Research On Short-term Wind Power Probabilistic Prediction Method Based On Combination Model

With the further development of global energy conservation,emission reduction,and energy transformation,the efficient use of renewable energy has received widespread attention from countries around the world,among which vigorously developing wind power generation is an effective way to realize energy transformation and take the path of sustainable development.Due to the intermittency,volatility,and randomness of wind power,large-scale wind power entering the power grid will affect the balance of the power system,reduce the reliability of the power grid,and pose major challenges to the security and stability of the power grid.Wind power prediction can improve the accuracy of wind power dispatching and further maintain the safe and stable operation of power grid.Therefore,accurate and reliable wind power prediction is very important in power system optimization and dispatching,network planning and other tasks.In order to improve the prediction accuracy of short-term wind power,this paper proposes a combination prediction method based on "data preprocessing-parameter optimization-deterministic prediction-probability prediction".The main research contents are as follows:1)The current research status and development trend of wind power prediction technology at home and abroad are analyzed;The prediction methods are classified according to different time scales,spatial scales,prediction models,prediction objects,and prediction forms,and some wind power prediction systems that have been successfully put into operation at home and abroad are listed.2)By analyzing the mechanism and physical model of the wind power generation system,the influencing factors of wind power are summarized.After that,the entropy weight-grey relational analysis method(EWM-GRA)is used to analyze the correlation between different environmental variables and wind power data,and select multidimensional input data.Then,in view of the strong randomness and volatility of the original wind power data,which lead to inaccurate prediction results,empirical mode decomposition(EMD)and variational mode decomposition(VMD)are proposed to preprocess the data respectively,and a simulation verification is conducted using the actual data of a wind farm in Hebei Province as an example.The comparison of simulation results shows that the VMD method has better noise reduction effect.Finally,summarize and analyze the problems existing in the VMD method,laying a theoretical foundation for the following research.3)Aiming at the problem that VMD is difficult to achieve optimal decomposition due to improper parameter settings during data decomposition,particle swarm optimization(PSO)and sparrow optimization(SSA)algorithms are constructed to optimize VMD parameters,respectively,so as to realize the adaptive determination of the number of modes K and the penalty factor α of VMD.Taking the actual data of a wind farm in Hebei Province as an example,experimental comparative analysis is conducted to provide reliable sample data for establishing a combined forecasting model.4)Select Support Vector Machine(SVM)and Long Short-Term Memory Neural Network(LSTM)as basic prediction models,and construct SSA-VMD-SVM,SSA-VMD-LSTM,EWM-GRA-SSA-VMD-SVM,and EWM-GRA-SSA-VMD-LSTM combined prediction models,respectively.By comparing and analyzing the prediction effects of different prediction models step by step,the advantages of the EWM-GRA-SSA-VMD-LSTM combined model is verified.5)Aiming at the mutability of input variables of wind power prediction model and the uncertainty of wind power prediction results brought by the robustness of the model itself,this paper uses a nonparametric kernel density estimation method(NKDE)to analyze the uncertainty of the output results of the wind power prediction model in order to determine its fluctuation range.Firstly,the prediction error distribution of the EWM-GRA-SSA-VMD-LSTM model is analyzed using NKDE,then the prediction interval under the given confidence was calculated,and the probabilistic prediction results are evaluated using the evaluation indicators of the probabilistic prediction model.Finally,the effectiveness of the proposed method is verified by comparing with the parameter estimation method.