[ebechhoefer]

Dear Sir,

This is interesting, in that is shows that its hard to test software for all contingencies – a run time bug on my part. In fact, you are correct. There was a small set of data that was “reused” unintentionally. The good news is that the set was small relative to the full data set, and, if you identify it correctly, won’t hurt your score.

Most Sincerely,

Eric

[ebechhoefer]

Dear Sir,

Two things: First, the data sets are from real operators/consultants. While the generally accepted practice is to mount the booms 90 degrees from each other – it is, essentially, at the preference of that operator/consultant. That said, I checked the provenance of the data files, and one operator did mount the booms 180 degrees out. Good eye. At some point, I will go back as ask why.

Most Sincerely,

Eric

[ebechhoefer]

I investigated the provenance of Shear2, Shear14,Sheaar26, Shear128, Shear163,Shear205, Shear213, Shear229, and Shear232. these came from the same logger, where I had seen one bad channel. The other channels of these files are OK…. Sorry about the inconvenience – sometime we just have to live with the bad data we have…

Sincerely,

Eric

[ebechhoefer]

Typically, the booms are mounted 90 degrees relative to each other. You can, through analysis of the paired data, determine the absolute direction. I am sorry that I can’t be more specific.

Most Respectfully,

Eric

[ebechhoefer]

Hi there,

The file format was designed to include only existent anemometers. Some data off of the logger is NaN. I have found a few small errors in the logger data – this NaN may well be one. Can you give me the specific shear test file number? Then I can go back to the original logger file and look at the provenance.

More Respectfully,

Eric

[ebechhoefer]

Dear Sir,

I have to say that for a moment, I was concerned that I had indexed into the master data set improperly when building the Shear training data sets 2 & 3. However, this is not the case (whew!). In looking at the original data, I find that the logger (channel 6, 10 meter, max, min) did in fact have error and should be ignored. I find that the mean and standard deviation for 10 meter data, though, is correct. I apologizes for that – the rest of the channels look good.

In terms of mean wind speed on the 10 meter anemometer data being low. Yes, this is correct. Depending on site conditions (presence of trees and shrubs) the shear between 10m and 30m can very high. Essentially, that why you want to put the wind turbine up at 80m!

[ebechhoefer]

Hi There,
This is a great questions. As noted, the anemometer rating is in percent error. That said, what one see with failed anemometers is that, because of the increase friction:
a) at lower wind speeds (2 to 5 m/s or so), its 2% relative, but
b) at higher speeds (5 to 11 m/s?), theu are slower by, perhaps, .2m/s.

Not a simple answer, to be sure…..

thanks

Eric

[ebechhoefer]

The training data set consists of nominal data under normal conditions and is less than 2% error. However, if icing is present (a rare event) then there could be training data that is in error. This suggest that some methodology to detect icing must be designed to filter out bad training data.

[Author]

Posts