| Abstract |
Accurate load forecasting is crucial in ensuring efficient, reliable, and cost-effective planning and execution of the power grids. The uncertain characteristic of consumers as well as the increasing distributed generation units makes precise load forecasting an increasingly challenging task. Additionally, due to the complex nonlinear relationship of electric load with weather factors, advanced load forecasting models are necessary to accommodate the fast-changing nature of power grids. This research proposes an improved model for short-term load fore-casting using deep learning. Important factors like electricity price, dew point, dry bulb, wet bulb, humidity, day of month, day of week, year, and time of day will be included in the model. First, a deep neural network is implemented with default hyperparameters. Then, hyperparameter optimization is applied, during which the model searches for the optimal combination of hyperparameters within the defined range. Finally, the model is run with the optimized hyperparameter values to predict the load. The proposed method is implemented on the Electrical Reliability Council of Texas system to validate its accuracy. The results show a significant improvement in system performance after hyperparameter optimization while considering diverse impact factors. |
, Nga Nguyen 
