Short Term Wind Power Prediction based on TCN-Informer Model Combining STL Decomposition and Frequency Adaptive Normalization

Accurate prediction of wind power plays a very important role in the new environment of the power system where a large amount of clean energy is integrated into the power grid nowadays. The wide application of a high proportion of clean energy and energy storage makes multi-time-scale wind power prediction a necessary basis for assisting the stable and efficient operation of the power grid. Therefore, this paper proposes an Informer prediction model combining STL decomposition and the Frequency Adaptive Normalization (FAN). The model first undergoes STL decomposition of the processed original sequence in terms of architecture, decomposing the original sequence into seasons, trends, and residual components. Then, the three decomposed components are subjected to frequency-adaptive normalization, that is, the influence of non-stationarity is mitigated by filtering the top K dominant components for each input instance in the Fourier domain. It is used to handle the unified non-stationary factors composed of trends and seasonal patterns. Finally, the processed three components are input as multi-dimensional features into the Informer model to predict the future values of the three components. After inverse normalization of the predicted components, they are combined to obtain the final prediction result. The model was verified with the wind power generation data of a certain place in Xinjiang. The results show that the model can effectively achieve accurate prediction at multiple time scales.