Short-Term Load Forecasting Based on Multi-Scale Ensemble Deep Learning Neural Network

• Published in: IEEE access Vol. 11; pp. 111963 - 111975
• Main Authors: Shen, Qin, Mo, Li, Liu, Guanjun, Zhou, Jianzhong, Zhang, Yongchuan, Ren, Pinan
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Computational modeling; Data mining; Deep learning; Differential equations; differential evolution algorithm; Electrical loads; Ensemble learning; ensemble model; Evolutionary computation; Forecasting; Load forecasting; Load modeling; Machine learning; Mathematical models; Model accuracy; Neural networks; Power; Predictions; Predictive models; Short-term load forecasting; Stochastic processes; stochastic weight averaging
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2023.3322167

High-precision load forecasting is crucial for the power system planning and electricity market transactions. Recently, deep learning models have been widely used due to their powerful data mining capabilities. However, the existing research mainly focus on model structure adjustment and input feature selection, ignoring the influence of model ensemble on prediction. A single deep learning model is not yet able to address the various complex challenges that arise in short-term load forecasting. To improve the accuracy of short-term load forecasting, this paper proposes a novel multi-scale ensemble method and multi-scale ensemble neural network. This neural network uses long short-term memory, gate recurrent units, and temporal convolutional network as the basic model. By coupling the stochastic weight averaging ensemble method and differential evolution ensemble method, these deep learning networks were assembled from single-model scale and multi-model scale, respectively, thereby effectively improving the model prediction accuracy. For predicting the power load of Hubei Province in China, meteorological features and time features were in consideration. The proposed model was trained and compared with eleven intelligent short-term load forecasting models, including machine learning, deep learning and ensemble deep learning models. Simulations show that the proposed model has the best comprehensive prediction performance. This study highlights the power of ensemble deep learning models coupled with multiple ensemble techniques and the promising prospect of our proposed model in short-term load forecasting.