Journal of Water Resources Planning and Management

In response to the destructive floods of 1983, 1986, 1995, and 1997, the U.S. Army Corps of Engineers and the Reclamation Board of the State of California are partnering a study to investigate flood damage reduction and ecosystem restoration opportunities in the Sacramento and San Joaquin River Basins, California. This paper continues presentation of Comprehensive Study methods and details the construction and application of basin-wide reservoir models in support of ongoing system analyses. Discussion emphasizes model development, simulation results with a focus on headwater and major terminal reservoirs, and potential improvements to the flood damage reduction system. Conclusions are drawn regarding the influence of reservoirs in flood hydrology, use of HEC-5 software in flood analyses, and the role of this effort in the Comprehensive Study.

In dynamic real-time control (RTC) of regional water systems, a multicriteria optimization problem has to be solved to determine the optimal control strategy. Nonlinear and/or dynamic programming based on simulation models can be used to find the solution, an approach being used in the Aquarius decision support system (DSS) developed in The Netherlands. However, the computation time required for complex models is often prohibitively long, and therefore such a model cannot be applied in RTC of water systems. In this study, Aquarius DSS is chosen as a reference model for building a controller using machine learning techniques such as artificial neural networks (ANN) and reinforcement learning (RL), where RL is used to decrease the error of the ANN-based component. The model was tested with complex water systems in The Netherlands, and very good results were obtained. The general conclusion is that a controller, which has learned to replicate the optimal control strategy, can be used in RTC operations.

Sampling design (SD) for water distribution systems (WDS) is undoubtedly an important issue, and has been addressed in the past by a number of scientists and practitioners. The aim of the SD methodology developed here is to find a set of optimal network locations at which to place measurement devices. Optimal locations are determined with the aim of collecting data that will be used later on in the calibration of the analyzed WDS hydraulic model. First, existing calibration and SD approaches in the case of WDS are reviewed. After that, SD is formulated as a two-objective optimization problem. The objectives are maximization of the calibrated model accuracy by minimization of the relevant uncertainties, and minimization of total SD costs. The optimal SD problem is then solved using a multiobjective genetic algorithm based on Pareto ranking, niching, and restricted mating. The methodology developed is applied and verified on a case study. At the end, a summary is made and relevant conclusions are drawn.

The quality of leak detection and quantification and calibration for friction coefficients in pipelines and networks by the inverse transient method are dependent on the quantity and location of data measurement sites. This paper presents an approach for determining the configuration of measurement sites that produces optimal results. Three performance indicators, two that are based on A- and D-optimality criteria and one that is based on the sensitivities of the heads with respect to the parameters, show which configurations are superior. These are illustrated by two case studies, the first of which is a small pipe network in which all configurations are considered directly (fully enumerable) and the second is a larger pipe network in which statistics are drawn from a sampling of configurations. For the large network, a genetic algorithm, with a new crossover operator, performs a search of possible measurement site configurations to determine the optimal measurement locations. The number of sites as well as time length of data at each site are also considered.

The specified range of free chlorine residual (between minimum and maximum) in water distribution systems needs to be maintained to avoid deterioration of the microbial quality of water, control taste and/or odor problems, and hinder formation of carcinogenic disinfection by-products. Multiple water quality sources for providing chlorine input are needed to maintain the chlorine residuals within a specified range throughout the distribution system. The determination of source dosage (i.e., chlorine concentrations/chlorine mass rates) at water quality sources to satisfy the above objective under dynamic conditions is a complex process. A nonlinear optimization problem is formulated to determine the chlorine dosage at the water quality sources subjected to minimum and maximum constraints on chlorine concentrations at all monitoring nodes. A genetic algorithm (GA) approach in which decision variables (chlorine dosage) are coded as binary strings is used to solve this highly nonlinear optimization problem, with nonlinearities arising due to set-point sources and non-first-order reactions. Application of the model is illustrated using three sample water distribution systems, and it indicates that the GA is a useful tool for evaluating optimal water quality source chlorine schedules.

Artificial neural networks (ANNs) are applied to estimating nitrate concentrations in a typical Midwestern river, i.e., the Upper Sangamon River in Illinois. Throughout the Midwestern United States, nitrate in raw water has recently become an increasingly important problem. This is due to recent changes in the U.S. EPA nitrate standard and to the increasingly widespread use of chemical fertilizers in agriculture. Back-propagation neural networks (BPNNs) and radial basis function neural networks (RBFNNs) are compared as to their effectiveness in water quality modeling. Training of the RBFNNs is much faster than that of the BPNNs, and yields more robust results. These two types of ANNs are compared to traditional regression and mechanistic water quality modeling, based on overall accuracy and on the frequency of false-negative prediction. The RBFNN achieves the best results of all models in terms of overall accuracy, and both BPNN and RBFNN yield the same false-negative frequency, which is better than that of the traditional models.

This paper presents a hybrid modeling approach for modeling flows within wastewater pipe networks. The approach utilizes the MOUSE model as a deterministic pipe network model and a radial basis function neural network (RBFNN) model as a stochastic error-correction model. Both models utilize rainfall as input, whereas the RBFNN utilizes MOUSE model flow predictions and its errors (residuals, differences) related to that prediction. The MOUSE model output provides an approximation of the hydrodynamic process, whereas the outputs of the trained RBFNN compensate for the output errors (residuals) of the MOUSE model. It is demonstrated that this approach is capable of reducing the model prediction error compared to the deterministic model when applied alone. It is also demonstrated that this approach generates more accurate results than hybrid models with linear stochastic components when compared for direct and iterative forecasts. The results achieved are promising, and the approach developed provides an innovative tool for achieving more accurate outputs.