Short-term offshore wind power multi-location multi-modal multi-step prediction model based on Informer (M3STIN)

In recent years, offshore wind power has become an important source of wind power generation. When large-scale offshore wind farms generate power in groups, the accuracy of wind power prediction is crucial for the stability of the system. This report focuses on multi-location multi-step spatio-temporal wind power prediction. It is designed to exploit spatial dependencies by the relative physical locations of offshore wind farms in order to improve prediction and generate forecasts for eight locations. This report proposes an Informer-based multi-location multi-modal multi-step prediction model (M3STIN). In this model, spatial correlation between offshore wind farms is considered using Pearson coefficient and the Gaussian kernel function. To address spatial and temporal dependencies, graph attention networks and Informer models are applied, respectively. Furthermore, the incorporation of multi-task learning with auxiliary tasks, along with the integration of multi-modal strategies, contributes to enhancing both accuracy and computational efficiency of the prediction model. To validate M3STIN, the model is compared with 10 benchmark models based on MAPE, MAE, R2, and computation time. The results demonstrate that the model achieves the lowest prediction error in multi-step short-term forecasting across all sites, highlighting its superior performance over existing models.