Prediction of Transformer Oil Temperature with CEEMDAN Decomposition and SSA Optimization

The distribution of electrical energy is one of the problems that power companies must deal with on a daily basis. Power distribution requires accurate prediction of future demand, but it is affected by many factors such as the environment, which can lead to waste of resources and depreciation of equipment. Transformer oil temperature is a subset of the key guidelines reflecting the operational status of power transformers, for this reason, this thesis proposes a deep learning based time series prediction method for oil temperature. In this paper, CEEMDAN is used to preprocess the data for the ETTh2 dataset, and the TCN-BiLSTM-Multi-Head Self-Attention model is optimized using the Sparrow Search Algorithm. The experiments are validated on the ETTh2 dataset and evaluated using four metrics: MSE, RMSE, MAE and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex>. Derived from the experimental findings, it was shown that the model introduced in this thesis exhibits higher accuracy and stability on the transformer oil temperature prediction task compared to traditional deep learning methods, providing an effective solution for the time series prediction task.