TCR-Net: Context-consistency risk grading for pre-transmission power inspection files

Digital power-operation workflows increasingly rely on team gateways that screen inspection files before they enter shared workspaces. In renewable stations, substations, and transmission-inspection tasks, an unsafe release may expose equipment status or work-plan information, whereas excessive blocking can delay maintenance. Existing format and policy checks detect explicit violations, but they are weak at identifying files that are syntactically valid yet inconsistent with their task stage, operator role, destination, or recent behavior. We formulate this as event-level risk grading into low, medium, and high risk, mapped to release, review, and block. The Temporal-Consistency-and-Rule Network (TCR-Net) represents each event through file structure, semantics, operating context, and short-term history and fuses a semantic-consistency score, a temporal-deviation score, and explicit rule-conflict evidence into a graded decision calibrated on validation data. It also reports component scores and triggered rule items with each decision for audit. On a 6144-event dataset generated from modeled grid operational logic, TCR-Net achieves macro-averaged F1 score (Macro-F1) =0.869±0.003, high-risk recall =0.696±0.034, and false alarm rate =0.002±0.002 over five training seeds. In the main test set, the fixed-parameter confirmatory experiment, and cross-scenario validation over three grid-operation profiles, TCR-Net consistently outperforms strong rule-aware and sequence-aware comparison methods. Code is available at https://anonymous.4open.science/r/TCR-Net-612B.