Bootstrap wind data to account for autocorrelation and obtain new slope

Question

mashtine 2015-8-26

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/236689-bootstrap-wind-data-to-account-for-autocorrelation-and-obtain-new-slope

回答： Cloud Blend 2017-3-27

Hi there,

I have a sample dataset (12783x2) containing the time and wind speed values overs 35 years (6 hour intervals). I have noted that there is a large amount of autocorrelation in the wind data and thus I would like to use bootstrapping re-sampling to perform a new regression on the data with time to find a new slope that describes the long term trend in the wind speeds.

I have been trying a few options with bootstrap but cannot get my polyfit function to output a new slope for the resampled dataset. Instead I get multiple slopes and intercepts.

bootstrp(1000,@polyfit,test(:,1),test(:,2),1);

I know I am writing this code incorrectly. Any idea how I can just get one new slope of the bootstrapped data?

I will also like to perform the wblft() function on the data as well to see if the resampled data with bootstrp has a similar distribution to the original.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Cloud Blend 2017-3-27

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/236689-bootstrap-wind-data-to-account-for-autocorrelation-and-obtain-new-slope#answer_260565

在 MATLAB Online 中打开

Because 1000 is your first argument, bootstrp is iterating on your data 1000 times, and returning 1000 slopes and 1000 intercepts. You could simply take the mean

mean(bootstrp(1000,@polyfit,test(:,1),test(:,2),1),1);

If your resamples are sparse enough, that may help with your autocorrelation problem. You may want to compute autocorrelation for a number of different time lags and decide what is best using a high p value or a rho that is close to zero:

maxSamples = 1000; %prevents very large correlations from taking too much memory
maxSkip = 1461; %1 year
rhos = NaN(1, maxSkip);
ps = NaN(1, maxSkip);
for frameSkip = 1:maxSkip
  sampleTotal = size(test,1)-frameSkip;
  sampleInterval = floor(sampleTotal / maxSamples);
  bufferSize = min(maxSamples, sampleTotal);
  buffer = NaN(bufferSize,2);
  sourceCounter = 1;
  for destCounter = 1:bufferSize
      buffer(destCounter,1) = test(sourceCounter, 2);
      buffer(destCounter,2) = test(sourceCounter + frameSkip, 2);
      sourceCounter = sourceCounter + sampleInterval;
  end
  [rho, p] = corr(buffer);
  rhos(frameSkip) = rho(2);
  ps(frameSkip) = p(2);
end