A wake prediction framework based on the MOST Gaussian wake model and a deep learning approach

Abstract

With the rapid development of wind energy, accurately predicting the wake speed distribution behind wind turbines is crucial for load assessment and coordinated control of wind farms. However, existing wake models still fall short in accurately predicting under the complex and variable inflow characteristics and turbine operating states in actual wind farms. To address this issue, this paper proposes a wake prediction framework that combines the Gaussian wake model based on Monin-Obukhov Similarity Theory (MOST) and deep learning approach. In this framework, the MOST Gaussian wake model is improved to account for yaw correction, and the one-dimensional convolutional neural network-bidirectional long-short-term memory (1DCNN-BiLSTM) deep learning model is employed to dynamically calibrate the wake expansion rate parameters using both inflow characteristics and turbine operating states as inputs. Validation with actual wind farm case studies shows the proposed framework achieves 95.35% wind speed prediction accuracy and 84.17% power accuracy at Penmanshiel wind farm, and 97.12% wind speed accuracy and 87.59% power accuracy at La Haute Born wind farm. The high prediction accuracy of this framework provides a reliable basis for future load assessment and coordinated control of wind farms and offers new ideas for optimizing wind farm design and operation strategies.