An IDBO-optimized CNN-BiLSTM model for load forecasting in regional integrated energy systems

• Published in: Computers & electrical engineering 2025-04, Vol.123, p.110013, Article 110013
• Main Authors: Su, Zhonge, Zheng, Guoqiang, Wang, Guodong, Hu, Miaosen, Kong, Lingrui
• Format: Article
• Language: English
• Subjects: Bidirectional Long Short-Term Memory; Convolutional neural network; Improved Dung Beetle Optimization algorithm; Integrated district energy systems; Multiple load forecasting
• ISSN: 0045-7906
• DOI: 10.1016/j.compeleceng.2024.110013

Accurate multi-energy load forecasting is a crucial prerequisite for ensuring integrated regional energy system planning and scheduling. Addressing the problem that hyperparameters are difficult to optimize when applying deep learning methods for load forecasting, the paper introduces a load forecasting approach utilizing the Improved Dung-Beetle Optimization (IDBO) algorithm to enhance the CNN-BiLSTM model and boost the forecasting accuracy. Firstly, the CNN-BiLSTM model is constructed by utilizing the powerful data space mining technique of CNN and the time series learning technique of BiLSTM. Secondly, the methods of Sobol sequence, curve adaptive strategy and cosine adaptive coefficient are used to improve the Dung Beetle Optimization (DBO) algorithm. Finally, a regional integrated energy load forecasting method based on the CNN-BiLSTM model optimized by IDBO is proposed, and relevant comparative experiments are conducted with other methods. The results indicate that the forecasting method proposed in this paper significantly improves prediction accuracy compared to other methods.