The Research On The Single Prediction Models Of The Wind Power Based On Structural Optimization

With the rapid development of economy, the requirement of energy will be growing. As the main body of the current world energy structure, fossil fuels has the characteristics of pollution and non-renewable. As a kind of new energy, wind power has the characteristics of clean and reusable, so the use of wind power has become an important form of many countries to improve the energy structure. But the wind is intermittent, randomness and uncertainty. When wind power in parallel operation, it will cause the problem of power grid voltage fluctuation, harmonic pollution, low voltage across and so on, and it is difficult to operate the whole scheduling. Therefore, predict the wind power can make power sector corresponding scheduling plan, on the premise of improve the utilization rate of wind power, guarantee the safe and stable operation of interconnected power grid on the premise of improving the utilization rate of wind power.By the support of National Natural Science Foundation of China(number:51277127), this paper sets up four kinds of single forecasting model of wind power. The model includes:wind power prediction based on improved BP neural network model, wind power prediction based on structural optimization of RBF neural network model, wind power prediction based on structural optimization of wavelet neural network model, and wind power prediction based on unitary time series of kalman filter.The main research contents of this paper include the following several parts:(1) This paper states the research background and significance, summarizes the current status of the development of wind power at home and abroad and wind power prediction of the characteristics of the classification and prediction model, and gives the four types of evaluation index to evaluate the results.(2) In view of the BP neural network input nodes and the number of hidden layer nodes to determine the lack of theoretical guidance, the steepest descent method of BP neural network easy to fall into local minimum value, multivariate time series were used respectively to determine the input variable, the grey relational-sensitivity pruning method to determine the number of hidden layer nodes and the improved particle swarm algorithm to optimize learning algorithm to the BP neural network was improved, the simulation shows that the prediction accuracy of improved BP neural network haves greatly improved.(3) In view of the RBF neural network and wavelet neural network input nodes and the number of hidden layer nodes to determine lack of theoretical guidance, this paper uses the multivariate time series and grey relational-sensitivity method to determine the number of input and hidden layer nodes, using determined structure of network to forecast wind power, the simulation results show the improved network improves prediction precision.(4) For the problem of unitary time series low order model forecasting precision and kalman filter algorithm’s state equation and observation equation is difficult to establish, this paper combines the two methods, through unitary time series to establish a low order model, then using kalman forecast iterative equation of wind power prediction, The simulation results show that the relative of a time series model prediction precision has greatly improved.