Study On The Method Of Power Forecasting In Wind Farm

Wind power is an inexhaustible and inexhaustible renewable energy. Current wind energy is equivalent to coal consumption of energy more than 1,000 times every year in the global. Its content is considerably more than the water, also greater than solid fuel and liquid fuel energy combined. Therefore wind power is more great importance to generate electricity at home and abroad.The characteristic of Wind power is distribution range, but the energy density is lower. For always in the atmospheric freedom movement state, so stability is low. Although it is everywhere, but it has very big random process. Therefore it creates wide fluctuations of output power of wind fans. It is lead to the power supply of wind generator cannot satisfy the requirements of grid's stability, continuity and adjustability etc. The constant change of output power is easy to bring shock to grid and aggravate the grid peak load regulating operation burden. It is also bringing great difficulties for dispatching work in the power sector. So finding a effective method for power forecasting of wind farm is the key point..Through the study found that application of time series method to find the law of historical samples of wind farm output power for a short time a more accurate prediction. But for a long forecasting, the wind farm output power has great randomness, and subject to weather and other factors, so a large forecast errors. The BP neural network in the application of wind power output forecasts, although the basis for numerical weather prediction as a forecast, but if the input space and network-related serious dimension since the higher, BP neural network prediction accuracy will drop. To solve this problem, this paper uses samples of K-S Methods, and then principal component analysis to extract useful information for solving the principal components, eliminating the correlation between samples. Proposed wind farm output power while the dual neural network model to compensate for defects in a single predictive model to improve prediction accuracy.Single model with the previous forecast methods on the instance results show that the dual neural network model for wind farm output forecasting model eliminates the relevance of input factors and to simplify the network structure, which greatly improves the accuracy of the prediction, with a possible and effectiveness. Finally the error is analyzed, it is found that wind power prediction error is not only with prediction model, but also related to the power curve.