Optimization of transmission maintenance schedule using transformer load forecasting with modified hybrid CNN-LSTM and integer linear programming

Abstract Transformers are important parts for maintaining the reliability of electrical power systems. However, their maintenance requires power outages, which impact electricity sales. The maintenance is carried out based on operator readiness and does not take load conditions into account. The purpose of this study is to obtain an optimal transformer maintenance schedule based on load forecasting results. The forecasting method used a modified hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) for a one-day forecast. The modification was made to optimize the model in capturing the temporal and nonlinear characteristics of the load. The factors used were load, current, reactive power, temperature, tap position, and time features. The use of current factors that were highly correlated with load, combined with time features, could improve forecasting accuracy. Therefore, it was chosen as the proposed model. Then, the performance results were compared with the existing model, showing a root mean squared error (RMSE) value of 2.03 (an improvement of 4.69%), a mean absolute error (MAE) of 1.58 (an improvement of 7.06%), and a mean absolute percentage error (MAPE) of 13.6 (an improvement of 7.92%) for 24-h predictions. The forecasting results were integrated with an integer linear programming (ILP) model to determine the maintenance schedule with a load minimization objective function. Maintenance was limited to 8:00 to 5:00 for 4 h. The results showed that maintenance at minimum load could provide energy efficiency of 3.8% compared to existing conditions. Further research could test the model on extended data and explore other parameters.