Research On Wind Power Prediction Based On NWP And The Improved BP Neural Network

As a clean non-pollution and renewable energy, wind power can solve the problems of environmental pollution and the lack of fossil energy. However, the volatility, intermittency, and uncontrollability of wind energy bring great challenges to the stable operation and generator schedule of the power grid system with inaccurate forecasts of wind power connected. Study on wind power prediction algorithm with greater accuracy is important for improving the utilization level of wind power and planning and dispatching power grid. This paper uses a new neural network which is improved by the topology of small-world network as the prediction model. To improve the accuracy of wind power prediction, this paper focuses on data processing methods and the prediction methods. The following researches have been carried out in this paper.1. This paper improved the BP neural network with the topology of small-world network based on the algorithm of BP neural network and small-world network. The improved BP neural network (S WBP) is set up as a predicting model.2. The mutual information feature select algorithm is improved and applied in the wind power input feature selection for there is no practical feature selection algorithm in wind power prediction field.3. The average algorithm is proposed to do the data preprocessing. This preprocessing algorithm speeds up the convergence speed of the neural network in the training stage is much faster, decreases the training error, and increases the prediction accuracy.4. The properties of the SWBP algorithm is verified in the short-term wind power prediction using the input features selected by the mutual information and processed by the average algorithm. The results proved that the SWBP algorithm is much better than BP algorithm in convergence, training error and prediction accuracy.5. The wind power prediction method is divided into scrolling real-time prediction and daily prediction according to the national energy administration files. The prediction properties of SWBP algorithm are compared with BP, PSOBP and RBF neural network algorithm in scrolling real-time wind prediction, which proved that the SWBP algorithm is more suitable for wind power real-time prediction. The numerical weather prediction data is applied into wind power daily prediction, which proved that SWBP neural network algorithm can meet the requirement of wind power daily prediction.