Abstract: In the process of perceiving abnormal operating states of equipment in thermal power plants, if sensors are arranged at the historical fault locations of the equipment and assigned the same weight, the threshold for abnormal scores will appear single due to the influence of different equipment perception quantities, thereby reducing the adaptability to different equipment abnormal perception quantities, increasing the relative error of state perception, and ultimately affecting the operating quality of thermal power plant equipment. Therefore, a method for perceiving abnormal operating states of thermal power plant equipment based on the isolation forest algorithm was designed. Deploy perception nodes on various devices to analyze the relevant perception quantities of similar device groups, and determine the weight vector for abnormal operation status perception of thermal power plant equipment based on the status values of similar device groups. In the abnormal device window, create an initial sliding window that stores the first data received in the data stream, use the isolation forest algorithm to randomly select an abnormal feature and a normal feature, and use the abnormal score threshold to perceive the abnormal score of the device operating state vector, thereby achieving accurate perception of the abnormal operating state of thermal power plant equipment. The final perception result shows that the relative error of abnormal operation status perception of thermal power plant equipment is within ±0.01%, and the perception accuracy is high, which plays an important role in improving the operation quality of thermal power plant equipment.