Ultra-short-term Power Forecasting Method For Wind Farm Clusters Based On Regional Wind Speed Spatiotemporal Matrix Modeling

With the increasing proportion of wind power in "source" of the power grid,the influence of randomness and volatility of wind power generation on the power system is gradually revealed.Accurate wind power forecasting is an important means of promoting new energy amazing and ensuring safe and stable operation of power systems.Most of the current research is carried out on a single wind farm.However,the current wind farms have the characteristics of clustering and large-scale development,so it is very necessary to forecast the power of wind farm clusters.Ultra-short-term power forecasting,as an important part,plays an important role in frequency regulation,reserve capacity regulation and real-time dispatching of power system.Wind power is mainly affected by wind speed,so it is also very important to obtain high-precision wind speed forecasting results.This paper focuses on deep mining of the spatiotemporal correlation of wind,and uses deep learning algorithms as research methods to conduct research on ultra-short-term power forecasting of wind power clusters,which has important practical value for obtaining highprecision wind power forecasting results and implementing optimal dispatching.This paper introduces the research background and significance of wind power forecasting,and establishes a wind power abnormal data identification model based on quartile method and DBSCAN clustering for the situation that there are many abnormal values in historical data.First of all,this paper classifies the abnormal data and analyzes the reasons through the measured wind speed-power scattergram.Then,the missing time data is completed to ensure the continuity of the data in time.Combined with the fact that the power of the wind power curtailment data remains constant for a period of time,the identification model of the wind power curtailment data is established.Aiming at the isolated points and deviation points in the data,the horizontal quartile method and the vertical quartile method are used to divide the wind power and wind speed into several small intervals for identification.However,some deviation points are close to the wind power conversion curve,and the quartile method cannot accurately identify them.Based on this,a deviation data identification model based on the DBSCAN algorithm is established.Finally,the missing and abnormal data are corrected by the cubic spline interpolation algorithm,which ensures the integrity,reliability and smoothness of the data.Considering that wind power is directly affected by wind speed,this paper proposes a wind speed forecasting model for multi-wind farms based on regional wind speed spatiotemporal matrix modeling.Firstly,the temporal correlation,spatial correlation and spatiotemporal correlation of wind were analyzed by using the measured data,and the spatiotemporal matrix sequence of regional wind speed was constructed.Combining the advantages of GCN in processing non-European spatial data,a multi-wind farm wind speed forecasting model based on GCN is established.The historical wind speed data of each wind farm is used as the node feature,and the correlation coefficient of the wind speed series of each wind farm is used as the edge feature.The spatiotemporal correlation contained in the wind speed spatiotemporal matrix sequence is deeply excavated.From the simulation results,the wind speed forecasting models for multi-wind farms based on the regional wind speed spatiotemporal matrix modeling proposed in this paper have high accuracy under the wind fluctuation types such as breeze,gale,sudden rise and fall.In order to make full use of the information of wind speed and wind power,this paper proposes an ultra-short-term forecasting model of wind power cluster power based on the fusion of CNN-LSTM algorithm.Firstly,this paper fully integrates the forecasting wind speed and historical power data of multiple wind farms,and constructs the forecasting wind speed spatiotemporal matrix and historical power matrix.Combined with the advantages of CNN and LSTM in extracting spatial and temporal features,the forecasting wind speed spatiotemporal matrix and historical power matrix are used as the input of the model,CNN and LSTM algorithms are used to extract wind speed and wind power features.Taking the forecasting deviation of wind power cluster power as the loss function,an ultra-short-term forecasting model of wind power cluster power is established.It can be seen from the simulation results that,compared with other models,the model proposed in this paper has a higher accuracy of ultra-short-term forecasting results,and also has better robustness for wind power forecasting with different time steps.