Adaptive Threshold Based Outlier Detection For Data Of Oxygen System In Metallurgical Industry

The accuracy of the data such as consumption of oxygen in the process of smelting acted an important role for production scheduling and optimal operation. But the data collected is often abnormal due to the instability of industrial data acquisition system and the underlying communication equipment are susceptible to interference and other reasons. Abnormal data will allow for personnel judgment misleading and may cause major accidents and economic losses in severe cases, causing the disturbance of system reliability. For this kind of situation, the workers tend to rely on experience to determine abnormal data, and a lot of huge amounts of data in the form of make artificial judgment difficult. So anomaly detection of the consumption of oxygen smelting process data is of great significance and research value.This paper proposes an outlier detection based on adaptive threshold method, the method firstly bases on K nearest neighbor based adaptive fuzzy c-means clustering algorithm. This method outputs fuzzy membership degree, fitness vector and clustering center. Then according to the fitness vector for each data point to be detected abnormal index and each index constitute the anomaly factor sequence. Then use the above factor sequence for data anomaly detection, concrete steps as follows:1) for all the data points to be detected respectively, using a sliding window to detect delimitation of their respective neighborhood;2) calculate the mean and standard deviation within the scope of the neighborhood anomaly index of all points;3) determine corresponding threshold for each point using the mean and standard deviation of the point to be detected, and compare the relationship between the index and the corresponding anomaly threshold to determine whether the measuring point is an outlier or not.In order to verify the effectiveness of the proposed method in this paper, we use a domestic iron and steel enterprise oxygen consumption of blast furnace and converter oxygen consumption data in the system as the research object; the data that exist in the abnormal situation for outlier detection, the results demonstrate the feasibility and effectiveness of the method. And it’s verification for optimal scheduling of oxygen system provides a scientific and effective guarantee.