Research On Power Transformer Temperature Prediction Based On ARIMA-LSTM Model

In recent years,with the continuous development of the computer field,the role of machine learning is increasingly prominent in various fields.As an important field in production and life,electric power resources are closely related to social development and people’s life,so it is of great significance to study the field of electric power by means of machine learning.At the same time,technological developments in the field of electricity will greatly promote progress in other fields.The role of the transformer is multifaceted-not only to increase the voltage to the supply area,but also to reduce the voltage for all levels of use to meet the demand for electricity.In short,the transformer must be able to raise and lower voltage.When power is transmitted,there will inevitably be voltage and power loss in the power system.When the same power is transmitted,the voltage loss is inversely proportional to the voltage,and the power loss is inversely proportional to the square of the voltage.The use of transformers can increase voltage and reduce power loss.Researches on deep learning in power mainly focus on power consumption and the prediction trend of future power consumption.However,due to the large number of indicators of transformer data and the linear and nonlinear parts of data itself,the prediction analysis of data only using traditional models has certain limitations.Deep learning provides an effective supplement for this purpose.In this thesis,ARIMA-LSTM fusion model is proposed and applied to power transformer prediction.Power transformer temperature(ETT)is an important indicator of long term deployment of power.To explore the granularity of the long series Time Series prediction(LSTF)problem,different subsets were created,{ETTh1,ETTh2} for the 1 hour level and ETTm1 for the 15 minute level.ARIMA-LSTM fusion model was used for prediction.The training set data was input into the ARIMA model to obtain the oil temperature prediction data.By comparing oil temperature prediction data and linear fitting data,the fitting error sequence was obtained.Combined with the power index,the error sequence fitting was realized according to the LSTM model prediction.The final power prediction result of the combined model is obtained by series model prediction.The weight of prediction error is determined by entropy method according to the set evaluation index,and the final error value of the model is calculated according to the weight.At the same time,the prediction accuracy of other traditional models is calculated and compared with the fusion model,and the application prospect and advantages of the fusion model are further expounded.The results show that the ARIMA-LSTM model has advantages in power data prediction,which can effectively reduce model errors and improve model accuracy,and can be used as a model reference for the same type of data.At the same time,the simplification of indicators by grey correlation analysis is conducive to the simplification of model modeling,which is of great significance for processing large data sets and complex models.Transformer is an important equipment for electric power transportation,and oil temperature is an important index of transformer efficiency.Real-time monitoring should be done,periodic modeling and analysis should be done,and power development should be planned efficiently.