Computing the appropriate Power Spectral Density

I am trying to compute the power spectral density for a given set of data, but I am getting a variety of results for several different methods, and I am unsure which produces the most relevant graph.

My data consists of fluctuating force measurements due to a turbulent wind, recorded in Newtons over 15 minutes, with readings taken every 1/800 of a second (a total of 721600 readings). Ideally, I would like to produce a graph of W/sqrt(Hz) by Hz. I have tried several different processes including but not limited to: taking the FFT of the autocorrelation function, using periodogram function, using pwelch function. I am still a little confused about the adjustments of the windows and how to know which graphs are producing relevant results. No matter which method I use, I seem to end up with negative y-axis values, which I do not want. Also, all of these commands seem to calculate in dB, whereas I need a conversion to Watts. Can someone please help shed some light on the methods for calculating PSD? Just using the MATLAB functions does not seem to work without further manipulation.