Random factors of SITL in ArduPlane

Hi, I am doing some experiments in SITL (ArduPlane and MAVProxy) for fixed-wing plane, to calculate the sum of errors in Autopilot (e.g. the sum of angle errors and rate errors in roll, pitch and yaw controllers in the same time), but I found it that the sum of errors are often different each time (especially in the take-off phase). I have ingored the wind turbulence during simulation, but it seems that there are other random factors causing inconsistent results. Can anybody help me to find out the random factors, thanks!