Research On Wind Power Combination Prediction Model Based On Neural Network

With the carbon peak,and carbon neutral"30-60 target"this economic and social systemic changes are proposed.It will further promote the optimization of China’s energy structure and accelerate the development of the wind power industry.Xinjiang is very rich in wind energy resources due to its unique geographical characteristics.It is one of the major wind power installation areas in China.Wind power generation itself has inherent characteristics such as randomness,volatility,and intermittency,presenting a significant anti-peak regulation.It imposes a disorderly impact on the stable operation of the power system.Limits the entry of large amounts of wind power into the power system.Accurate wind power prediction can effectively solve these problems.It can provide both more accurate forecast values for power system dispatch and reference for a unit mix.Thus,it can reduce the cost of power generation and improve the market competitiveness of wind power generation.Deterministic prediction method of wind power based on combined prediction model of data decomposition-prediction technique.The time-frequency analysis method is used to smooth the original signal and realize the efficient separation of multi-component signals.Different methods are used to model the submodular data characteristics with high prediction accuracy and prediction efficiency.It shows good prediction performance in wind power prediction.In this thesis,the research is related to data pre-processing by SSA,VMD,and PE techniques and classification prediction methods by GRU and SVR techniques.The main research contents of this thesis are as follows.(1)in data preprocessing.For the problem that the VMD cannot be decomposed adaptively,an optimized variational modal decomposition algorithm is proposed.The method of adaptively determining the hyperparameters K andαof VMD based on SSA is investigated.The original wind power time series is decomposed into K smooth subseries by the optimized variational modal decomposition algorithm.The smoothness of the original wind power time series is achieved.It is also integrated with the PE method.The recombination of the decomposed subseries is completed according to the closeness of PE values.The constructed SSA-VMD-PE data preprocessing model improves the predictability and robustness of wind power prediction.(2)In the prediction method.For the problem of co-integration of prediction effectiveness and prediction efficiency.The sub-modes are classified into high and low frequency according to the PE value size.The classification of high and low frequency sub-modes prediction is implemented.High-frequency sub-modes data have high complexity and non-smoothness.The prediction is done by the gated recurrent unit neural network wind power prediction model.the GRU neural network can effectively mine the wind power series data features.The low-frequency submodular data has low complexity and strong nonlinearity.The prediction is done by the support vector regression wind power prediction model.the SVR prediction is fast and can solve the regression problem at the same time.Through the classification prediction method,the prediction efficiency of wind power is improved while ensuring the prediction effect.(3)The proposed combined wind power prediction model based on SSA-VMD-PE-GRU-SVR.The application study was carried out with the actual wind power generation power of the Darshancheng wind farm according to four seasons respectively.At the same time,seven comparative models setting approximate parameters were constructed.The comparative analysis was carried out by test sets.The application effectiveness of the proposed model is verified based on three error evaluation indexes,MAE,RMSE,and R~2,and the corresponding improvement percentages.The experimental results show.The proposed model is more effective and stable in extracting detail information and trend information in wind power series compared with the comparison model.It has the optimal multi-step prediction performance and has good practical engineering application value.