Condition monitoring and early fault warning of power plant auxiliary equipment using LSTM-SDAE model with an adaptive threshold

Abstract

With the increasing penetration of renewable energy in the power grid, which makes power plant equipment is always in changing operating conditions. The correlation between the main and auxiliary equipment of the unit is easy to lead a potential fault, therefore, the safety and reliability of the auxiliary equipment of thermal power units have become a more challenging issue in the case of large-scale renewable energy. Adaptive condition monitoring of the auxiliary equipment can reduce maintenance costs and improve reliability in the thermal power units. Most existing studies perform poorly at extracting features from distributed control systems data and make less use of time series data. A novel adaptive condition monitoring framework and early fault warning method based on long short-term memory and stack denoising auto-encoder network has been proposed for auxiliary equipment of power plant unit. The proposed framework has two main parts, which contain condition monitoring and adaptive early fault warning. The Mahalanobis distance of a reconstruction error is defined as the monitoring indicator to reflect the condition of the equipment. The Chebyshev inequality determines an adaptive threshold for early anomaly detection that applies to changeable working conditions. The effectiveness of the proposed method was verified by the actual case of the coal mill. The adaptive threshold method can obtain the advance time of 42s and 108s, respectively.