Hi @HUI_HUI, welcome!
How many times is this pattern shown above repeated?
It looks highly periodic, so it shouldn’t take many repetitions of the pattern for TM to learn it.
If the anomaly scores haven’t settled down after a few repeats I’d look at the encoding parameters, and make sure they make sense given the distribution of your values. For example, if the values for one metric tend to fall around 80-120 mostly, it wouldn’t make sense if the encoder min & max were 0 & 1.
So first thing I’d do is ensure that the pattern repeats several times for TM to learn, and then make sure the encoding params match the raw data distribution. In my experience, the problem is most likely to be one of these rather than any TM parameter.