Various models don't yield all that great a result.
f(x) = ax + by
f(x) = a/x + by
f(x) = aln(x) + by
f(x) = a/ln(x) + by
Pretty much the same result. Some clue might be good. Check some first and second differences and see if you can find some sort of pattern.
It does look like a quadratic relationship to 'y' is better than a linear relationship. I didn't test that. You have to do some work.