Ultra-Short-Term Multi-Step Prediction of Wind Power Based on ModernTCN and Multi-Task Learning

The intermittent and fluctuating nature of wind energy makes it difficult to accurately predict wind power, leading to low penetration in the wind power grid. To address this problem, this paper designs ultra-short-term multi-step prediction of wind power based on ModernTCN and multi-task learning to improve the prediction accuracy. Firstly, ModernTCN's ability to have a large receptive field and capture cross-variate timing dependencies is exploited to obtain multivariate timing feature information and establish a multi-step mapping of input features to power. Then it is inputted into the MMoE multi-task learning method, and the prediction results are corrected to obtain the ultra-short-term prediction of wind power using multiple expert networks and gating units. The experimental results show that the method proposed in this paper can effectively improve the ultra-short-term prediction accuracy of wind power, and can provide a favourable reference for the scheduling centre to adjust the power generation plan in real-time response to the power demand.