation of reservoir operation in a real case study indicates about 15% improvement in objective function value and elimination of infeasible conditions by using RISDP operating policies.     

Single Reservoir Operating Policies Using Genetic Algorithm

To obtain optimal operating rules for storage reservoirs, large numbers of simulation and optimization models have been developed over the past several decades, which vary significantly in their mechanisms and applications. As every model has its own limitations, the selection of appropriate model for derivation of reservoir operating rule curves is difficult and most often there is a scope for further improvement as the model selection depends on data available. Hence, evaluation and modifications related to the reservoir operation remain classical. In the present study a Genetic Algorithm model has been developed and applied to Pechiparai reservoir in Tamil Nadu, India to derive the optimal operational strategies. The objective function is set to minimize the annual sum of squared deviation form desired irrigation release and desired storage volume. The decision variables are release for irrigation and other demands (industrial and municipal demands), from the reservoir. Since the rule curves are derived through random search it is found that the releases are same as that of demand requirements. Hence based on the present case study it is concluded that GA model could perform better if applied in real world operation of the reservoir.     

Comparison of Policies Derived from Stochastic Dynamic Programming and Genetic Algorithm Models

A comprehensive Genetic Algorithm (GA) model has been developed and applied to derive optimal operational strategies of a multi-purpose reservoir, namely Perunchani Reservoir, in Kodaiyar Basin in Tamil Nadu, India. Most of the water resources problem involves uncertainty, in order to see that the GA model takes care of uncertainty in the input variable, the result of the GA model is compared with the performance of a detailed Stochastic Dynamic Programming (SDP) model. The SDP models are well established and proved that it takes care of uncertainty in-terms of either implicit or explicit approach. In the present study, the objective function of the models is set to minimize the annual sum of squared deviation from desired target release and desired storage volume. In the SDP model the optimal policies are derived by varying the state variables from 3 to 9 representative class intervals, and then the cases are evaluated for their performance using a simulation model for longer length of inflow data, generated using a Thomas–Fiering model. From the performance of the SDP model policies, it is found that the system encountered irrigation deficit, whereas GA model satisfied the demand to a greater extent. The sensitivity analysis of the GA model in selecting optimal population, optimal crossover probability and the optimal number of generations showed the values of 150, 0.76 and 175 respectively. On comparing the performance of SDP model policy with GA model, it is found that GA model has resulted in a lesser irrigation deficit. Thus based on the present case study, it may be concluded that the GA model performs better than the SDP model.     

Optimization of Exclusive Release Policies for Hydropower Reservoir Operation by Using Genetic Algorithm

The efficient utilization of hydropower resources play an important role in the economic sector of power systems, where the hydroelectric plants constitute a significant portion of the installed capacity. Determination of daily optimal hydroelectric generation scheduling is a crucial task in water resource management. By utilizing the limited water resource, the purpose of hydroelectric generation scheduling is to specify the amount of water releases from a reservoir in order to produce maximum power, while the various physical and operational constraints are satisfied. Hence, new forms of release policies namely, BSOPHP, CSOPHP, and SHPHP are proposed and tested in this research. These policies could only use in hydropower reservoir systems. Meanwhile, to determine the optimal operation of each policy, real coded genetic algorithm is applied as an optimization technique and maximizing the total power generation over the operational periods is chosen as an objective function. The developed models have been applied to the Cameron Highland hydropower system, Malaysia. The results declared that by using optimal release policies, the output of power generation is increased, while these policies also increase the stability of reservoir system. In order to compare the efficiency of these policies, some reservoir performance indices such as reliability, resilience, vulnerability, and sustainability are used. The results demonstrated that SHPHP policy had the highest performance among the tested release policies.     

Integrated Reservoir Management System for Adaptation to Climate Change: The Nakdong River Basin in Korea

This study begins with the premise that current reservoir management systems do not take into account the potential effects of climate change on optimal performance. This study suggests an approach in which multi-purpose reservoirs can adapt to climate change using optimal rule curves developed by an integrated water resources management system. The system has three modules: the Weather Generator model, the Hydrological Model, and the Differential Evolution Optimization Model. Two general circulation models (GCMs) are selected as examples of both dry and wet conditions to generate future climate scenarios. This study is using the Nakdong River basin in Korea as a case study, where water supply is provided from the reservoir system. Three different climate change conditions (historic, wet and dry) are investigated through the compilation of six 60 years long scenarios. The optimal rule curves for three multi-purpose reservoirs in the basin are developed for each scenario. The results indicate that although the rule curve for large-size reservoir is less sensitive to climate change, medium or small-size reservoirs are very sensitive to those changes. We further conclude that the large reservoir should be used to release more water, while small or medium-size reservoirs should store inflow to mitigate severe drought damages in the basin.     

Improved Reservoir Operation Using Hybrid Genetic Algorithm and Neurofuzzy Computing

A hybrid genetic and neurofuzzy computing algorithm was developed to enhance efficiency of water management for a multipurpose reservoir system. The genetic algorithm was applied to search for the optimal input combination of a neurofuzzy system. The optimal model structure is modified using the selection index (SI) criterion expressed as the weighted combination of normalized values of root mean square error (RMSE) and maximum absolute percentage of error (MAPE). The hybrid learning algorithm combines the gradient descent and the least-square methods to train the genetic-based neurofuzzy network by adjusting the parameters of the neurofuzzy system. The applicability of this modeling approach is demonstrated through an operational study of the Pasak Jolasid Reservoir in Pasak River Basin, Thailand. The optimal reservoir releases are determined based on the reservoir inflow, storage stage, sideflow, diversion flow from the adjoining basin, and the water demand. Reliability, vulnerability and resiliency are used as indicators to evaluate the model performance in meeting objectives of satisfying water demand and maximizing flood prevention. Results of the performance evaluation indicate that the releases predicted by the genetic-based neurofuzzy model gave higher reliability for water supply and flood protection compared to the actual operation, the releases based on simulation following the current rule curve, and the predicted releases based on other approaches such as the fuzzy rule-based model and the neurofuzzy model. Also the predicted releases based on the newly developed approach result in the lowest amount of deficit and spill indicating that the developed modeling approach would assist in improved operation of Pasak Jolasid Reservoir.     

Climate Change Water Resources Planning Impacts Incorporating Reservoir Surface Net Evaporation Fluxes: A Case Study

The majority of published studies on the impacts of climate change on reservoired water resources systems have concentrated on the influence of the climate- change-modified inflow series. However, for reservoirs the direct net evaporation (i.e. evaporation less rainfall) fluxes on the reservoir surface are also affected by climate change and, depending on the magnitude of the change, could have significant effects on the assessed impacts. In this study, we have performed reservoir storage-yield-reliability planning analyses on two multiple reservoir systems, one in England and the other in Iran, to investigate the possible effects of reservoir surface net evaporation flux for both baseline and climate-change conditions. The results showed that, under baseline conditions, consideration of net evaporation will require lower storages for the English systems and higher storages for the Iranian systems. The practical significance of this is that English systems analysed without consideration of surface fluxes represent an over-design which can provide a buffer against future shortages, whereas the under-design caused by ignoring surface fluxes in the Iranian systems will exacerbate the problem of such shortages. Perturbing the baseline inflow and climatological time-series data using a number of recently published climate-change scenarios produced different impacts at high and low yields for both systems. Possible explanations are offered for these impacts and suggestions are made for further studies.     

Multireservoir System Optimization using Fuzzy Mathematical Programming

For a multireservoir system, where the number of reservoirs islarge, the conventional modelling by classical stochastic dynamicprogramming (SDP) presents difficulty, due to the curse ofdimensionality inherent in the model solution. It takes a longtime to obtain a steady state policy and also it requires largeamount of computer storage space, which form drawbacks inapplication. An attempt is made to explore the concept of fuzzysets to provide a viable alternative in this context. Theapplication of fuzzy set theory to water resources systems isillustrated through the formulation of a fuzzy mathematicalprogramming model to a multireservoir system with a number ofupstream parallel reservoirs, and one downstream reservoir. Thestudy is aimed to minimize the sum of deviations of the irrigationwithdrawals from their target demands, on a monthly basis, over ayear. Uncertainty in reservoir inflows is considered by treatingthem as fuzzy sets. The model considers deterministic irrigationdemands. The model is applied to a three reservoir system in theUpper Cauvery River basin, South India. The model clearlydemonstrates that, use of fuzzy linear programming inmultireservoir system optimization presents a potentialalternative to get the steady state solution with a lot lesseffort than classical stochastic dynamic programming.     

Three Level Rule Curve for Optimum Operation of a Multipurpose Reservoir using Genetic Algorithms

Finding optimal policies for real-life reservoir systems operation (RSO) is a challenging task as the available analytical methods cannot handle the arbitrary functions of the problem. Most of the methods employed are numerical or iterative type and are computer dependent. Since the computer resources in terms of memory and CPU time are limited efficient algorithms are necessary to deal with the RSO problems. In this paper we present a Genetic Algorithms (GA) optimized rule curve (RC) model for monthly operation of a multipurpose reservoir which maximizes hydropower produced while meeting the irrigation demands with a given reliability. Instead of the usual single target storage for each period the proposed model considers three sets of target storages, namely dry, normal, and wet storages, based on the beginning of the period storage level. The reservoir considered is Bhadra Multipurpose Reservoir, in the state of Karnataka, India, which supplies water to irrigation fields through two canals while generating hydropower with turbines installed at each of the canal heads and at the river bed. Optimization ability and robustness of GA-RC approach are ascertained through simulation with a different inflow sequence for which global optimum is computed using Dynamic Programming. Further, a 15 year real-time simulation of the reservoir using historical inflows and demands showed significant improvement in the benefit, i.e. power produced, without compromising on the irrigation demands throughout the operation period.

     

Predictive Temporal Data-Mining Approach for Evolving Knowledge Based Reservoir Operation Rules

The persistent problem in reservoir operation is that the derived optimal releases fail to incorporate the decision maker or reservoir operators’ knowledge into reservoir operation models. The reservoir operators’ knowledge is specific to that particular reservoir and incorporating such an experienced knowledge will help to derive field reality based operation rules. The available historical reservoir operation databases are the representative samples of reservoir operators’ knowledge or experience. Thus, an attempt has been made that deals with the development of a methodological framework to recover or explore the historical reservoir operation database to derive the reservoir operators’ knowledge as operational rules. The developed methodological framework utilizes the strength and capability of recently developed predictive datamining algorithms to recover the knowledge from large historical database. Predictive data-mining algorithms such as a) classifier: Artificial Neural Network (ANN), and b) regression: Support Vector Regression (SVR) have been used for single reservoir operation data-mining (SROD) modelling framework to explore the temporal dependence between different variables of reservoir operation. The rules of operation or knowledge learned from the training database have been used as guiding rules for predicting the future reservoir operators’ decision on operating the reservoir for the given condition on the inflow, initial storage, and demand requirements. The developed SROD model was found to be efficient in exploring the hidden relationships that exist in a single reservoir system.     

Modeling Multi-objective Pareto-optimal Reservoir Operation Policies Using State-of-the-art Modeling Techniques

A novel challenge faced by water scientists and water managers today is the efficient management of the available water resources for meeting crucial demands such as drinking water supply, irrigation and hydro-power generation. Optimal operation of reservoirs is of paramount importance for better management of scarce water resources under competing multiple demands such as irrigation, water supply etc., with decreasing reliability of these systems under climate change. This study compares six different state-of-the-art modeling techniques namely; Deterministic Dynamic Programming (DDP), Stochastic Dynamic Programming (SDP), Implicit Stochastic Optimization (ISO), Fitted Q-Iteration (FQI), Sampling Stochastic Dynamic Programming (SSDP), and Model Predictive Control (MPC), in developing pareto-optimal reservoir operation solutions considering two competing operational objectives of irrigation and flood control for the Pong reservoir located in Beas River, India. Set of pareto-optimal (approximate) solutions were derived using the above-mentioned six methods based on different convex combinations of the two objectives and finally the performances of the resulting sets of pareto-optimal solutions were compared. Additionally, key reservoir performance indices including resilience, reliability, vulnerability and sustainability were estimated to study the performance of the current operation of the reservoir. Modeling results indicate that the optimal-operational solution developed by DDP attains the best performance followed by the MPC and FQI. The performance of the Pong reservoir operation assessed by comparing different performance indices suggests that there is high vulnerability (~?0.65) and low resilience (~?0.10) in current operations and the development of pareto-optimal operation solutions using multiple state-of-the-art modeling techniques might be crucial for making better reservoir operation decisions.

     

Optimal Pre-storm Flood Hedging Releases for a Single Reservoir

Flood hedging reservoir operation is when a pre-storm release creates a small flood downstream to reduce the likelihood of a more damaging but uncertain larger flood in the future. Such pre-storm releases before a flood can increase reservoir storage capacity available to capture more severe flood flows, but also can immediately increase downstream flood damage and reduce stored water supply. This study develops an optimization model for pre-storm flood hedging releases and examines some necessary theoretical conditions for optimality, considering hydrologic uncertainty from flood forecasts and engineering uncertainty from flood failures. Theoretically, the ideal optimality condition for pre-storm flood hedging releases is where the current marginal damage from the hedging release equals the future expected marginal damage from storm releases. Additional water supply losses due to pre-storm releases tend to reduce pre-storm flood hedging releases. The overall flood damage cost to be minimized must be a convex function of pre-storm hedging releases for flood hedging to be optimal. Such convexity is determined by the overall flood risk together with the probability distribution of storm forecasts. Increasing the convexity of the failure probability function can induce more pre-storm hedging release. Categorized by flood risk likelihood downstream, forecasted storms that are large, but not yet overwhelming flood management systems, drive optimal flood hedging operation. A wide range of near-optimal hedging releases is observed in numerical examples, providing options for more rational water resources management decisions.     

Development of Operational Policy for a Multi-reservoir System in India using Genetic Algorithm

A genetic algorithm (GA) and a backward moving stochastic dynamic programming (SDP) model has been developed for derivation of operational policies for a multi-reservoir system in Kodaiyar River Basin, Tamil Nadu, India. The model was developed with the objective of minimizing the annual sum of squared deviation of desired target releases. The total number of population, crossover probability and number of generations of the GA model was optimized using sensitivity analysis, and penalty function method was used to handle the constraints. The policies developed using the SDP model was evaluated using a simulation model with longer length of inflow data generated using monthly time stepped Thomas–Fiering model. The performance of the developed policies were evaluated using the performance criteria namely, the monthly frequency of irrigation deficit (MFID), Monthly average irrigation deficit (MAID), Percentage monthly irrigation deficit (PMID), Annual frequency of irrigation deficit (AFID), Annual average irrigation deficit (AAID), and Percentage annual irrigation deficit (PAID). Based on the performance, it was concluded that the robostic, probabilistic, random search GA resulted in better optimal operating policies for a multi-reservoir system than the SDP models.     

An Effective Approach to Long-Term Optimal Operation of Large-Scale Reservoir Systems: Case Study of the Three Gorges System

A new approach for optimization of long-term operation of large-scale reservoirs is presented, incorporating Incremental Dynamic Programming (IDP) and Genetic algorithm (GA) . The immense storage capacity of the large scale reservoirs enlarges feasible region of the operational decision variables, which leads to invalidation of traditional random heuristic optimization algorithms. Besides, long term raised problem dimension, which has a negative impact on reservoir operational optimization because of its non-linearity and non-convexity. The hybrid IDP-GA approach proposed exploits the validity of IDP for high dimensional problem with large feasible domain by narrowing the search space with iterations, and also takes the advantage of the efficiency of GA in solving highly non-linear, non-convex problems. IDP is firstly used to narrow down the search space with discrete d variables. Within the sub search space provided by IDP, GA searches the optimal operation scheme with continuous variables to improve the optimization precision. This hybrid IDP-GA approach was applied to daily optimization of the Three Gorges Project-Gezhouba cascaded hydropower system for annual evaluation from the year of 2004 to 2008. Contrast test shows hybrid IDP-GA approach outperforms both the univocal IDP and the classical GA. Another sub search space determined by actual operational data is also compared, and the hybrid IDP-GA approach saves about 10 times of computing resources to obtain similar increments. It is shown that the hybrid IDP GA approach would be a promising approach to dealing with long-term optimization problems of large-scale reservoirs.