Multi-physics coupling calculation has applications in many important research fields. If particle transport process is included in this calculation, Monte Carlo method is often used to simulate this process and usually a large amount of calculation time is needed. So, efficient Monte Carlo algorithm for time-dependent particle transport problem is important for an efficiently coupling calculation, which inevitably relies on large-scale parallel calculation. Based on the characteristic of time-dependent particle transport problem, two methods are proposed in this paper to achieve high- efficiency calculation. One is a tally-reducing algorithm which is used in the coupling of transport simulation and burnup calculation. By reducing the quantity of data which should be reduced necessarily, this method can reduce the calculation time largely. It can be seen that a new coupling mode for these two processes in MPI environment has a larger value when model scale is larger than the sample size. The other method is an adaptive method of setting the sample size of Monte Carlo simulation. The law of large number assures that the Monte Carlo method will obtain an exact solution when the sample scale tends to infinity. But generally, no one knows which sample scale is big enough for obtaining a solution with target precision in advance. So, the common strategy is to set a huge-enough sample scale by experience and conduct the posterior check for all results. Apparently, this way cannot be efficient because the calculation will go on after the precision of solution has reached an object value. Another popular method is to set the sample size to rely on the relative error of some single calculation. The sample size is enlarged without a break until the relative error is less than some presetting value. This method is not suitable either, because Monte Carlo particle transport simulation will gives feedbacks to other process which is composed of many tallies. It is inappropriate to adjust the sample size according to the relative error of any calculation. Relying on the generalization of the Shannon entropy concept and an on-the-fly diagnosis rule for a entropy value sequence, the adaptive method proposed in this paper can reduce the original huge sample scale to a reasonable level. By numerically testing some non-trivial examples, both algorithms can reduce the calculation time largely, with the results kept almost unchanged, so the efficiency is high in these cases.