A Stacking-Based Ensemble Model for Short-Term Load Forecasting

Short-term load forecasting plays an important and indispensable role in the daily operation planning of power grid because it allows grid operators to predict electricity demand a few hours to one week in advance. Although statistics-based methods and machine learning-based methods have been widely used in short-term load forecasting, a single model may have difficulty capturing all underlying dynamics, causing reduced prediction accuracy. Therefore, a stacking-based ensemble model that improves prediction accuracy by integrating multiple base prediction models is proposed in this study for short-term load forecasting. Firstly, for data preprocessing, data normalization is used to scale the raw load data to a range of 0 to 1. Data imputation is used to ensure data integrity. Secondly, base prediction models including logistic regression, decision tree, random forest, multilayer perceptron, convolutional neural network, and long short-term memory are utilized to train the prediction models. Thirdly, the stacking-based ensemble learning method is utilized to integrate these base prediction models to further predict electric load. The results of comparative experiments and error analysis show that the stacking-based ensemble learning model outperforms the base prediction models for the majority of the evaluation metrics. Additionally, the analysis of curve fitting results demonstrates the high level of agreement between the actual values and the predicted values for the stacking-based ensemble learning model.