Document Actions
Optimization hangs
Up to Bug Reporting
Posted by
Nick,
I'm optimizing a project that has environmental sensors (noise) and a fairly complicated set of constraints that I turned into a raster layer. I set the optimization to growing layouts with loose constraints on array losses and capacity factor to get a maximum layout using the noise and other constraints. I optimized it once and got 52 turbines, and then discovered that I was missing a constraint layer, which took away a small chunk of buildable area where a turbine was placed. I set the optimization to run again with the 52 turbines - openWind told me there were no layouts possible. So I deleted the one turbine and let it optimize, and it stayed at 51 turbines.
So, copied the entire layer and took it down to 10 turbines and let it reoptimize with growing layouts. And now it's at 53 turbines, despite that I have less land than before. Also, in reviewing the position of similarly placed turbines and the power density layer, it appears that the 53 turbine layout has slightly better placement.
As a test, I ran the process again. This time, the project capped out at 45 turbines. I'm currently running it once more from one turbine to see the results.
My concern is that there's a certain point where the optimizer gets "stuck" - While this is easy to see in a layout that uses set constraints, it gets tricky with noise, and I can never be confident that the layout has indeed optimized.
I'm optimizing a project that has environmental sensors (noise) and a fairly complicated set of constraints that I turned into a raster layer. I set the optimization to growing layouts with loose constraints on array losses and capacity factor to get a maximum layout using the noise and other constraints. I optimized it once and got 52 turbines, and then discovered that I was missing a constraint layer, which took away a small chunk of buildable area where a turbine was placed. I set the optimization to run again with the 52 turbines - openWind told me there were no layouts possible. So I deleted the one turbine and let it optimize, and it stayed at 51 turbines.
So, copied the entire layer and took it down to 10 turbines and let it reoptimize with growing layouts. And now it's at 53 turbines, despite that I have less land than before. Also, in reviewing the position of similarly placed turbines and the power density layer, it appears that the 53 turbine layout has slightly better placement.
As a test, I ran the process again. This time, the project capped out at 45 turbines. I'm currently running it once more from one turbine to see the results.
My concern is that there's a certain point where the optimizer gets "stuck" - While this is easy to see in a layout that uses set constraints, it gets tricky with noise, and I can never be confident that the layout has indeed optimized.
Hi Matt,
There was a certain point where the optimiser gave up trying to add turbines. I have removed that feature in the latest version. It was an attempt to save some processing in sites which were already full but I suspect it doesn't make all that much difference and so now I leave it to the user to switch it off if the extra processing is a concern.
Please bear in mind that it is a stochastic optimiser and as such it can get caught in local optima or trapped by random events in the early part of the optimisation. I suspect that the change that I outline above will make subsequent runs more similar.
I will update your executable later today.
Nick
Powered by
Ploneboard