You estimate power the same way as for any other study. The effect of interest is for an interaction which created as a product term of two other variables. So you if you have the effect size then it is straight-forward.

Of course in practice the utility of the calculation depends entirely on how good the effect size estimate is. Generally it makes sense to calculate power for the smallest effect you'd like to be able to detect (e.g., the smallest effect of clinical or practical utility) unless you have a quantitative model of the size of effect. Another alternative is to use existing data (but that can be very unreliable if you use pilot data because the samples are too small to get accurate estimates). Finally, you might be tempted to use t shirt effect sizes (small, medium, large). This is generally a  bad idea but particularly so for moderator analyses as the effects tend to be smaller than for main effects - because power for moderator analyses is highest when data are drawn from the extremes of the distributions of the original variables. If both variables are, say, approximately normal then this guarantees pretty low power.

Simulation is useful because you can model the impact of a range of circumstances (distribution, correlation between predictors, sample size and so on) on the power.

Richard, the threshold rules of thumb such as N = 10k will rarely be sensible (well, even less sensible than usual) for moderator effects. Tests of moderator effects are generally lower power than other effects. This could be built into a power analysis through reducing the estimate of the standardised effect size, but a rule based only on predictors will fail.

Hi Thom, thanks for your detailed response. I am not familiar with using simulation to calculate the effect size, do you mind recommending some papers or tutorial to me? Thanks again for your kind response!

Hi Richard, thanks for attaching the paper and providing your response! I will download and read it through.

You can use statistical calculators available online, as for example in this website: https://www.danielsoper.com/statcalc/related.aspx?id=16

This is very helpful.

Thanks for this usefull tool!