A Simulated Current–Voltage Data Set of a Photovoltaic Array Under Compound Electrical Faults and Partial Shading: Generation, Validation and Reproducibility

Learning-based fault diagnosis of photovoltaic (PV) arrays needs labelled current–voltage (I–V ) curves that cover faults, partial shading and their combinations. Such curves are usually simulated, but the curves, the code and the evidence that the simulation is correct are seldom published, so results cannot be compared or reproduced. This paper describes PVCF-4x4, an open data set of 96 000 simulated I–V curves of a 4×4 array of 60-cell modules with three bypass diodes each, given with and without string blocking diodes. Each curve has 331 points of the array current and of the four string currents. Every scenario carries a multi-label target for partial shading, short circuit, open circuit and degradation, with all 16 combinations equally represented, and the drawn irradiance, temperature, shading pattern, fault type, location and resistance. The curves come from a nodal Newton solver of the same circuit as a Simscape Electrical model. The two agree to within 0.33 mA (0.001% of the rated current) on 320 independent scenarios, and the solver agrees with an explicit solution of the healthy array to 4 × 10-8 A. Module ratings, temperature coefficients, the numerical health of all curves and the fault signatures predicted by circuit theory were also checked. Every scenario has its own random seed, and 400 scenarios regenerated on a single process were identical to the stored ones. The data, an official split, baseline results and the MATLAB code are released under open licences.