Hyperparameter Optimization for Deep Learning Modeling in Short-Term Load Forecasting

The evolution of new technologies has made short-term power load forecasting an essential part of the streamlining process in the management of power grid systems. Machine learning algorithms have been applied widely in this area but with little success towards achieving better accuracy rates. These gaps point out the necessity for better forecasting methods . This study is about the power grid system from Ho Chi Minh city in Vietnam. Ho Chi Minh operates as a metropolitan area on the rise with economic activity and seasonal factors greatly influencing electricity consumption. Due to its intricate fluctuations in consumption pattern, the city is known for having a high level of energy. This makes the citysuitable for an in-depth investigation regarding a case study on short-term load forecasting approaches. In this study, the goal is to evaluate the effectiveness of three hyperparameter optimization methods: Random Search, Grid Search, and Bayes Search. All these methods optimize the performance of Convolutional Neural Network (CNN) models for short-term electricity load forecasting in Ho Chi Minh City. The results obtained through this work can also be used as a basis for introducing the methods to other locations in Vietnam. The assessment of the techniques is performed using fundamental error measures such as Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Bayes Search completed with an MAEof77.93, MAPE of2.94%, MSE of 10,376.7, and RMSE of 101.9. These results indicate a noticeable enhancement in prediction accuracy when compared with the outcomes from Grid Search and Random Search. Grid Search provided an MAE of 106.23, MAPE of 3.95%, MSE of 17,033.7, and RMSE of 130.5. Random Search produces results of an MAE of 96.8, MAPE of 3.57%, MSE of 14,951.0, RMSE of 122.3. These results are evidence that Bayes Search is better for short-term electricity load forecasting in Ho Chi Minh City. The study also proposes an evaluation framework, which is meant for load forecasting in Vietnam. It is designed for Ho Chi Minh City predicting purposes, thus, integrating innovative concepts with actual forecasting functions. The framework is also applicable to other areas in Vietnam, both rural and urban, having different power consumption patterns. The reduction in forecasting inaccuracies through the use of Bayes Search is found to be promising as observed in the research. This automation supports better decision-making in energy management. It helps reduce costs in dynamic and complex power grid environments. These findings have practical value. They support efforts to build more flexible and efficient energy grids in Vietnam.