Research on machine learning-based curtailment analysis for optimal operation of wind farm

Novel controllers have been introduced in energy systems to enhance efficiency and manage uncertainties in renewable sources. These controllers incorporate real-time meteorological and SCADA data for more accurate wind turbine output predictions. This paper presents a curtailment weighting control module using a CNN-BiLSTM model trained on data from the Dongbok wind farm. The objective is to minimize active power loss during curtailment using a sequential quadratic programming algorithm, focusing on improving prediction accuracy and operational efficiency. The system integrates a real-time simulation environment, including hardware-in-the-loop simulation. Communication between wind farm controllers, RTDS grid models, and the management system is via the Modbus TCP/IP protocol. Comparative analysis with traditional methods validates the proposed method's advantages. Real-time simulation results indicate that the curtailment weighting control module reduces power losses and improves operational efficiency. The proposed method achieved a power loss reduction of 2.5813 kWh compared to traditional PD control methods and demonstrated a 99.9% availability across all wind turbines during curtailment scenarios. This study highlights the importance of combining machine learning-based control with advanced strategies to address challenges in renewable energy integration and maintain a stable power supply.