Enhancing Short-Term Load Forecasting with Technical Indicators and Tree-structured Parzen Estimator

Short-term load forecasting in the power sector plays a crucial role in decision-making for power companies and enterprises, including economic dispatch, power scheduling, purchasing, and selling strategies. Therefore, the development of accurate short-term load forecasting models is of paramount importance, as it is closely related to grid stability and costs. In recent years, with the advancement of technology, machine learning techniques have been gradually adopted to address the nonlinear nature of load forecasting problems, leading to more accurate predictions. However, the features provided to machine learning models have become a critical factor affecting the accuracy of the model. Hence, we propose a novel approach that utilizes technical analysis indicators from the stock market as features for short-term load forecasting, resulting in improved performance. Additionally, while the inclusion of holidays as a feature has shown significant enhancements, there is a lack of research explaining the underlying improvements. Therefore, we compare the performance of including holidays as a feature. Furthermore, over time, the forecasting model needs to be updated. To address this, we propose the use of Tree-structured Parzen Estimation (TPE) to automate the hyperparameter tuning process, reducing the burden of manual tuning. Finally, we evaluate the proposed approach on two datasets, demonstrating the improvements achieved by our proposed methods.