Study on Reliability of Online Monitoring System for Transformer Oil Chromatogram Based on Machine Learning

Online oil chromatograph monitoring devices, as an effective access for monitoring transformers, have been widely applied in recent years. However, effected by electromagnetic fields, mechanical vibration, temperature, working time and so on, the measurement results of the oil chromatogram devices will exist obvious errors. For operation and maintenance engineers, accurate measurement results are extremely important because it can help engineers to judge the operation states of transformers. Therefore, it is very important to evaluate the reliability of the online oil chromatograph monitoring devices. Based on machine learning algorithm and semi supervised learning strategy, this paper proposes a novel online oil chromatography equipment reliability monitoring method to achieve real-time monitoring of online oil chromatography equipment reliability. Because of the high cost of data acquisition, this paper generates the data set based on Latin hypercube sampling (LHS), picks up the environmental variables that may affect the reliability of oil chromatography equipment and the monitoring value of oil chromatography equipment as the input variables, and the percentage error between the gas monitoring value of oil chromatography equipment and the laboratory calibration value as the output, and expands the data set through self-training methods. Based on the expanded data set, the random forest algorithm is used for modeling, and the percentage error of online oil chromatography equipment is output by inputting realtime parameters of the environment and online data of oil chromatography. The proposed approach is verified through test dataset. The result shows the proposed approach’s advantage of high efficiency and accuracy.