I haven’t tested yet to see how (or in what circumstances) it might improve performance, but depending on your dialect you can wrap the concatenation in a hashing function (like MD5()) and get a virtually-guaranteed-to-be-unique-in-most-applications ...

I’d also consider trimming off any time periods with incomplete measures (or at least giving users a toggle filter so they can do so). The “hooks” at the ends of the various trend lines can be misleading.