| Abstract:To address the problems of inconsistent limit scales, false alarms caused by single-point spikes, and difficulty in identifying future risk persistence in overtemperature early warning for the water wall, economizer, superheater, and reheater tubes of thermal power boilers, a four-tube overtemperature trajectory early warning method based on PatchTST quantile regression is proposed. The method constructs a safety margin using the 95th-percentile high-level wall temperature of tube-type measurement points, and introduces temperature-rise trend features, tube-type embedding, and PatchTST patch-based temporal encoding to characterize wall-temperature thermal inertia and response differences among heating surfaces. A softplus quantile monotonic structure is used to output 0.1, 0.5, and 0.9 quantile safety-margin trajectories, and a continuous over-limit area is constructed based on the 0.1 quantile trajectory to suppress false alarms caused by isolated over-limit points. Time-sequence replay verification is conducted using hot-state operation records of four-tube temperatures in a boiler. The results show that the proposed method achieves an RMSE of 7.03 ℃, an event recall rate of 92.1%, a false alarm rate of 5.8%, and an average lead time of 10.4 min, verifying its effectiveness for rolling early warning. |
点此下载全文
位访问者,今日访问
次