Multifactor Short-Term Wind Power Interval Prediction Based on IFC-TCN-BiGRU-attention

Accurate wind power prediction is crucial for the safe and efficient operation of modern power systems dominated by renewable energy sources. To achieve more precise forecasting, this paper adopts a multi-factor wind power prediction approach and introduces the Improved Feature Cross (IFC)-Temporal Convolutional Network (TCN)-Bidirectional Gated Recurrent Unit (BiGRU)-attention model. Firstly, IFC is introduced to uncover hidden relationships between features and identify the features that need to be combined. This approach enriches the originally diverse set of features. To avoid data redundancy, Pearson correlation analysis is employed to examine the correlation between wind power and various factors, as well as historical wind power. Simultaneously, the feature importance of random forests is utilized for feature selection, reducing the dimensionality of the features. Subsequently, the TCN-BiGRU-attention prediction model is employed for point forecasting. In point forecasting, it is challenging to avoid prediction errors. To quantify the uncertainty of predictions and provide decision-makers with more information, point predictions are ultimately transformed into interval predictions based on quantile regression. Through a case analysis of actual measurement data from a wind farm in Hami, Xinjiang, the results indicate that the proposed method can better perceive the potential impact of secondary factors, make more accurate predictions during power fluctuations, and simultaneously reduce training time, enhancing overall prediction accuracy.