An Early Warning Model for Induced Draft Fan Based on Multivariate Time Series Forecasting and Dynamic Threshold

The induced draft (ID) fan is an important piece of auxiliary equipment in thermal power plants, directly impacting the safety and economy of the power production. Early detection of potential faults in the ID fan is crucial to guide predictive maintenance, reduce unscheduled shutdowns, and improve the reliability and availability of thermal power plants. This paper proposes a model for early warning of ID fan faults based on the multivariate time series (MTS) forecasting method using normal operating data of the fan. To generate early fault warnings for various operating conditions, the sliding window method was used to create a dynamic threshold for improved fault detection. The proposed method was applied to the actual operating data of an ID fan in a 300 MW thermal power plant, demonstrating high accuracy in forecasting normal operating data and the capability to detect potential faults in advance.