pvlib's fit_cec_sam is a wrapper for the SAM SDK's parameter estimator. When it fails, pvlib can't do much besides inform you of the failure. I suspect the problem here is numerical overflow when evaluating the exponential term with the scaled voltage and current.

Instead, I would get the parameters for a single module, then scale them to "harness level".

• multiply photocurrent (I_L_ref) and dark current (I_o_ref) by n_strings_per_harness

• multiply series resistance (R_s) and the diode voltage (a_ref) by n_panels_per_string

• scale the shunt resistance (R_sh_ref) by n_panels_per_string / n_strings_per_harness

You can work out these scaling rules by substituting V= V' / Npanels and I = I' / Nstrings in the single diode equation, where V' and I' are the harness DC voltage and current.

For the temperature coefficients, its a similar exercise in algebra to arrive at

alpha_isc' = alpha_isc * Npanels

beta_voc' = beta_voc * Nstrings