Question: Running simulations in different number of seeds to increase the robustness of results

[GVer11]

Dear all,

I am a MSc student, currently working on my thesis project and I'm going to use OpenFAST in conjunction with SIMPACK to calculate the internal loads and fatigue of the drivetrain of bottom-fixed and floating offshore wind turbines. To that end, I'm going to use all the required modules and TurbSim to generate the turbulent conditions.

Since I'm new to OpenFAST, I am a little confused with the seeds notion in a simulation. According to IEC Standards, to get robust results, we should use a minimum of 6 seeds. In the TurbSim input file, the terms "RandSeed1" and "RandSeed2" are included. According to the TurbSim's User guide, it is recommended to assign the "RanLux" value to the "RandSeed2" term.

According to my understanding, to satisfy the Standard's recommendation, I should run 6 simulations with the same inflow conditions, but different values for "RandSeed1", namely 6 sets of seeds. Is that correct?
If yes, in order to reduce the computational time while still getting valid results, could I compare one parameter considering gradually the averages from 1 to 6 seeds, for example for 2 seeds, consider the average value of the parameter for the first 2 simulations (2 sets of seeds), for 3 seeds, calculate the average value of the first 3 parameters (3 sets of seeds) etc and see under which number of seeds the average values converges?

Otherwise, what would be a correct way to approach that requirement?

Thank you very much in advance for your time and help!

[jjonkman]

Dear @GVer11,

Clause 7.5.6 of the offshore wind turbine design standards (IEC 61400-3-1 for fixed bottom and -3-2 for floating) state how to handle the seeds. Some load cases involve choosing the worst case (absolute maximum) load across all seeds; some load cases involve choosing the mean of the worst loads across seeds.

From my experience, the minimum number of seeds specified in the standard are less than would be required for convergence of the statistics (i.e., running less than the 6 seeds is not advised, and running more than 6 seeds would be required if your goal is to ensure convergence).

Best regards,

[GVer11]

Dear @GVer11,

Clause 7.5.6 of the offshore wind turbine design standards (IEC 61400-3-1 for fixed bottom and -3-2 for floating) state how to handle the seeds. Some load cases involve choosing the worst case (absolute maximum) load across all seeds; some load cases involve choosing the mean of the worst loads across seeds.

From my experience, the minimum number of seeds specified in the standard are less than would be required for convergence of the statistics (i.e., running less than the 6 seeds is not advised, and running more than 6 seeds would be required if your goal is to ensure convergence).

Best regards,

Dear Dr. Jonkman,

Thank you very much for your prompt and detailed reply. It is now clearer to me and I will look into the specified segments in more detail.

To be sure about the first part of my question, using different seeds means to run the simulation under the same conditions but different value of "RandSeed1" (or respectively different set of "RandSeed1" and "RandSeed2" if "Ranlux" is not used)?

Best regards,
George

[jjonkman]

Yes, I agree regarding RandSeed1 and RandSeed2.

Best regards,