Journal of Water Resources Planning and Management

A classifier system for automatic operation of canal gates was developed and tested through simulation modeling. The classifier system manipulates a population of rules that are trained to “learn” appropriate operational responses to unsteady hydraulic conditions. Each rule has one condition and one associated action. The condition and action pair were applied by matching rules to the current hydraulic status, then taking the gate action specified by the matching rule. Sets of gate operational rules that define appropriate responses to different environment situations, represented by hydraulic transients due to changing water deliveries along the canal, were generated through the classifier system. An apportionment-of-credit algorithm was designed by applying a combination of immediate and delayed rewards. Three components of the delayed reward were used to quantify performance in terms of quick stabilization to target water depths in the canal, and small depth fluctuations. A genetic algorithm was applied to inject new rules. In the best cases, the classifier system produced operational rules that stabilized the simulated canal system within 2% of the target levels in 95% of the simulations. Compared to three local automation methods, the classifier system showed the best overall performance in terms of hydraulic stabilization time and matching target water levels.

Optimal groundwater remediation design problems are often complex, nonlinear, and computationally intensive. Genetic algorithms allow solution of more complex nonlinear problems than traditional gradient-based approaches, but they are more computationally intensive. One way to improve performance is through inclusion of local search, creating a hybrid genetic algorithm (HGA). This paper presents a new self-adaptive HGA (SAHGA) and compares its performance to a nonadaptive hybrid genetic algorithm (NAHGA) and the simple genetic algorithm (SGA) on a groundwater remediation problem. Of the two hybrid algorithms, SAHGA is shown to be far more robust than NAHGA, providing fast convergence across a broad range of parameter settings. For the test problem, SAHGA needs 75% fewer function evaluations than SGA, even with an inefficient local search method. These findings demonstrate that SAHGA has substantial promise for enabling solution of larger-scale problems than was previously possible.

In groundwater problems, there always is some uncertainty associated with appropriate values for aquifer parameters. Therefore an optimal remediation strategy identified by assuming a deterministic description of the system may not yield an optimal and feasible design. This work develops a robust genetic algorithm (GA) approach that takes into account the uncertainty of hydraulic conductivity values when determining the best remediation design possible. Within a generation of the robust GA, all designs are evaluated using the same realization of the heterogeneous hydraulic conductivity field, but the realizations vary between GA generations. Ongoing performance of the designs is measured and is used in the GA evolution process. While the robust GA is a multiple realization method, minimal additional computation effort over that of a basic GA is required to identify robust designs. The robust GA is applied to two cases of varying heterogeneity of an example contaminated aquifer remediated by a pump-and-treat system. The goal of the optimization problem is to identify low cost designs that meet constraints on contaminant concentrations, hydraulic heads, and well extraction rates given uncertainty in hydraulic conductivity values. In addition, the basic GA and noisy GA are applied to the same problem for comparison. The robust GA identified designs with lower costs and using fewer objective function evaluations than the noisy GA designs. However, the noisy GA designs generally have lower concentration constraint violation and higher reliability than the robust GA designs, particularly in the more heterogeneous case. As heterogeneity and uncertainty in hydraulic conductivity increases, remediation costs increase as more wells are required to satisfy the remediation goal and reliability decreases. Nevertheless, both the robust GA and noisy GA identified designs that performed better in constraint feasibility and reliability than the solutions found by the basic GA, which assumed deterministic conditions. Additionally, this work shows that assuming a deterministic description of the aquifer—either homogeneous or heterogeneous—can result in significant underdesign and poor remediation performance.

An integrative computational methodology is developed for the management of nonpoint source pollution from watersheds. The associated decision support system is based on an interface between evolutionary algorithms (EAs) and a comprehensive watershed simulation model, and is capable of identifying optimal or near-optimal land use patterns to satisfy objectives. Specifically, a genetic algorithm (GA) is linked with the U.S. Department of Agriculture’s Soil and Water Assessment Tool (SWAT) for single objective evaluations, and a Strength Pareto Evolutionary Algorithm has been integrated with SWAT for multiobjective optimization. The model can be operated at a small spatial scale, such as a farm field, or on a larger watershed scale. A secondary model that also uses a GA is developed for calibration of the simulation model. Sensitivity analysis and parameterization are carried out in a preliminary step to identify model parameters that need to be calibrated. Application to a demonstration watershed located in Southern Illinois reveals the capability of the model in achieving its intended goals. However, the model is found to be computationally demanding as a direct consequence of repeated SWAT simulations during the search for favorable solutions. An artificial neural network (ANN) has been developed to mimic SWAT outputs and ultimately replace it during the search process. Replacement of SWAT by the ANN results in an 84% reduction in computational time required to identify final land use patterns. The ANN model is trained using a hybrid of evolutionary programming (EP) and the back propagation (BP) algorithms. The hybrid algorithm was found to be more effective and efficient than either EP or BP alone. Overall, this study demonstrates the powerful and multifaceted role that EAs and artificial intelligence techniques could play in solving the complex and realistic problems of environmental and water resources systems.

Identifying contaminant sources in groundwater is important for developing effective remediation strategies and identifying responsible parties in a contamination incident. Groundwater source identification problems require solution of an inverse problem. Gradient-based local optimization approaches are among the most popular approaches for solving these inverse problems. While these methods are sometimes appropriate, they are not effective for problems that contain several local minima and for problems where the decision space is highly discontinuous or convoluted. For these types of problems, heuristic global search approaches such as genetic algorithms (GAs) are more effective. But methods such as GAs are inefficient for fine-tuning solutions once a near global minimum is found. For problems that contain several local minima, a hybrid approach starting with a global method and then fine-tuning with a local method may be more attractive, especially if the decision space is reasonably well behaved near the solution. In this paper, we compare several popular optimization methods for solving a simple groundwater source identification problem and show that hybrid GA-local search (GA-LS) approaches are generally more effective than using stand alone versions of each method. Some variants of the GA-LS approaches are then implemented on a parallel supercomputer to solve a more complex three-dimensional problem.

A water distribution system is a collection of hydraulic control elements jointly connected to convey quantities of water from sources to consumers. Such a system can be described as a graph with the nodes representing the sources and consumers, and the arcs—the connecting elements (e.g., pipes, pumps, and valves). Theoretically, the flow in each arc can reach either direction, resulting in $2^n$ possible digraphs, where $n$ equals the number of arcs. However, this number is substantially reduced as Kirchoff’s Laws 1 and 2 (continuity of mass and energy, respectively) hold, and as in certain arcs the flow is constrained to only one direction (e.g., the pipe leading out of a well). This study describes a methodology for establishing the most flexible pair: Operational and backup digraphs of a water distribution system that maintains Kirchoff’s Laws 1 and 2, and yields (if possible) a one-level system redundancy (i.e., if one arc fails, at least one path from at least one source to all consumers is retained by the operational or backup digraphs). The proposed methodology is cast in a genetic algorithm framework and demonstrated through two example applications.

Presented is a simulation/optimization (S/O) model combining optimization with BIOPLUME II for optimizing in situ bioremediation system design. The S/O model uses a new hybrid method combining genetic algorithms and simulated annealing to search for an optimal design and applies the BIOPLUME II model to simulate aquifer hydraulics and bioremediation. This new hybrid method is parallel recombinative simulated annealing, which is a general-purpose optimization approach that has the good convergence of simulated annealing and the efficient parallelization of a genetic algorithm. We propose a two-stage management approach. The first-stage design goal is to minimize total system cost (pumping/treatment, well installation, and facility capital costs). The second-stage design goal is to minimize the cost of a time-varying pumping strategy using the optimal system chosen by the first-stage optimization. Optimization results show that parallel recombinative simulated annealing performs better than simulated annealing and genetic algorithms for optimizing system design when including installation costs. New explicit well installation coding improves algorithm convergence. Threshold accepting reduces computation time 43% by eliminating unnecessary simulation runs. Applying the optimal time-varying pumping strategy in the second stage reduces pumping cost by 31%.