Role of Evolutionary Computation in Environmental and Water Resources Systems Analysis

Since the 1960s, systems analytic methods have played a key role in environmental and water resources planning and management. An array of systematic search procedures as well as statistical methods continues to be the focus of investigation in environmental and water resources systems (EWRS). With the advent of modern and faster computing resources at a high degree of affordability, the system simulation models are now able to incorporate more processes and their interactions, resulting in relatively more complex model structures. Therefore, the coupling of simulation models with search methods for optimization have become increasingly challenging. In many cases, complexities that arise from, for example, nonlinearity, discontinuity, and discreteness in modern simulation models, limit the application of traditional search methods, e.g., mathematical programming procedures. Such limitations have been overcome recently by directly coupling the simulation models with heuristic search procedures. While this directly coupled simulation–optimization (S–O) approach is, in general, computationally demanding, it has proven to be a viable approach given cheaper and faster computational resources.

Among the array of modern heuristic approaches (e.g., simulated annealing, tabu search, genetic algorithms, evolutionary strategies, particle swarm method, and ant colony optimization) that support an S–O framework for EWRS problems, the collection of methods—namely, genetic algorithms, evolutionary strategies, genetic programming, and evolutionary programming—encompassed within the broad category of evolutionary computation (EC) offers a multitude of capabilities. Starting in the early 1990s, evolutionary algorithms (including genetic algorithms and evolutionary strategies) have been applied and demonstrated for system optimization in the context of EWRS problems. Numerous studies over the past decade in areas that include groundwater monitoring and remediation design, water distribution network design, reservoir optimization, watershed management, and air pollution control show not only the viability of applying evolutionary algorithms to these challenging problems, but also the added benefits of using a directly coupled S–O approach enabled by these techniques. This special issue also represents a cross section of recent investigations related to EC methods and their applications in EWRS.

In addition to system optimization, many EWRS problems pose several interesting scenarios that call for additional systems analytic capabilities. For example, most EWRS problems consider multiple competing objectives, requiring the systems analytic methods to provide multiobjective optimization capabilities. The structure of EC methods readily support efficient search for Pareto optimal solutions to a multiobjective optimization problem. This is currently an active area of research in the EC research community, and the results and findings from those efforts are already making their way into multiobjective analysis of EWRS problems. It is also important to note that recent research investigations in multiobjective optimization in EWRS areas have contributed to EC-based multiobjective optimization methodologies.

Another scenario commonly encountered in EWRS problems is optimization under uncertainty or noisy conditions. Again, the typical structure of EC-based search procedures facilitates these methods to be readily suitable for convenient adaptation to search under noisy conditions. A branch of EC research is focused on developing search methods that perform robustly in the presence of noise and dynamic variations. While this topic is still in its infancy, several promising methodological advances have been reported by researchers in both the EC and the EWRS communities. An immediate challenge associated with this topic is the increase in computational needs brought upon by a larger number of simulations required to represent multiple system states. More research and development is needed to enable newer methods with enhanced performance. As this topic is extremely important in most of the EWRS problems, this area of research in EC-based methods is anticipated to grow.

As EWRS problems are typically plagued with the complexities of decision making (e.g., ill-defined or incompletely defined models due to unquantifiable issues) in public sector problems, the “optimal” solution alone may prove to be insufficient. In such cases, the model should be used to explore the decision space to identify “good” alternative solutions to support the decision-making process. The population-based search conducted in EC search methods provide a convenient framework to explore simultaneously for multiple solutions to the search problem. The EC research community has been investigating operators to find alternative niches in the decision space. While this offers a potential approach, it does not fully exploit the capabilities of EC-based methods to identify maximally different alternatives. Recent investigations by EWRS researchers have resulted in useful EC-based approaches to not only find the “optimal” solution to a modeled problem, but also to generate a few maximally different and near-optimal solutions.

In addition to numeric search capabilities offered by the EC-based methods, they also support specific approaches to conduct symbolic search. For example, genetic programming and evolutionary programming methods can operate on symbols to search for rules and model structures. These capabilities are beneficial in addressing EWRS problems that require empirical model induction, symbolic regression, data mining, or data-driven model development. Unlike other classes of search methods, EC-based methods incorporate operators to facilitate solutions to these challenges. These symbolic search methods are part of a major arm of EC research community, and therefore offer an array methodological development that are readily beneficial to the EWRS community. While this is already reflected by the numerous examples of model induction, symbolic regression, rule induction, and function fitting applications in the EWRS literature, more development and advanced applications can be expected in this area.

As EC-based methods enable an inherently parallelizable algorithmic structure, they hold immediate promise for implementations that offer significant computational enhancements. Most realistic EWRS problems demand significant computational resources when represented in a directly coupled S–O framework. While existing studies in the EC community offer some potential approaches for parallel∕distributed implementations of EC-based search methods, they are still limited in scope when faced with complexities such as coarse-grained structure of the search, heterogeneity of processors and communication protocols in the computer network, multilevel parallelism, and synchronization, if any, of the search algorithm. Further, the increasing availability of the grid-computing environment on the Internet could not only support large-scale S–O applications in EWRS problems, but also make it feasible to solve them in reasonable execution times. While some research in this area is underway, more investigations and findings are expected to lead to contemporary parallel∕distributed EC-based methods that would better harness the available network of computational resources.

While rapid progress in integration of EC in EWRS has been made in the past decade as described and represented by the collection of papers in this special issue, the full potential is yet to be reaped. As the “young” field of evolutionary computation continues to grow and new contributions are realized in the EC community, the EWRS researchers could push the boundaries of analysis and solution of complex environmental and water resources planning and management problems by appropriately bridging the gap between these two disciplines. One could readily draw a parallel between this emerging opportunity for cross-fertilization and the interfacing of operations research and EWRS disciplines since the 1960s. Using that as a guideline, the EWRS researchers could anticipate what new areas would be important, and proactively influence and shape the direction of a beneficial integration of EC into EWRS. The good and exciting times are ahead of us!