Research On The Application Of Neural Network Algorithm In Wind Power Prediction

Accurate wind power forecasting can effectively reduce or avoid the adverse effects of power system while improving wind power's competitiveness in the electricity market. Approaches based on historical data are only suitable for ultra short-term wind power prediction (1～6 hours), but if the prediction time is between 0 and 24 hours or even longer (0 to 72 hours), numerical weather prediction must be used.1. Data pretreatment method research.Numerical weather prediction and wind field measurement data are pre-processed, a relatively complete database is established, and the data pre-processing principles are studied.2. Neural network algorithms of wind power prediction.Firstly, the BP neural network is used in wind power prediction research. Given that RBF is good at handling complicated nonlinear problem, its application is explored in wind power prediction. Advantages and disadvantages are compared through examples. Results show that RBF neural network is more suitable for based on numerical weather prediction of wind power prediction. Results show that RBF neural network is more suitable for wind power prediction based on numerical weather prediction.3. Analysis of the main factors influencing the wind power prediction accuracy①data influence:input variables and the training samples.②eural network model parameters:network structure, hidden layer neurons, weights and threshold selection are key factors in building BP neural network models; hidden nodes data center, expand constants and output layer weights of control are key factors in building RBF neural network models.4. BP and RBF neural network model optimization algorithms.Use genetic algorithm to optimize BP neural network weights and threshold. Propose the use of orthogonal least squares and the gradient descent training algorithm in optimizing RBF neural network models respectively. Practical example results indicate that the above two optimized method can both improve the accuracy of the predictions, while especially the orthogonal least-square training algorithm is best of above methods at the prediction effects of the optimized models.