Hot Spot Temperature Prediction Models Including Loss Of Life Of Transformers Using Asaba Transmission SubStation As A Case Study

This study is on hot spot temperature (HST) prediction models, including loss of life (LoL) of transformers, and uses the Asaba transmission substation as a case study. One year loading profile data was collected from the substation while Artificial Neural Network (ANN) and Bagging Regression (BR) algorithms were modeled using this data on MATLAB and Python to predict the HST for Day-Ahead (24-Hours), Week-Ahead (120-Hours), and Weekend (48-Hours). The results show the nonlinear relation between Load and HST, as an increase in load values from 305.8A – 328.9A resulted in an increase in HST from 52.23oC – 54.88oC. Model performance was compared using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE). Using MATLAB indicated ANN had MAPE values of 2.51, 2.74, and 2.48 with RMSE values of 1.83, 1.71, and 1.85, respectively, for Day-Ahead, Week-Ahead, and Weekend prediction. In contrast, BR had MAPE values of 5.93, 5.83, and 5.77 with RMSE values of 4.30,4.86, and 4,43, respectively, for the same prediction period. Using Python indicated ANN had MAPE values of 6.67, 7.29, and 6.96 with RMSE values of 3.34, 4.27, and 1.95, respectively, for Day-Ahead, Week-Ahead, and Weekend prediction, while BR had MAPE values of 6.16, 6.07, and 6.5 with RMSE values of 5.7, 5.85 and 4.5 respectively for same prediction period. MATLAB software gave better results than Python, which had a higher percentage error for the ANN and BR models, likely due to the model's ability to better capture consistent thermal responses during sustained high loads. LoL for a 24-hour load cycle gave a cumulative age hour of 0.029959 hours, revealing how the HST significantly impacts the transformer’s insulation and life expectancy. It is recommended that different modeling techniques be explored for prediction analysis for medium-term transformer HST prediction if better results are desired. Also, operators should incorporate monitoring systems like Arduino to shut down the network and protect the transformer anytime the temperature exceeds set limits. This study presents the following contribution to knowledge: Data bank of a 60 MVA transformer can be obtained; Correlation between HST and Load demand has been established.