Ultra-short-term Forecasting Of Wind Power Based On Graph Neural Network Wind Pattern Prediction

As global climate change intensifies,countries around the world are committed to reducing carbon emissions,and low-carbon green energy represented by wind power has ushered in a huge development opportunity.However,compared to conventional energy sources,wind power generation is entirely weather-dependent and difficult to control artificially.The randomness of its fluctuations and uncertainty of changes can impact the stable operation of power grids.Especially in areas rich in wind power resources,the quality of power can be seriously threatened due to the high penetration of wind power.For this reason countries have invested in the field of wind power forecasting,by obtaining the expected value of future wind power can optimise system dispatch and weaken the negative impact of wind power randomness on the grid.After decades of development,the field of wind power forecasting has developed its own architecture,but there are still some shortcomings in the specific applications.Based on this situation,this paper conducts an in-depth study on ultra-short-term forecasting of wind power cluster power.In the past wind power cluster power prediction,the area where the cluster is located is usually divided into grids or blocks,or a power station is selected as a typical station,and its predicted power is directly multiplied by a multiple as the predicted value of the overall output of the cluster.The temporal and spatial correlation between the stations in the cluster does not realize the predictive effect of the upstream stations on the output of the downstream stations,and does not make full use of the strong correlation between the adjacent stations,which severely limits the power prediction of the wind power cluster further improvement in accuracy.In addition,the wind power varies widely,and it is difficult to fit all the variation laws of wind power with only one model.Therefore,this paper divides wind power fluctuations into a variety of different patterns and proposes an ultra-short-term prediction model for wind power clusters based on graph neural network wind pattern prediction.The model proposed in this paper consists of two main innovations: firstly,the wind power is divided into patterns,with different patterns representing different patterns of variation;secondly,the wind power cluster is modelled using graph neural networks to fully exploit the spatial and temporal correlation between the sites in the cluster.The model consists of the following four steps: firstly,the wind speed and power data are noise-reduced using the wavelet transform to remove the random high frequency fluctuations in the data and obtain smoother data that retains most of the information of the original data;afterwards,the wind power fluctuation process is divided into various fluctuation patterns using two methods: thresholding and clustering,and a prediction model is built for each pattern;then the wind power cluster is modelled using the graph neural networks are then used to model the wind power clusters to obtain the preliminary prediction results of the wind power clusters,and to determine the fluctuation mode to which they belong according to the prediction results;finally,the prediction values are obtained by selecting the corresponding prediction models according to the mode to which they belong,and the prediction values of each model are combined to obtain the final prediction results.This paper uses data from20 wind power plants located in northwest China,based on which modelling and simulation experiments are carried out and compared with other models,and the final results validate the effectiveness of the method.With the deepening of the reform of the electricity market,it is crucial for wind power companies to achieve better and more accurate power forecasting,which is a key issue for their profitability and survival.The method proposed in this paper has certain reference significance,and this method can not only be used for wind power forecasting,but also can be applied to a wider range of time series forecasting.