I don’t think it’s smoothing-related because the graph itself isn’t smoothed as far as I know. The power data recorded is just raw power coming from your power meter (unless you have a smart trainer that does smoothing itself). In endless discussions on the vector 3 forums Garmin has claimed that spikes can be a result of the receiver not getting the data for a second or two and then playing catch-up. No idea if that’s true but if it is, I suspect that TR’s interval changes cause a slight delay in processing power data.
But to be honest, I also wouldn’t preclude the possibility that my brain is playing tricks on me and my legs are actually anticipating the interval change.