TCN prediction method for non-smooth wind power sequences based on CEEMDAN-DBO co-optimisation

With the increasing integration of wind energy into the grid, high-precision wind power forecasting has become increasingly important for grid security. To address the randomness of wind speed and the limitations of traditional models, a CEEMDAN-DBO-TCN short-term forecasting method is proposed. The wind power sequence is first processed through the CEEMDAN method, which employs adaptive noise to decompose the signal and extract multiscale intrinsic mode functions (IMFs), thereby reducing non-stationarity. This technique extracts the multi-scale intrinsic mode function (IMF) to minimize the non-stationarity.Secondly, the Dung Beetle Optimization (DBO) algorithm is applied to globally optimize the hyperparameters of the Temporal Convolutional Network (TCN), This approach solves the tuning problem of TCN and avoids local optima through the simulation of rolling and foraging behaviors. Finally, the TCN’s dilated convolutions and residual connections are incorporated to capture long-term temporal dependencies. Experimental results based on 2018 data from a wind farm demonstrate that the proposed algorithm and model combination outperforms LSTM and CNN models in terms of RMSE and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, validating its advantages in feature extraction and accuracy improvement.