EVENT: A Modular, Semi-Automated Methodology for Anomaly Detection and Event Characterization in Thermal Power Plant Operational Data with an Explicit Separation of the Analyst and Expert Roles

Operational data from thermal power plants are interpreted through expert judgment, following procedures that are often only partly documented and hard to reproduce. The literature offers many algorithmic techniques, yet workflow-level methodologies binding validation, detection, event creation and expert review into an auditable whole remain scarce. This article proposes EVENT (Expert eValuation of ENumerated Typed events), a semi-automated methodology for offline analysis of operational time-series data. It is organized as five steps—data validation, data preparation, anomaly detection, event creation and report generation—linked by machine-readable data contracts covering schematic, semantic, physical and temporal checks. Two design choices set it apart. The reporting unit is the event, not the isolated sample, and the analyst’s procedural role is kept distinct from the domain expert’s interpretive role. As a proof of concept, the method ran on a one-year anonymized dataset identified with the KKS coding standard. The contract checks caught all seventy constructed faults injected across eleven categories, and the pipeline ran end to end on the annual record. The workflow was additionally replicated on public wind-turbine data by three separately implemented instances. These results verify the workflow but are not a statistical validation of detection performance.