基于期望实现深度学习的风电功率与爬坡率确定性及概率预测

Accurate day-ahead forecasting of wind power generation, both deterministically and probabilistically, is crucial for reliable and efficient power system operations, and its significance will increase in renewable energy-dominant power systems. Furthermore, inherent variability of wind farms necessitates day-ahead ramp rate forecasting to ensure stable energy balancing. To address these needs, we propose an hourly day-ahead forecasting framework that concurrently predicts the wind power generation and ramp rate. This framework leverages expectation-implemented deep learning models trained by the custom loss functions tailored to the different signal attributes and our distinct expectations. Additional strategies, such as feature generation and a feedforward error learning model, are implemented to enhance forecasting performance while maintaining a balance between tasks. Finally, our framework integrates heterogeneous generation and ramp rate forecasting results, incorporating probabilistic ramp rate forecasting derived from the synthesized output. We validate our approach using real wind power data and assess the impact of each proposed method individually. The results demonstrate that the proposed framework can significantly contribute to identifying the intrinsic volatility of wind power, thereby fully exploiting its potential benefits.