| As the key equipment in the power system,oil immersed transformer plays an important role in maintaining the safe and stable transmission of power.Condition monitoring and risk early warning of oil immersed transformers is not only an inevitable requirement to find potential safety hazards in time and prolong the service life of transformers,but also an inevitable move to promote the modernization of system and capacity for operation and maintenance.In recent years,with the wide application of data-driven technology in electric power industry,transformer monitoring and early warning based on thermal parameters has become a new research hotspot in the field of operation and maintenance.However,there are still some problems in the existing research results,such as inadequate mining of oil temperature original data,single index selection in the process of early warning and poor reliability of early warning.Therefore,this paper takes the oil-immersed transformer as the research object,carries on the accurate monitoring to its thermal state,and realizes the early fault warning to the oil-immersed transformer.The main research work is as follows:Firstly,based on the analysis of heat distribution and temperature rise law of oil-immersed transformer,an improved heat path model considering solar radiation is established based on hot spot analogy.Considering the influence of time series data decomposition,time step of model input and other factors on the improvement of prediction accuracy,an oil temperature combined prediction model based on CEEMDAN-SE-IPSO-LSTM is proposed.The proposed model and other comparison models are applied to the same data set,and the simulation results verify the superiority of this model in predicting oil temperature at a specific time.Secondly,aiming at the problem of error accumulation in predicting oil temperature at multiple times,a rolling prediction model of oil temperature based on similar day error correction is established.By means of grey relational analysis and cluster analysis,the similar day group of the forecast day was determined,and the similar day was determined based on the sum of Euclidean distance of oil temperature and ambient temperature at the same time scale.With the help of rolling prediction model,the similar days are modeled to obtain the prediction errors,and the errors are accumulated to the rolling prediction results of each moment corresponding to the date to be predicted as the final prediction results.The simulation results show that the proposed model can improve the poor performance of multi-time oil temperature prediction.Finally,two kinds of early warning models for oil immersed transformers are established from the perspective of risk early warning based on thermal state.Based on the difference of statistical characteristics of residuals under different sliding window lengths,a risk early warning model based on oil temperature residuals processing with double sliding windows is proposed.In addition,the residual information entropy and RMSE index are introduced into the risk early warning system.Through case analysis,the effectiveness of the two models in condition monitoring and fault early warning of oil immersed transformer is confirmed. |
