Research On Ultra-Short-Term Wind Power Forecasting Method Based On Combined Model

With the promotion of the energy revolution,wind power has been the fastest growing renewable energy source due to its low construction cost,non-pollution,and no usage restrictions.However,the inherent characteristics of wind power output,such as randomness,volatility and uncertainty,may restrict the integration of large amounts of wind power into the power system.Ultra-short-term wind power prediction is the crucial technology to alleviate this problem.Accurate wind power forecasting can enhance the power system’s adaptability to the inherent characteristics of wind power and enable safe access to the grid for wind power system,further enhancing wind power’s competitiveness in the power market.Therefore,this paper investigates ultra-short-term point prediction of wind power as well as interval prediction techniques,and the main work is as follows:1.The outliers in the original wind power series are filtered based on the quartile method,and then the outliers are filled with the linear interpolation method.To solve the problem of introducing noise in the acquisition process of the original wind power data,the Wavelet Transform(WT)is used to eliminate the noise in the original wind power data,which can reduce its volatility at the same time,and provide data preparation for the establishment of the subsequent prediction model.2.Complete Ensemble Empirical Mode Decomposition With Adaptive Noise(CEEMDAN)adaptively decomposes the original wind power time series into a series of modal components to reduce the nonlinearity and non-smoothness of the original series.Meanwhile,according to the Weighted Permutation Entropy(WPE),the similarity between the modal components is calculated and the similar components are reorganized to correct the over-decomposition problem of the adaptive noise complete set empirical mode decomposition,making the modified modal components more regular.Then,the reconstructed components are input into the Convolutional Neural Networks Long Short-Term Memory(CNN-LSTM)for time series modeling,and the neural weights of the CNN-LSTM network are optimally allocated using the self-attention mechanism to improve the adaptability of the CNN-LSTM network to the uncertainty of input characteristics.A multi-perspective validation of the point prediction performance of the proposed combined model using measured wind power data from the Belgian ELIA website.The experimental results show that the proposed model has a reasonable and effective structural design,and the MAPE decreases by an average of 55.27% and 24.41% respectively compared to the optimal results of the competing models and existing research results,verifying that the proposed combined model has a superior point prediction performance.3.For the wind power output has the characteristics of randomness,fluctuation and uncertainty,the deterministic point prediction cannot effectively reflect its fluctuation information,which is difficult to meet the requirements of actual power system operation and dispatch.However,the interval prediction can not only obtain the deterministic point prediction results,but also represent the corresponding fluctuation range of the point prediction results at the predetermined confidence level,so as to provide more comprehensive information for power system dispatching.Therefore,this paper proposes a wind power interval fusion prediction method combining adaptive bandwidth with Nonparametric Kernel Density Estimation(NKDE)on the basis of the proposed deterministic point prediction model.The verification by a arithmetic case shows that the proposed interval prediction model has higher reliability index,narrower interval width,and better prediction performance,and it is suitable for the randomness,volatility and uncertainty of the wind power output.