I meant if we have a pdf (probability distribution func) of generation and load, how can we sample it without using Monte Carlo method?
I will try an answer even if I am not sure of understanding well your question. A pdf has the main property of being integrated to get a cumulative distribution with which is possible to perform a statistical simulation. An alternative way of performing the same computation is a deterministic method. The comparison between these two types of solution becomes more clear when applied to a simple case like the computation of an integral. I would suggest reading of the article of F. James 1980 Rep. Prog. Phys. 43 1145, specially the part on numerical quadrature. A nice modern reference is the book Exploring Monte carlo Methods by William L. Dunn and Kenneth Shultis.
A very simple way of sampling from an arbitrary pdf is this:
1. draw a random number, r, from the uniform dist. on [0, 1]
2. find the value x, for which the cdf(x)=r where cdf is the cumulative dist. function
the random number x has the desire probability distribution. If the pdf has a simple functional form, it might be possible to find an analytical expression for the inverse of the cdf. Otherwise, this needs to be done numerically for each draw step.
The random sampling of random numbers of a distribution function can lead to a bundling in the range of a mode value. Especially large or small values could then be underrepresented in a small number of iterations. With Latin Hypercube Sampling, the range between 0 and 1 (in which the random numbers are generated) is subdivided into equal areas (corresponding to the number of iterations).
As a result, a better representation of the desired distribution can be achieved with a smaller number of iterations.
See also for example: http://kb.palisade.com/index.php?pg=kb.page&id=28
- Badges
- Science topic
- Similar topics
- Computer Science
- Research Topics