I'm looking to use a multifactor interest rate model to first fit to historical data, then use the fitted values to simulate future interest rates, for risk management purpose. Any one knows what is the most popularly used model in practice?

have you tried PCA?

There is a wide range of models that can be calibrated for this purpose. An easy choice is a HJM calibrated with PCA. The problem is that in long horizons, it will frequently generate negative rates, but having recent levels of forwards in CHF that should not be considered as a problem (lognormal model would not be able to reproduce real market data). If you consider negative rates as a serious obstacle, Black-Karasinski is a pretty simple alternative, but du to lognormal nature of rates, some scenarios are exploding, so you need to set up some policy what is happening with your portfolio in case of very high rates.

Thanks for the ideas. I found a paper by Algorithms, titled "A Multi factor statistical model for interest rates". They used pca to derive a set of state variables, looks theoretically appealing, but when I tried to used the same set of data to calibrate for the mean reversion parameter, we do not reach the same results as produced in the paper. Actually, they used equation (4) to derive the mean reversion parameter. which I could not get any sense out of it.The paper is attached.