When the Load history deflection predicts significantly larger deflection than a standard combinations times the creep factor, it's usually a matter of cracking. I use the Service Design - ECR plots to gauge the amount of predicted cracking, and in your model at many of the supports, the cracking results in approximately double curvature, over 50% cracked.
The other thing to watch out for are sections with no equilibrium. In your Load History Calc log there are dozens of these warnings. Here is the first:
Calculating Instantaneous Cross Section Load History Curvatures for Load Step Maximum Short Term Load Iteration 3 Cross section 437 has no equilibrium at (-251,9.199) (-257.6,9.199). Setting element stiffness to be very small.
So with all these fully cracked sections it's no surprise that the final load history results are so large. To improve the behavior, you might have to add some more user reinforcement, assuming there are no problems with the loads, load history steps or strip layout. As a test I increase the #5 @ 12 bottom bars to #6 at 12 and the max long term deflection dropped slightly to 3.0".