• No results found

The Evolutionary Computing in the design context

SECTION VIII – Evolutionary computing

1. The Evolutionary Computing in the design context

In the following paragraphs, I give an overview of what Evolutionary Computing is. I indicate the areas of its application and explain its basic mechanisms. Moreover, I show how this programming technique can be applied to the generative design system.

A number of biological mechanisms, such as growth, differentiation, wholeness, hierarchical order dominance, control and competition appear to be applicable in a variety of disciplines and in this sense may be said

‘universal’ (Bertalanffy, 1968). Identifying, defining and examining these concepts and their applications in different disciplines were the aim of systems theory. These mechanisms, generalized, gave rise to new methods for solving problems. Evolutionary Computing is one of them. It is a research area within computer science, which is inspired by the mechanisms of natural evolution. EC is successfully applied to many types of problems in fields as diverse as art, biology, chemistry, economics, engineering, genetics, operations research, physics, robotics and social sciences. When it comes to generative design systems, EC finds its application in research rather than in commercial software. Selected EC applications to generative design systems are discussed in section IX.

Although architecture generated by these systems can be called organic, it does not necessary imitates shapes of biological organisms. Rather, these systems apply the processes and mechanisms that are inspired by biology.

Generally, there is a change in recent design thinking on organic design:

“‘nature’ as a source of shapes to be copied” has been replaced by “‘nature’

as a series of interrelated dynamic processes that can be simulated and adapted for the design and production of architecture” (Hensel et al. 2010:

27).

In the natural evolution model, a fundamental concept is the environment filled with populations consisting of individuals. The time of individuals’

lives is limited and thus they reproduce in order to preserve the population.

The reproduction, if not stopped, would result in a population size that grows exponentially. But the environment can host only a limited number of individuals, because of the limited amount of resources. The individuals who are most likely to survive are those who utilize the resources most effectively.

Each individual in a population varies from the others. The set of behavioural and physical features of an individual is called phenotypic traits. The

phenotypic traits directly respond to the environment, determining the fitness

of the individual. The chance of survival and reproduction of each individual is thus determined by its fitness. The better-fitted individuals proliferate, copying the desired features to their offspring. The offspring are not exact copies of their parents, but they are randomly modified, and it happens that some of the children are more fitted than their parents. These children in turn remain in the selection process, giving birth to next generation. In this manner, the population changes over time such as the average fitness of an individual increases (Eiben and Smith, 2007).

GENERAL PROBLEM SOLVING IN EC

Following the evolutionary mechanism, one can construct a universal algorithm for improving the quality of a solution for a number of problems. A starting point for such an approach would be the generation of a number of random solutions for a given problem. For example, if the problem were to find a best sequence of the next three moves in a chess game, the algorithm could generate a hundred random sequences of possible next three moves (anticipating the opponent’s responses). These random sequences of three moves could be referred to as candidate solutions. The level of how well a particular candidate solution (a particular sequence) solves the problem would define the candidate’s quality. The algorithm would then select a certain number of solutions (the most beneficial sequences of moves), reproduce them in a number of copies, and modify the copies. The modified copies of the best solutions would become a new collection of the candidate solutions, presumably with a higher quality on average. The cycle would be repeated many times, until a satisfactory solution is found.

The fitness of each individual is determined by the environment that the individual lives in – the same individual placed in a modified environment would have a different fitness value. Analogously, the same solution would have a different quality if the problem were formulated differently. The analogy described above is conventionally called the main Evolutionary Computing metaphor (Figure 11).

EVOLUTION PROBLEM

SOLVING

environment problem individual candidate solution

fitness quality of solution

Figure 11. The Evolutionary Computing metaphor (after Eiben and Smith, 2007).

EC APPLICATION TO THE GENERATIVE DESIGN SYSTEM In the following paragraphs I will establish a link between the problem solving mechanism and the elements of the generative design system developed so far, such as: a building model, building characteristics and a building environment. Firstly, a problem definition corresponds to the definition of a building environment. The design requirements (site

constraints, building codes, master plan regulations) and the design intentions (client’s intentions and architectural qualities) form the design problem. A candidate solution for such defined design problem would be a building model. A set of initial solutions for the problem would be a set of randomly generated building models. The solution finding procedure would be analogous to the natural selection process: the building models that best fit the building environment would be reproduced and modified, while the less fitted would gradually disappear. As a result, a new ‘generation’ of building models would appear. This generation would undergo the same procedure, until a satisfactory building model had been found.

Furthermore, determining the fitness of an individual building model would consist in evaluating of how similar the building characteristics of the individual building model are to the building characteristics defined in the building environment. The fitness of a building model cannot be measured in absolute terms, but is relative to the building environment, just as the fitness of an individual organism depends on the environment the organism lives in.

Because the building environment consists of selected design objectives, the fitness of a building model would be measured by the design objectives.

Figure 12 summarises the basic evolutionary computing metaphor and applies it to the domain of architectural design.

Figure 12. Extension of the main Evolutionary Computing metaphor to architectural design.

EVOLUTION PROBLEM

SOLVING ARCHITECTURAL

DESIGN environment problem building environment

(design objectives) individual candidate solution building model

fitness quality of solution fitness of building model

2. BASIC NOTIONS OF EC IN THE CONTEXT OF A