Pattern recognition and similarity scoring for time series data

I have a project where I am collecting motion data from IMU sensors (accelerometers, etc). I collect short spans of ~250 or so data points from the sensor, forming a time series.

Given 10 or so time series, I want to make a classifier so that I can take the 11th time series and give it a similarity score to the previous 10. Ideally this similarity score ranges from 0 to 100: this is the most important part.

How should I go about doing this? So far I have tried dynamic time warping, which gives good relative similarity scores for my time series, but the result I find incredibly hard to scale to the 0 - 100 range. E.g. for my "best" matches, my DTW score is 30 - 50. For my worst, it's 400+. And for complete fails it's 800+. But then for a different set of 10 time series, "good" will mean < 10, and "bad" will mean > 40!

Hopefully I've made my problem clear. Basically I'd love a way to score similarity of novel time series to a pre-saved set, and have that score scale out of 100. I've tried DTW for this purpose which is good for relative similarity, but I can't figure out how to do the scaling I want.

Thanks for any help.