Deciding by RMSE how many simulation results to use for a representative sample

So, I'm doing a brute force Monte Carlo simulation to vary certain parameters and running a model through a thermal computational test bed. At the end I get temperatures for every hour. We have field data of the actual temperatures, so I do a RMSE to decide which parameters best fit the actual data. I'm not getting really strong trends. So what I've been doing is just taking the average of the top 10. This feels pretty sloppy and unscientific to me, and by that I mean it is. I really don't know much about statistics so I'm wondering if there is a better way to fundamentally have a cut off line by something like the best RMSE to where it is a certain amount higher? That is just one idea I had, but I'm a bit unsure of where to even start. Thank you in advance!