Research On Short-term Wind Speed And Wind Power Forecasting Based On Hybrid Models

Wind power, as one kind of environment-friendly and renewable energy, has received wide attention. And the capacity of wind power integration keeps increasing in recent years. Due to the intermittency and randomness of wind power, effective wind speed and wind power forecasting is the precondition of large-capacity wind power integration. Short-term forecasting method with high accuracy and strong generalization ability can ensure electric system’s safe and stable running, provide information for the adjustment of power grid and decrease the cost of wind farm’s operation.In this paper, based on the historical data collected from three wind farms in Northwest China, East China and Southwest China, several short-term forecasting methods are researched. The main work conducted is as follows:Firstly, one-hour-ahead prediction based on ARMA model is conducted. The prediction errors are stable and meet the engineering requirement. Some data preprocessing methods are also included, and the improvements of prediction accuracy based on them are verified.Then basic forecasting results based on a three-layer BP network are collected. Since the BP network has low convergence speed and easily falls into local optimal solution, Genetic Algorithm is used to optimize its weight and threshold values.Then, in order to improve wind speed forecasting results remarkably, a hybrid model based on single models above is adopted. In this model, with the help of wavelet analysis, the wind speed series is decomposed and reconstructed into some sub-sequences. Then, the sub-sequences are respectively predicted by ARMA model and GA-BP model. As the prediction results show, hybrid model can offer much better accuracy than single models ahead of both one hour and several hours, and it is appropriate for data from all the three wind farms.After getting the one-hour-ahead wind speed prediction values, linear fitting methods are adopted to obtain the power curve, so that wind power prediction values can be attained indirectly. The prediction results prove that short-term wind power forecasting based on the hybrid model is practical.