Self‐Training Partial Least Squares Model for Semi‐Supervised Multivariate Spectroscopic Calibration With Application to Power Grid Inspection Order Accuracy Assessment

ABSTRACT Power grid marketing inspection depends on work order data to detect rule violations, yet large‐scale manual annotation is expensive. This paper proposes a self‐training partial least squares (STPLS) model that exploits both labelled and unlabelled samples for semi‐supervised calibration. High‐confidence pseudo‐labels are iteratively incorporated to improve model reliability under limited supervision. Experiments on more than 10,000 real‐world grid work orders show strong performance in audit‐topic alignment and responsibility attribution, while tests on a benchmark wheat NIR dataset demonstrate cross‐domain generalizability. The proposed framework provides an efficient and scalable solution for intelligent inspection in power grid applications.