I am working on ultra high performance concrete, I would like to know which prediction model is more suitable for concrete in terms of compressive strength
Work by Benjamin Graybeal has shown that the compressive strength is dependent on the curing temperature and curing days in the following way.
f'c,t = f'c,28d x [1 - exp[-((t - tstart)/a)b]]
where the values of f'c,28d, "a" and "b" are dependent on the thermal treatment and curing age. T is the time in days and tstart is the time of strength gain initiation. See link for more information
There are lots of other parameters you can consider to build a model. You mentioned you are working on UHPC, the fiber content, aspect ratio ( fiber reinforced index) and other factors of plain concrete (i.e. w/c ratio, density of hardened concrete) aggregate size. You can use all these to form an empirical model to predict the f'c of your UHPC. Also, you may like to try other modelling method ( ANN, GP)
Try this paper done by my supervisor, may help you a bit.
Stress-Strain Model for Normal- and Light-Weight Concretes under Uniaxial and Triaxial Compression." Construction and Building Materials.71: 492-509
The nonlinear regression might be a good method if your specimens are not a lot. However, if your specimens become large in number, the regression model will not answer to them. Then you can use the Artificial Neural Network Method for predicting your compressive strength. I have recently published a paper exactly in the same area of the question you are asking. You can take a look at it and I hope it can help you.
In addition, if you have the ultrasonic pulse velocity of the concrete, you can predict the compressive strength using the exponential formula, as ultrasonic pulse velocity and compressive strength of concrete have exponential relationship.