Targeting at the problem of imprecise power prediction of wind power generation caused by weather fluctuations and nonlinear features, a new hybrid approach for wind farm power prediction is proposed by a combination of convolutional neural network (CNN), bi-directional long and short-term memory network (BiLSTM) network, and Whale Optimization Algorithm (WOA). The model takes into account both the spatial and the temporal features of the sequence data. The input layer feeds in a large amount of historical power generation, weather and date information, the CNN layer extracts the spatial features, and then the forward and reverse sequences of the data are processed by the BiLSTM layer to capture long term temporal dependence. Meanwhile, the global search of WOA is utilized to optimize the parameters of the CNN-BiLSTM model, and finally, the WOA-CNN-BiLSTM model to predict the wind power sequence. The experiments utilize authentic wind power generation data supplied by Longyuan Power Group Corporation, and the RMSE sum of the experimental results are compared and analyzed by comparing the BP neural network, LSTM, BiLSTM, CNN-LSTM, and BiLSTM-Attention models. The results show that the RMSE of this hybrid model is 4.425, is 0.965, and the root mean square error is decreased by ${2. 8 \%}$ in comparison to the conventional LSTM model, resulting in improved prediction accuracy and stability.
Research and Application of WOA-CNN-BiLSTM Based Model for Wind Power Prediction
2024-10-23
1833530 byte
Conference paper
Electronic Resource
English