A hybrid deep learning-based online energy management scheme for industrial microgrid. (15th December 2021)

• Record Type: Journal Article
• Authors: Lu, Renzhi; Bai, Ruichang; Ding, Yuemin; Wei, Min; Jiang, Junhui; Sun, Mingyang; Xiao, Feng; Zhang, Hai-Tao
• Is Part Of: Applied energy. Volume 304(2021)
• Publication Date: 2021-12-15
• Subjects: Online energy management -- Demand response -- Industrial microgrid -- Deep learning -- Convolutional neural network -- Long short-term memory; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2021.117857
• Languages: English
• ISSNs: 0306-2619

Abstract

Abstract: The fluctuations in electricity prices and intermittency of renewable energy systems necessitate the adoption of online energy management schemes in industrial microgrids. However, it is challenging to design effective and optimal online rolling horizon energy management strategies that can deliver assured optimality, subject to the uncertainties of volatile electricity prices and stochastic renewable resources. This paper presents an adaptable online energy management scheme for industrial microgrids that minimizes electricity costs while meeting production requirements by repeatedly solving an optimization problem over a moving control window, taking advantage of forecasted future prices and renewable energy profiles implemented by a hybrid deep learning model. The predicted values over the control horizon are assumed to be uncertain, and a multivariate Gaussian distribution is used to handle the variations in electricity prices and renewable resources around their predicted nominal values. Simulation results under different scenarios using real-world data verify the effectiveness of the proposed online energy management scheme, assessed by the corresponding gaps with respect to several selected benchmark strategies and the ideal boundaries of the best and worst known solutions. Furthermore, the robustness of the scheme is verified by considering severe errors in forecasted electricity prices and renewable profiles. Highlights: Propose an adaptable online energy management scheme for industrial microgrids. A hybrid deep learning model is developed to forecast future unknown information. The optimization problem is solved in real time over a moving control window. Prediction errors are included to verify the robustness of the proposed scheme. Five benchmarks are selected to compare the proposed energy management scheme.