Abstract:Outlier detection is an important part of data processing in thermal process, and is also the basis for system modeling, optimization and control. Aiming at the problem that the operational condition of the thermal process changes frequently, which causes the difficulty of outlier detection, this paper proposes a thermal process outlier detection method combining signal decomposition method and densitybased detection method. Firstly, the empirical wavelet transform method is used to extract the operational trend of the thermal process time series. After removing the sequence operational trend, the local outlier factor method is used to obtain the local outlier values for the data points. Finally, the box plot method is used to determine the sequence outlier points. The load data of the 1000MW unit in a certain power plant was used as the experiment data, five errors of 05%, 1%, 2%, 5% and 10% were set respectively to verify the effectiveness of the method. The experiment results show that besides having applicability to both dynamic and steady state processes, the outlier detection method proposed in this paper achieves high detection accuracy under the above five error conditions.