Channel clustering based LSTM joint time-frequency decomposition for ultra-short-term forecasting of offshore wind power

Offshore wind power forecasting is fundamental for ensuring the safe dispatch and economic efficiency of power systems. It is more challenging due to the extreme intermittency and volatility of offshore wind power and the sever operating environment. This paper proposes a Time-Frequency Decomposition and Channel Clustering Long Short-Term Memory model (TF-CC-LSTM) for ultra-short-term forecasting of offshore wind power, which integrates time-frequency decomposition and channel clustering to address wind power uncertainty. Firstly, the meteorological and historical power data are preprocessed by the time-series decomposition module to extract key features and capture long-term trends and periodic fluctuations. Secondly, multilayer perceptron (MLP) is applied combined with time-domain and frequency-domain analysis techniques to enhance feature extraction. The channels of these extracted features with a higher degree of similarity are clustered by K-means clustering, and the offshore wind power forecasting model is established with the features based on LSTM model for each cluster. Finally, a series of experimental results prove the effectiveness of the proposed model.