Evolutionary Modeling of Routes: The Case of Road Design

In this study a route is defined as the path that a linear structure or facility follows in the terrain. Linear structures comprise facilities such as roads, motorways, railways, pipelines, electrical power lines, and telephone cables, each of these structures requiring specific technical parameters in what concerns the geometry of the path and having different effects on the terrain they traverse. Amongst these structures, roads and motorways are the group that creates the greatest overall impact; accordingly Portuguese legislation requires an environmental impact assessment (EIA) process as part of the necessary licensing approval. Usually the alternative (or alternatives) that undergo the EIA process is justified in terms of technical and economical issues. The result is that if major environmental impacts are identified by the EIA study, a myriad of mitigation measures are proposed, very seldom the redesign of the path being carried out (Guimarães Pereira & Antunes, 1996). Preliminary studies that precede the implementation of these types of projects are technically detailed and often come together with economical feasibility studies, shelving environmental issues for later assessment. In the methodology proposed in this chapter a multidimensional evaluation methodology, multicriteria evaluation, will be combined with the robustness of a search methodology, genetic algorithms (GAs) to generate alternative road routes that take into consideration environmental, economical, technical, and social criteria. These criteria are referenced to the physical space where the road is to be placed and therefore this methodology is embedded into a geographic information system (GIS). Genetic algorithms are particularly attractive to apply to multi-modal problems, allowing the exploration of spatial features to eventually find “best compromise” alternatives because these algorithms proceed their search by maintaining a population of solutions, that they can simultaneously exploit for their efficiency.1 Moreover, the particular mixing mechanism provides the means to recombine solutions and explore the search space. The remainder of this chapter describes evolutionary modeling of road routes, in particular the coding onto a GA of the geometric algorithm that accounts for the technical aspects of motorway siting. The details of the implementation of the MCDA-GA methodology, running within the GIS GRASS 4.1 (Geographic Resources Analysis Support System) and its application to generate and evaluate alternative routes of a section of a Portuguese complementary itinerary (IC7) will be presented.

Designing Genetic Algorithms to Solve GIS Problems

What makes a problem hard for a genetic algorithm (GA)? How does one need to design a GA to solve a problem satisfactorily? How does the designer include domain knowledge in the GA? When is a GA suitable to use for solving a problem? These are all legitimate questions. This chapter will offer a view on genetic algorithms that stresses the role of the so-called linkage. Linkage relates to the fact that between the variables of the solution dependencies exist that cause a need to treat those variables as one “block,” since the best setting of each individual variable can only be determined by looking at the other variables as well. The genes that represent these variables will then have to be transferred together. When these genes are set to their optimal values, they constitute a building block. Building blocks will be transferred as a whole during recombination and the building blocks of all the genes make up the optimal solution. As will become apparent, knowing the linkage of a building block is a big advantage and will allow one to design efficient GAs. Sadly, in the majority of problems, the linkage is unknown. This observation has given rise to a lot of development in linkage learning algorithms (for an example, see Kargupta 1996). However, there is a specific class of problems that allows for relatively easy determination of linkage: spatial problems. This is because in these problems, the linkage is geometrically defined. We will focus in this chapter on certain hard problems that arise in the context of geographical information systems and for which the linkage can be easily found. Specifically, we will fully detail the design of a GA for the problem of map labeling, which is an important problem in automated cartography. The map labeling problem for point features is to find a placement for the labels of a set of points such that the number of labels that do not intersect other labels is maximized.

A Genetic Algorithm to Design Optimal Patch Configurations Using Raster Data Structures

The design of optimal patch configurations is a generic problem relevant to many spatial planning exercises. Spatial pattern affects processes in the natural and manufacture of environment and should be incorporated as a criterion in planning. Currently, while geographic inormation systems (GISs) are adequate for data storage, analysis, and visualization they do not provide sophisticated spatial decision-making functions. With the help of GISs, pattern can be incorporated into spatial decision-making explicitly, using ad hoc procedures, or implicitly, through visualization of alternative plans. Other computer technologies like remote sensing and decision support systems facilitate decision-making by supplying timely data and techniques for solving multi-criteria evaluation problems. There are now a number of artificial intelligence techniques that can be coupled with GIS to address a variety of hard spatial problems. Genetic algorithms are particularly attractive for optimization problems because they are efficient and effective in complex search spaces. Landscape ecologists use the twin concepts of patch and matrix to describe the spatial structure of the environment (McGarigal & Marks, 1994). The matrix is the dominant landscape element and patches are distributed within it. Patches can be crisp objects with well-defined boundaries, such as administrative areas, or inferred objects with fuzzy boundaries, such as vegetation or habitat patches in natural environments. In the former case, patches can be adequately represented by polygons in a vector GIS. In the second case patches are inferred from a continuous spatial distribution of attribute values. The raster data model is the most common representation of continuous fields within a GIS and is preferred to vector models in environmental applications because it is a better representation of the continuous variation characteristic of natural phenomena. There is a need for decision support tools that use raster GISs when spatial criteria relate to natural phenomena. Patch design involves many complexities: in a raster GIS it is also a complex problem in spatial geometry. This chapter describes a genetic algorithm for designing patch configurations in raster GISs. The genetic algorithm is coupled with GIS and multicriteria evaluation functions to build an autonomous system that explicitly includes pattern as a criterion in the design of patch networks. Conceptually, the problem is to extract from an infinite set of possible spatial patterns a single pattern that is optimal by some criterion.

Modeling Spatial Phenomena

In part II we describe some possible methods of modeling spatial phenomena with spatial evolutionary algorithms. We will explain what spatial evolutionary models and spatial evolutionary algorithms are and how they can be designed. We will also provide a general framework for spatial evolutionary modeling. We believe that this framework can be used to create evolutionary models (and algorithms) of spatial phenomena that will reach well beyond the model discussed in the book. Wherever possible we will give examples to illustrate the concepts, terms, and procedures we discuss. In fact, by the end of part II we will have built, using presented principles, a complete spatial evolutionary model—a spatial evolutionary model of a wireless communication system. We shall begin our discussion with an explanation of the distinction between spatial evolutionary models and evolutionary models of spatial phenomena. As we shall see, the difference between these two terms, while subtle, is very important for the understanding of spatial modeling in general and evolutionary spatial modeling in particular. . . . "Spatial Evolutionary Models" Versus "Evolutionary Models of Spatial Phenomena" . . . The differences between the terms spatial evolutionary models and evolutionary models of spatial phenomena extend well beyond their lexical dissimilarities and touch upon very basic issues of evolutionary and spatial modeling. The term spatial evolutionary model, as used here, refers to an evolutionary model that constitutes a separate, distinct class of computer evolutionary models. In contrast, the term evolutionary models of spatial phenomena denotes applications of existing evolutionary methods (or mere extensions of established evolutionary methodologies) to problems defined in space. Our view of the science of spatial modelling is driven by the choice of which definition, along with its consequences, that we accept. If we accept that spatial evolutionary models constitute a separate and distinct class of evolutionary models, then we will also have to accept the proposition that they possess unique rules governing their behavior, a unique genome design to represent a model-specific data structure, and a set of unique operators that cannot be readily applied to nonspatial problems. Moreover, it will follow that these evolutionary models also possess problem-specific language, that is language specific to the domain of spatial evolutionary models.

Concepts of Evolutionary Modeling and Evolutionary Algorithms

This book is about evolutionary algorithms as applied to spatial and geographic phenomena. Why are we writing this book? Do these new algorithms deliver solutions to our modeling and data analysis problems that conventional methods cannot handle? Or will they just fade away, as have so many other “new” ideas from the past, some eventually finding their way into a museum of computer and conceptual contraptions? It is not our purpose here to attempt to present answers to all of these questions. This is not because we lack expertise but because we do not yet know the answers. However, what we can do, and what we intend to do in this book, is to offer the reader a proper perspective within which to look at evolutionary models. For some, this perspective may prove disappointing, as we will not to solve all known information modeling ills. However, our perspective will help the reader to understand evolutionary computer methods and related concepts, and to use them in appropriate applications and models. In other words, we guarantee that the reader will not be disappointed in coming to a clear understanding as to what evolutionary modeling is all about. Admittedly, other books already serve this purpose. What we offer here, which we feel is unique, is a perspective on the new area of the application of evolutionary models—the area of spatial and geographic phenomena. We shall begin with a broad introduction to models and modeling. This introduction will go well beyond the scope of computer science and geographic information systems (GISs) and will touch upon wider philosophical issues. We believe that modeling is a serious undertaking and it may have serious consequences for the modeler, the modeling subject, and even the lay public. In this introductory chapter, our objective is to assist the reader in coming to understand the modeling business as we see it (and would like others to see it). Modeling is as old as human civilization. Models shape what we are, how we define our reality, known saying: and what forms our thoughts may take.

Airspace Sectoring by Evolutionary Computation

When joining two airports, aircraft must follow routes and beacons; these beacons are necessary for pilots to know their position during navigation and because of the small number of beacons on the ground they often represent crossing points of different airways. Crossing points may generate conflicts between aircraft when their trajectories converge on it at the same time and induce a risk of collision. At the dawn of civil aviation, pilots solved conflicts themselves because they always flew in good weather conditions (good visibility) with low-speed aircraft. In contrast, modern jet aircraft do not enable pilots to solve conflicts because of their high speed and their ability to fly with bad visibility. Therefore, pilots must be helped by an air traffic controller on the ground who has a global view of the current traffic distribution in the airspace and can give orders to the pilots to avoid collisions. As there are many aircraft simultaneously present in the sky, a single controller is not able to manage all of them. Airspace is then partitioned into different sectors, each of them being assigned to a controller. Sectoring is currently done in an empirical way by some airspace experts who apply rules they have learned with experience. The sectoring modifications are usually due to traffic evolution over long period and when a sector is regularly overloaded it has to be modified. To reach this aim, an ad hoc commission meets to identify new boundaries for the sectors in order to balance the workload. Afterward, sectoring is updated (until new problems arise). This way of working is relevant because it takes into account several practical aspects but has a limited effect on the local zone it treats. This process can be improved with an automatic approach in order to give a solution to the sectoring problem in the whole airspace and that solution could be refined by experts. Before specifying a mathematical description of our problem, it is necessary to set out our framework to introduce some simplifications for our model.

Beyond Data: Handling Spatial and Analytical Contexts with Genetics-Based Machine Learning

Geographic information systems (GISs) are fairly good at handling three types of data: locational, attribute, and topological. Recent work holds promise for adding temporal data to this list as well (e.g., see Langran, 1992). Yet the unprecedentedly vast resources of geographically referenced data continue to outstrip our ability to derive meaningful information from such databases, despite dramatic improvements in computer processing power, algorithm efficiency, and parallel processing. In part this is because such research has emphasized improvements in processing efficiency rather than effectiveness. We humans are slow-minded compared with our silicon inventions; yet our analytical capabilities remain far more powerful, primarily because we have evolved elaborate cognitive infrastructures devoted to ensuring that we leverage our limited processing power by focusing our attention on the events and information most likely to be relevant. In GIS use, so far only human perception provides the requisite integration of spatial context, and human attention directs the determination of relevance and the selection of geographic features and related analyses. Understanding of spatial context and analytical purpose exists only in the minds of humans working with the GIS or viewing the displays and maps created by such operations. We still extract information from our geographic data systems primarily through long series of relatively tedious and complex spatial operations, performed—or at least explicitly preprogrammed—by a human, in order to derive each answer. Human integration of analytical purpose and spatial and attribute contexts is perhaps the most essential and yet the most invisible component of any geographic analysis, yet it is also perhaps the most fundamental missing link in any GIS. Only humans can glance at a map of a toxic waste dumps next to school yards, or oil spills upstream from fisheries, and recognize the potential threat of such proximity; human cartographers understand the importance of emphasizing either road or stream networks depending on the purpose of a map; humans understand that “near” operates at different scales for corner stores versus cities, or tropical jungle habitat versus open savannah. Given a GIS with the capability to deluge any inquiry with myriad layers of extraneous data, this natural human ability to filter data and manipulate only the meaningful elements is essential.