2

I have two datasets namely reference dataset and new dataset; All datasets have power and speed for a power curve (Power Vs Speed) of wind turbine. Power is in vertical axis and Speed is in horizontal axis. I want to use my new incoming data points of speed to find the probability of its associated power value using pdf function based on my calculation on reference data (Calculated binning, mean value & std value for old data).

In short, this is the code I have written so far and after that I am clueless or not getting what to do next : 

numbin = 30;
Edge = linspace(min(Speed_old),max(Speed_old),numbin);
[h,whichBin] = histcounts(Speed_old,Edge);
mu = Speed_mean_Ref;
sigma = std_ref;
f = exp(-(Edge- mu).^2./(2*sigma^2))./(sigma*sqrt(2*pi));

After that what should I do for a given a new data point (speed_new), find the probability density of this data point based on the kernel density estimator (or pdf) and then its associated power value.

EBH
  • 10,350
  • 3
  • 34
  • 59
ravi pandit
  • 117
  • 12
  • dear @Ares I have written my code(via updating my question) so far please have a look. After that I clueless :( – ravi pandit Aug 01 '16 at 15:54
  • I've never done this exact thing, but with similar things I have used the `ecdf` function to find the cumulative density function and gone from there. – Matt Aug 01 '16 at 16:45
  • @Matt thank you for your kind help... But I wonder whether ecdf is suitable for this or not because I am assuming and want to also verify the distribution is Gaussian and I think this can only be done with PDF. If you can share some example what you have done with ecdf then I would be grateful... Thank you :) – ravi pandit Aug 01 '16 at 20:56
  • There is no big difference between pdf and cdf, and you can jump between them [as you like](http://www.mathworks.com/help/stats/normal-distribution-1.html). The question is what are you trying to do? The distribution of the speed is one dimensional, so given new speed data, all you can say is if it com from the same distribution. If you want to relate speed to power then you look for some regression method. Testing for the "normality" of the distribution is done by chi^2 test or k-s or other tests, it has nothing to do with the power values. – EBH Aug 10 '16 at 19:27

0 Answers0