Power transformer is the key equipment in power system. The oil temperature is an important factor affecting the service life and load capacity of oil immersed power transformer. Applying big data technology to oil temperature prediction system can more reliably and efficiently evaluate the thermal characteristics of transformer. Therefore, based on Hadoop + spark big data platform and its ecological components, this paper uses convolutional neural networks (CNN) and other algorithms to compare, evaluate and predict the long-time series data of oil temperature at the top of power transformer. The final results show that: CNN algorithm model built in this paper has higher prediction accuracy than other related algorithm models. For quarterly short time series prediction, the average value of root mean square error (RMSE) of the CNN model is 0.907 and the average mean absolute percentage error (MAPE) value is 2.8%. The platform can accurately predict the oil temperature of transformer combined with CNN algorithm. This paper can provide a certain reference value for power grid big data analysis and stable operation of power equipment.

Spark; Oil Temperature;CNN Algorithms；;Prediction；;Power Transformer