Short‐Term Offshore Wind Power Prediction Based on Significant Weather Process Classification and Multitask Learning Considering Neighboring Powers

Abstract

Offshore wind power is an important technology for low‐carbon power grids. To improve the accuracy, a short‐term offshore wind power prediction method based on significant weather process classification and multitask learning considering neighboring powers is presented in this paper. First, a novel weather process classification method, in which the samples are divided into pieces of waves based on extreme points and are quantified with labels of energy level and fluctuation level, is proposed to classify samples into multiple types of significant weather processes for independent modeling. Second, a multitask learning method, in which the power sequences in neighboring offshore wind farms are innovatively introduced as a new input feature, is proposed for modeling wind power prediction for each wind farm inside a neighboring region under each weather process class. Case studies are presented to verify the effectiveness and superiority of the proposed method. Based on this new method, the 4‐h ultra‐short‐term root mean squared error (RMSE), 24‐h day‐ahead RMSE, 4‐h ultra‐short‐term mean absolute error (MAE), and 24‐h day‐ahead MAE can be reduced by 1.45%, 2.1%, 1.15%, and 1.85%, respectively, compared with benchmark methods, which verify the effectiveness of the proposed method.