String-Conditioned Multi-Edge Causality for Early Detection of PV DC Optimizer Anomalies Using DC Telemetry Only

DC optimisers used for maximum power tracking in medium- to large-scale PV systems often exhibit a decoupling-loss/pass-through tendency, in which an optimizer's output voltage becomes unduly driven by its own module-side voltage or abnormally decoupled rather than reflecting appropriate conversion behaviour. This work proposes a DC-telemetry-only detection pipeline that isolates device-specific effects from string-level constraint propagation using string-conditioned multi-edge causality. Data from a real site was investigated. We segment data to active regimes via a robust current-deviation mask, then test device-specific causality with a string-conditioned nested model. We quantify device vs. string influence using dominance factors and device loss evidence score (DLES). The scaled through candidates are subjected to Benjamini-Hochberg False Discovery Rate to detect false positives when testing several devices. This approach delineates the effects of environmental conditions, actual PV module performance and is driven by global performance observations. Results are reported as (a) an operator-ready shortlist, (b) a full dominance decomposition per device, and (c) directionality sanity checks. The approach is designed for digital twin-enabled field deployment where only DC telemetry is available.