Short-Term Power Forecasting and Uncertainty Analysis of Wind Farm at Multiple Time Scales

Wind power poses a challenge to the stability of the power grid due to its unpredictability and intermittency. This study aims to analyze the forecasting law and uncertainties of short-term wind farm power forecasting (WFPF) at various time scales, in order to support the stability of energy generation. To achieve this, we propose a framework for short-term WFPF and uncertainty analysis, utilizing the whale optimization algorithm (WOA), convolutional neural network-bidirectional long short-term memory network (CNN-BiLSTM), cloud model (CM), and non-parametric kernel density estimation (NPKDE). The data is trained using a hybrid model of CNN-BiLSTM with multiple convolution and pooling methods, while the parameters are optimized using the WOA algorithm. The uncertainty of WFPF is described qualitatively by the expectation, entropy, and hyper-entropy of the cloud model, and quantified through the confidence interval based on non-parametric kernel density estimation. Test results show that the proposed WOA-CNN-BiLSTM model achieves RMSE forecasting errors of 3.79%, 4.52%, and 5.12% at 4 hours, 24 hours, and 72 hours, respectively. The maximum peak errors are less than 10.5758MW, 21.128MW, and 20.0292MW, and are better than other models. Additionally, the WOA optimization performance is superior, consistent with the results described by the cloud model. Furthermore, the RMSE forecasting value of WFPF increases with the time scale, while the growth rate of RMSE decreases with the increase of time scale. This study provides valuable insights into the uncertainties of short-term WFPF and offers a robust framework for improving the stability of energy generation.