Study On The Prediction Of The Wind Speed Series

Wind power, as a kind of promising renewable energy, gets widespread attention around the world, and it is applied in the power generation industry instead of the traditional fossil fuels to reduce carbon emissions. However, because the wind has the characteristics of intermittent and uncontrollability and randomness, wind power in large scale grid causes huge impact to the power grid, which has become one of the main bottlenecks limiting wind power utilization. The present study shows that wind speed series, as a meteorological data, has good predictability. Therefore, the wind speed prediction has become one of the valid methods to resolve the problem above.The prediction methods of the wind speed series are studied in this thesis. First, the principle of BP neural network is investigated, and hysteretic neural network, combining the hysteresis in the natural biological nervous system, is proposed to improve the prediction performance of the forward neural network. Second, combing the chaos theory, the prediction of the wind speed series with chaotic characteristic is investigated. Finally, hybrid prediction methods are proposed to overcome the disadvantages of prediction method, such as different prediction mechanism, losing information in the prediction process. The work can be summarized as follows:(1) The principles of times series analysis method and BP neural network are analyzed. The ordering method of ARIMA model is investigated. Based on the BP neural network, hysteretic characteristic is brought into neural network, and hysteretic neural network is proposed to overcome some problems in BP neural network, such as local minimum problem, and false saturation problem. The hysteretic characteristic can enhance the utilization of the information and the storage ability, so it can improve the generalization ability and prediction performance of the prediction network.(2) The prediction of the wind speed series with chaotic characteristic is investigated based on chaos theory. Chaotic operator network is proposed to perform the prediction. The characteristic of the network can be changed to follow to that of the predicted wind speed series by training the parameters of the network. In this way, the prediction network with the similar characteristic to the predicted wind speed series can be constructed, and its prediction performance can be improved. Furthermore, the best unstable period contained in the wind speed series with chaotic characteristic can be determined. And the prediction of the wind speed can be done by the unstable period. Because the unstable period is longer than prediction steps, the relatively stable prediction results can be obtained when the prediction step is limited in some range.(3) Two hybrid prediction methods, weighted hybrid prediction method and time-sharing hybrid prediction method, are proposed to overcome the disadvantages, different mechanism and losing information, of single prediction method. Furthermore, the fusion prediction method, combining Kalman Filter method, is proposed. Hysteretic neural network and ARIMA model are fused by Kalman Filter to perform the prediction of the wind speed series. Besides, the state equation on the wind speed and acceleration is established. The wind speed series and acceleration series are predicted separately, the optimal prediction estimation of the wind speed series can be obtained by Kalman Filter method.