Research On The Ultra-short-term Wind Power Forecasting Method

With the consumption of fossil energy and the increasingly prominent environmental pollution problem,wind power,as a huge renewable clean energy,has been widely concerned all over the world.The total installed capacity of wind power in China has ranked the first in the world.Due to the development mode of the wind power,which is combined to the grid intensively,the problem of power grid affected by wind power wave is very serious.The development of wind power itself is also influenced by this problem.Using wind power forecasting technology to achieve accurate prediction of wind power is conducive to wind power online bidding,reducing wind abandoning and mitigate the adverse effects of wind power to the power grid.Achieve a win-win goal by this way.At present,the combination forecasting model is a key point in the research of wind power.However,how to choose single prediction models and combination way is also worth further exploring.At the same time,it is still worth exploring how to improve the performance of ultra-short-term multi-step prediction of wind power prediction model.Around the above problems,the research on the ultra-short-term prediction of wind power is carried out in this paper.In this paper,a single prediction model selection method based on grey relational degree theory is proposed.Before establishing the combined forecasting model of wind power,it is necessary to select some single prediction models from the set of prediction models.The proposed method gives the selection criteria of single prediction models,avoiding the subjectivity and blindness of being given the prediction models directly,and ensuring the rationality and effectiveness of the selected individual prediction models.In this paper,a nonlinear combined prediction model of wind power based on different optimization criteria and generalized regression neural network is established.Based on the three selected models,different linear combined forecasting models are built by different optimization criterions.Then,the GRNN is used to establish the nonlinear combined forecasting model by those linear combined models and then obtain the optimized model.The verification with the real measured data of a wind farm shows that all the indexes of the proposed optimization model have been improved compared with before improvement,and the accuracy of ultra-short-term prediction of wind power has been effectively improved.Considering the defect that gradient descent algorithm is easy to fall into local minimum when solving weights and thresholds of adaptive wavelet neural network,an improved PSO-DE algorithm is proposed to optimize the weights and thresholds of adaptive wavelet neural network.The particle swarm algorithm and differential evolutionary algorithm for generating new individuals in different ways.The proposed method is a combination of the two algorithm,and it builds a bridge to realize the free exchange of information between different kinds of groups.It's helpful to a single heuristic algorithm to avoid being into local minimum by error of judgment information.Using the improved model to realize ultra-short-term multi-step prediction for wind power.The simulation results show that the improved PSO-DE algorithm effectively optimizes the adaptive wavelet neural network weights and thresholds,the proposed model has high multi-step ahead prediction accuracy and good practical value of engineering.