Thermal Modelling for Ampacity Prediction of 132 kV Directly Buried Aluminium Cable Systems

This study develops a reduced IEC 60287-based thermal model and an IEC-informed neural-network approximation for ampacity prediction in a 132 kV directly buried aluminium cable reference system. The analytical implementation is calibrated at a single CYMCAP baseline condition for a 132 kV, 1000 mm2 aluminium, XLPE-insulated and HDPE-sheathed cable, for which CYMCAP gives 799.486 A. The numerical design is a deterministic full-factorial Cartesian grid over five variables: soil thermal resistivity, burial depth, ambient soil temperature, allowable conductor temperature, and a bounded screen-loss scenario factor. This design yields 221,760 parameter combinations spanning the prescribed installation and thermal ranges. The model-development run underlying the reported accuracy metrics used 16,000 fitting cases and a 4000-case hold-out interpolation set. Against the reduced IEC 60287 targets, the neural model achieved a mean absolute percentage error of 0.141%, a root-mean-square error of 1.008 A, and R2 = 0.9999. The CYMCAP comparison is explicitly limited to the single calibration point and is not presented as independent multi-point CYMCAP validation. Sensitivity results for soil thermal resistivity and burial depth reproduce the expected monotonic trends, but burial-depth results are interpreted as sensitivities of the calibrated reduced model because the empirical mutual-heating factor is held fixed. The study therefore demonstrates high-fidelity approximation of the stated reduced IEC 60287 function within the sampled domain, rather than universal cable-system verification. The historical subset-selection seed was not retained, so exact reproduction of the neural-network run is not possible.