Probabilistic Prediction of Unit-level Power Scenarios in Wind Farms

The majority of wind power prediction technologies currently available focus on predicting the total power output of wind farms, disregarding the prediction requirements of wind farm unit-level scenarios. To address this gap, this paper proposes a probabilistic prediction method for wind farm unit-level power scenarios, leveraging the convolutional neural network (CNN) and kernel density estimation techniques. Firstly, a CNN is utilized to establish a mapping model between the wind farm tower data and the unit-level power scenarios, enabling a deterministic prediction of unit-level power. Building on this, the power prediction error is calculated to construct an error dataset, which is stratified using the concept of a prediction box for reference. Subsequently, the kernel density estimation method is employed to derive the probability density function and cumulative distribution function of the power prediction error, resulting in the generation of wind farm unit-level power scenario intervals across various confidence levels, which facilitates the probability prediction of unit-level power. Finally, the proposed method's effectiveness is verified using one year's worth of operation data from a wind farm.