Journal
APPLIED ENERGY
Volume 314, Issue -, Pages -Publisher
ELSEVIER SCI LTD
DOI: 10.1016/j.apenergy.2022.118997
Keywords
Deep learning; Virtual power plant; Integrated energy systems; Zero-carbon; Electric vehicles; Robust-stochastic programming
Categories
Funding
- National Key Research and Development Program of China [2019YFB1705401]
- Natural Science Foundation of China [61873118]
- Science, Technology and Innovation Commission of Shenzhen Municipality [20200925174707002, ZDSYS20200811143601004]
Ask authors/readers for more resources
This paper proposes a deep learning approach and optimization model for the optimal day-ahead scheduling of zero-carbon multi-energy systems (ZCMES) virtual power plants. The proposed model considers carbon capture, electric vehicle flexibility, and clean energy marketing strategy to ensure system reliability, and achieves optimal decision-making using integrated recurrent unit-bidirectional longshort term memory (GRU-BiLSTM), autoencoder (AE), and robust-stochastic modelling approaches. The model is verified using historical multi-energy data, demonstrating its superiority in achieving high self-consumption and load cover ratios, as well as cost reduction in day-ahead scheduling. The model also strengthens carbon-neutral feasibility and provides a reference tool for sustainable energy policymakers.
The zero-carbon multi-energy systems (ZCMES) have received attention due to developed countries' promulgated carbon-neutral policy. Thus, This paper proposes a deep learning approach and optimization model for the optimal day-ahead scheduling of ZCMES virtual power plants. Technically, a carbon capture system (CCS) is introduced to harness the carbon emission associated with some equipment, consideration of electric vehicle multi-flexible potentials, followed by a clean energy marketer (CEM) strategy to ensure system reliability sustainably. For day-ahead multivariable time-series prediction, an integrated recurrent unit-bidirectional longshort term memory (GRU-BiLSTM) is developed. This is followed by an autoencoder (AE) for scenario generation and scene reduction using the fast forward reduction algorithm. A robust-stochastic modelling approach is then applied for optimal decision-making. As a case study, the proposed model is verified using accurate historical multi-energy data of a district in Arizona, the United States. The results show that the proposed model outperformed other scenarios by achieving a 76% average self-consumption ratio and 0.85 average multi-energy load cover ratio. Also, the proposed method obtains a 10.74% reduction in day-ahead scheduling cost by considering the CEM trading period and EV flexibility. Further, a 36% reduction is observed using a robuststochastic approach, which is more robust and economical than deterministic, stochastic, and robust methods. Remarkably, it was observed that the CEM trading period restriction influenced the scheduling behaviour of ZCMES and the charging pattern of EVs. However, the integration of EV flexibility reduces dependency on the external grid and optimize the power consumption of CCS using part of cogeneration electrical output instead of total reliance on the external grid. Thus, the proposed model strengthens carbon-neutral feasibility in urban centres and serves as a reference tool for sustainable energy policymakers.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available