Session: 01-02: AI for Energy Sustainability II

Paper Number: 131171

131171 - An Ultra-Short-Term Power Prediction Method for Wind Farms in Northwest China Based on Federated Learning 

Abstract: 

With the increasing depletion of Earth's resources, energy issues have become increasingly prominent. The continuous extraction and consumption of fossil fuels are leading to the depletion of primary energy sources and environmental pollution. Solar and wind power generation stand as important directions in the field of new energy. Wind power is highly favored due to its minimal environmental impact, great economic benefits, and mature technology. China is developing large-scale wind farms in its northwestern region as a significant part of its Carbon Neutrality strategy. Accurate power prediction for large-scale wind farms is one of the keys to the safe operation of China's future power grid. Accurate prediction of wind power can reduce the impact of wind power generation uncertainty on the grid, ensuring power supply reliability and energy quality. This assists relevant departments of the power system to adjust scheduling plans promptly and formulate appropriate control strategies, reducing instances of curtailed wind power, making it more controllable energy sources to ensure power system stability, and thereby increasing the prevalence of renewable energy generation systems. Additionally, precise wind power forecasts can assist wind farms in participating in electricity market bidding, laying the foundation for China's electricity market reform. With the rapid construction of wind farms across Northwestern China, farms of diverse scale and a variety of start-up time have to overcome communication barriers. These factors collectively present challenges to achieving precise power predictions for wind farms. To address this, this paper presents a universal ultra-short-term prediction method for multiple wind farms using an encoder-decoder network based on federated learning. The prediction power sequences are separated into trend and fluctuation components to be predicted separately. While considering the data privacy of different farm owners, this approach overcomes the low accuracy and lack of model universality in power predictions for various wind farms. In case studies of four wind farms spanning diverse climate zones and geographical locations in Northwestern China, this method achieves a prediction accuracy of 85%, a 5%-10% improvement compared to conventional wind power prediction methods. This approach also addresses the challenges of small-sample predictions in newly established wind farms, reaching an 80% accuracy rate in two case studies with limited data. The proposed method overcomes the challenge of data heterogeneity among different wind farms. It involves collaborative learning across multiple wind farm entities while respecting the privacy of individual data, enabling more precise and universal power predictions for wind energy systems.

Presenting Author: Xiaojie Lin Zhejiang University

Presenting Author Biography: 2021.2-present: College of Energy Engineering, Zhejiang University Associate researcher (in tenure-track)
2020.2-2021.2: Zhejiang University Changzhou Institute of Industrial Technology Senior expert
2018.3-2020.2: College of Energy Engineering, Zhejiang University Postdoctoral fellow
2012.8-2017.8: University of Maryland Ph.D

Working in modeling and control of multi-energy flow systems in industrial park integrated energy
Published 35 SCI/EI indexed papers as the first or corresponding author in the past three years
Received a third-class prize of Jiangsu Science and Technology Award.

National Key R&D Program of China ‘Joint Research and Demonstration Project on Collaborative Energy Management and Operational Optimization of National City Smart Energy Networks under the 'Belt and Road' Initiative’Core Member
National Key R&D Program of China ‘Modeling and Distributed Collaborative Control Methods for Energy Systems in Manufacturing Enterprise Clusters’Core Member

Authors:

Wu Rong Zhejiang University
Xiaojie Lin Zhejiang University
Feiyun Cong Zhejiang University
Wei Zhong Zhejiang University