Prediction of Power Transformer Oil Chromatography based on LSTM and RF Model
| Autor: | Gao Xue Lian, Abdul Wahid Mahrukh, Sohail Sajeel Bin |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
010302 applied physics
business.industry Dissolved gas analysis 0211 other engineering and technologies 02 engineering and technology 01 natural sciences Square (algebra) Random forest law.invention Variable (computer science) law Logic gate 0103 physical sciences Environmental science 021108 energy Real-time data Time series Transformer Process engineering business |
| Zdroj: | 2020 IEEE International Conference on High Voltage Engineering and Application (ICHVE). |
| DOI: | 10.1109/ichve49031.2020.9279968 |
| Popis: | The insulation strength of the power transformer depends on the oil-immersed of the transformer. The oil of the transformer becomes contaminated with moisture and the dissolved gases increase and the insulation gets weak with the time. Thus, it is necessary to predict the concentration of dissolved gases to avoid internal insulation failure of the power transformer. The behavior of dissolved gases is non-linear therefore, machine learning models is the best way out to predict the concentration of dissolved gases in the transformer. The proposed approach used AI algorithm to detect the inceptive failure of transformer early, using the time series Long Short-term Memory (LSTM) model and Random Forest algorithm. The approach uses online real time data that is acquire from a dissolved gas analysis (DGA) online monitoring system. The data is preprocess to obtain highly correlated variable using statistical correlation and regression square output. The time series model of random forest and LSTM model used to train the model, both techniques perform effectively during testing. Finally, comparison of two methods has been made, the results indicate the Random forest method captivate a better forecast the dissolved gases content in power transformer to avoid insulation failure. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|