Please note that the interface and visualization of the application are in a totally initial stage and therefore not polished at all.
Test 1, 2: Connectivity
In the first test, the goal was set to achieve as many of the connections specified in the original matrix as possible. The original matrix was semi-randomly generated, by filling a percentage of 3 randomly selected rows/columns while leaving the rest blank. This arrangement corresponds to a centralized circulation layout with a few spaces being having the role of a "hub", and the rest being passive in terms of circulation.
The first test resulted into a solution with very high fitness value, satisfying all connection requirements except two. In order to validate this, a second test with the same parameters was performed. However, in the second test fitness did not turn out as high as in the first. This can be attributed to two reasons:
The evolutionary process is in it's early stage and needs further study to increase the effectiveness of the search for optimal solutions.
An optimal solution is not achievable with the specific connectivity scheme
Results
In the images below you can seen the original fittest and final (after evolution) fittest solutions. Next to the arrangements is a bar displaying the fitness value and an overlay of the required and actual connectivity matrices.
 |
| Test 1: Initial Fittest Solution |
 |
| Test 1: Fittest Solution after approx. 10000 generations of 239 members (10 min runtime) |
 |
| Test 1: Parameters |
 |
| Test 2: Initial Fittest Solution |
 |
| Test 2: Fittest Solution after approx. 10000 generations of 239 members (10 min runtime) |
 |
| Test 2: Fittest Solution, alternate viewpoint |
Discussion of the Results
It is clear in the first test that, while most connectivity requirements are satisfied, spaces with irrational characteristics appear. To satisfy their connectivity, some spaces have gained so much height that the solution is no realistic. Moreover, there are spaces with no adjacency whatsoever to the outside, a feature which further decreases the feasibility of the solution.
It is clear in the first test that, while most connectivity requirements are satisfied, spaces with irrational characteristics appear. To satisfy their connectivity, some spaces have gained so much height that the solution is no realistic. Moreover, there are spaces with no adjacency whatsoever to the outside, a feature which further decreases the feasibility of the solution.
Test 3, 4: Connectivity + Adjacency to Surface
A second performance factor has been added to thess tests, which requires at least one block of every different space to be in contact with the surface of the building volume. Again, the connectivity matrix was semi-randomly generated as in the first test.
A second performance factor has been added to thess tests, which requires at least one block of every different space to be in contact with the surface of the building volume. Again, the connectivity matrix was semi-randomly generated as in the first test.
 |
| Test 4: Initial Fittest Solution |
 |
| Test 4: Fittest Solution after approx. 10000 generations of 250 members (10 min runtime) |
 | |
|
Discussion of the Results
While it seems again that the solution found is performing very well, spaces outside the allowable height range still appear. Essentially, all spaces that are required to have a large number of connections are oriented vertically, so that they can be in contact with as many other spaces as possible.
Test 5, 6: Connectivity + Adjacency to Surface + Height Limit
In the last two tests the aim was to produce solutions that adhere to height restrictions for each space. The representation used so far is dimensionless, therefore "floor" has been chosen as the height unit in this case. This equals roughly 3 meters for a typical office floor. Once again, test was performed using a semi-random connectivity matrix.
 |
| Test 5: Initial Fittest Solution |
 |
| Test 5: Fittest Solution after approx. 10000 generations of 252 members (10 min runtime) |
 |
| Test 5: Initial Parameters |
Discussion of the Results
It is evident that much more feasible solutions begin to appear with the introduction of the height limitation via penalization. Still, probably due to the number of unique spaces and the intense connectivity requirements of some of them, the connectivity satisfaction is impossible to achieve in it's entirety.
Next tests will be after a basic data input interface is implemented, so that working on actual configuration data becomes possible.
Check back soon!
0 σχόλια:
Post a Comment