I have been testing for some time already different approaches for the optimization of exterior fix shading systems. I believe this is one of the most discussed topics and probable the most challenging due to the number of parameters and variables that are involve in this decision.
I have tested Octopus in the past for multiobjective optimization and Colibri along with Design Explorer, I think all these are very powerful tools. However, these are very time consuming approaches and I was wondering if there are any simplistic approach that can be summarize in one single objective that can be use as a fitness in Galapagos.
So far, I have seen different ways to do this, but I am a bit confused about which one is more appropriate for my case.
In my case, I have an office space with southwest orientation located near Dublin and I want to reduce/minimized the harmful radiation during the summer (i.e. when Tout > 16C) and maximised the daylight access. For this, I have the cumulative incident radiation in the windows during the harmful radiation hours for a scenario with and without shades, so I can normalize both daylight and radiation. My indicator for daylight is a normalized DA average for a series of points at 6m distance from the windows against a scenario with no shades.
Based on this, I have two normalized “ratios” (between 0-1) that show how much my shades minimize daylight and radiation against a no shading scenario. I know that If I divide my radiation ratio by my daylight ratio I can obtain the optimal solution in terms of minimum radiation maximum daylight if I look for the minimum result, however, this solution does consider equally important daylight and radiation. I want to give priority to daylight, since it is more important to obtain an acceptable daylight level than reducing solar gains in this situation.
I have tested some ways to do this, but the results just doesn’t make much of a sense and I was wondering if anybody have carryout some similar studies successfully where you need to prioritize one variable more than the other.
They report different results, and I am confused whether I should use one or the other. Any insight is very much appreciated. Single objective optimization test.gh (101.0 KB)
I don’t think you should overcomplicate this. I doubt you will create an optimization which you can sell as professional unless you really know what you are doing. And especially in this case I don’t think you consider all relevant information. I found Galapagos very easy to use, if you think about the fitness function as a “score” system. Think about it. Getting enough daylight can give you 3000 points, whereas less radiation can only give you 2000 at best. Since they contradict each other, you will likely get a better maximum for the daylight somewhere around the center. You can apply a weighting function like in the second example to reduce the occurrence of edge cases. You can add multiple objective like this, and I doubt this simplistic way of seeing it, is by any means worse than using algorithms like Pareto-front. See a generic solver is just a smarter way in doing trial and error. You automate trying to set different slider states, by preventing to repeat the worse states. To not stay in a local extrema, you mutate. But the quality does not come with weighting or the algorithm, but by specifying the parameters and their domain in first place.
Hi @TomTom,
Thank you for your reply. I agree with you that this should not be overcomplicated. I believe meaningful results can be obtained by rather simple methods.
What do you mean by “a weighting function like in the second example”? are you referring to this approach?
Could you also please explain what do you mean by: “But the quality does not come with weighting or the algorithm, but by specifying the parameters and their domain in first place.”?
No I refer to the other image. I proposed you (re-)map a given output into a scoring range. Usually this is linear. The more daylight the higher the score. But essentially you can use a curve/function( “graph-mapper”) to map non-linear. E.g. If you believe that there is something like having too much daylight, then you could represent that with a curve, or that the score raises slower in the beginning and gets faster high in end. Maybe because almost no daylight is as bad as having no daylight. Therefore you can control the scoring in a non-linear manner (if this something to you need).
This just means, you need to consider the correct things in an appropriate context. Now energy sims are not in my professional domain, but I could imagine that if you find an optimal setup, but the cost and the complexity rises exponentially because of requiring custom elements, then you found a theoretical optimum, but a total impractical solution. This is also why I’m a bit against these “optimisations” in first place. Its the other way you should work. Try some realistic configurations and see how good they perform. So give the system realistic bounds, then it doesn’t matter if you take the third or fifth best solution. Hope that makes sense…
Hi @TomTom,
Regarding the first question:
If I am not mistaken the average weighted method that I show in the image is a linear function re-map, not sure why that is not correct. For now, I want to keep things simple and therefore daylight will be more important (more weight) than radiation for all cases and this will be proportional. This is what you mean by second example, right?
Regarding the second question:
I agree with you on this to an extent. For now this is an early stage analysis, and being theoretical it can be very useful in order to have an idea about which parameters are the ones with highest impact so to reduce the reduce the number of variables/options. However, I believe this is very useful to have a general idea of what strategy is effective and for comparison purposes but also taking into account that in most cases custom shading sizes are generally not a cost-effective solution.
