Prediction Of Dissolved Gas Concentration In Transformer Oil Considering Various Factors

As one of the crucial equipment in the operation of the power grid,the working status of power transformers is closely related to the security and stable operation of the power system.Once a fault occurs,it will cause massive losses to the safety of social electricity consumption and economic construction.Therefore,the research on transformer fault diagnosis technology is of immense significance.Transformer fault diagnosis often uses dissolved gas analysis methods in oil.With the popularity of smart grids and the development of online monitoring technology,it is possible to accurately predict the dissolved gas in transformer oil using big data.In this paper,XGBoost algorithm is used for the first time to predict the concentration of dissolved gases in oil,and the content of dissolved gases in oil is predicted considering external environmental factors and the situation of transformers with different voltage levels.First of all,this paper proposes an improved XGBoost model,which verifies the feasibility and reliability of XGBoost’s prediction of gas concentration.At the same time,in order to make up for the shortcomings of the original model,the Isolation Forest algorithm is used to amend the gas concentration data,the polynomial fitting method is used to de-trend the gas time series,feature selection is carried out to eliminate redundant features and add important features that are missing in the prediction,Bayesian Optimization algorithm is applied to find the best combination of parameters of the model.According to the prediction results,the prediction accuracy of XGBoost is continuously improved during the above optimization procedure,and the error is effectively reduced.Finally,compared with the original XGBoost model,the improved XGBoost model has a reduction rate of root mean square error between 13.2%and 91.4%,and absolute percentage error between 20.1%and 97.5%for the prediction of gas concentrations.The model fitting effect is significantly upgraded,which verifies the effectiveness of the method in this paper.Secondly,in order to compensate for the lack of considering the impact of external environmental parameters on the prediction of dissolved gases in oil,the paper works to obtain multi-dimensional input variables based on data of dissolved gases in transformer oil from four different geographical locations,combined with local daily average temperature,daily average relative humidity,daily cumulative precipitation,and daily average atmospheric pressure.To observe the impact of meteorological data on gas concentration prediction,and infer the relationship between environmental factors and gases based on experimental results.Research has shown that methane,hydrogen,and carbon monoxide are more susceptible to environmental data,and the input of meteorological data can effectively improve the accuracy of their prediction,while ethane,ethylene,and carbon dioxide cannot reduce errors through meteorological data.Finally,the paper explores the prediction rules of dissolved gases in transformer oil with different voltage levels.Based on real-time sampling data from two 110kV transformers,three 220kV transformers,three 500kV transformers,and two 1000kV transformers,the prediction effect of applying the improved XGBoost model on gas concentration will vary depending on the voltage level,and the optimal model prediction accuracy for different types of gases is also closely associated to the transformer level.However,when predicting gas concentration for 1000kV transformers,the error is invariably relatively large.The research results of this paper provide a useful reference for the prediction of dissolved gases in the oil of different grades of transformers.