Abstract: To tackle the problems of stochasticity and intermittency characteristic of wind power data, this paper introduces an innovative short-term wind power forecasting model which employs variational mode decomposition（VMD） combined with Transformer-BiGRU architecture to generate a hybrid forecast. Based on the principle of variational mode decomposition, the original data is decomposed into a series of frequency subsequences, and signal features are fully extracted. On this basis, a Transformer-BiGRU model is constructed that uses bidirectional gating units to capture bidirectional temporal dependencies. Then, four key hyperparameters are optimized through Bayesian optimization. Finally, the predicted values of the decomposed sub signals are accumulated to obtain the final prediction results, thus constructing a Transformer BiGRU model based on bidirectional gating units capturing bidirectional temporal dependencies.The experimental results show that the proposed model is significantly better than the comparative model, accurately predicting wind power and effectively enhancing the stability and accuracy of short-term wind power prediction, providing a technical optimization path for efficient wind power prediction.

Key words: short-term power forecasting, variational mode decomposition, Transformer, bidirectional gated unit, Bayesian optimization algorithm