Research On Dispersed Wind Power Forecasting Methods Based On Improved Artificial Bee Colony Algorithm

With the rapid development of wind power, the capacity of wind power interconnection is constantly increasing, especially in recent years, large scale wind farms have already been constructed. In order to excavate new energy power generation, China begins to make full use of dispersed wind farms. However, due to the volatility and uncertainty characteristics of wind speed, wind power generated by wind is also very unstable, which disturbs the power grid. At present, many dispersed wind farms have no equipment for short-term wind power forecasting, which bring many problems to electrical dispatch system. Therefore, it is imperative to develop short-term forecasting methods of dispersed wind power. Accurate forecasting of wind power will provide a reliable foundation for the wind power merging into the grid and play a profound role in the realization of the dispatching automation and on-the-spot guidance.Recently, the short-time wind power forecasting mainly rely on time series, BP neural network and support vector machine. But the time series algorithm has the problem of setting the order, the BP neutral network has the problem of the selection of weight and the algorithm of SVM need set the appropriate parameters. In the light of the present situation and forecasting demand of different wind farms, the improved artificial colony optimization algorithm is incorporated into the three different kinds of wind power prediction methods. First of all, time series method based on improved artificial colony is adopted for short-term wind power prediction, which is directly based on historical data to predict the wind power. It is found that it could decrease the intermediate error term and have a good performance of predicting the power generation after a few minutes or within one hour. Secondly, BP neural network based on improved artificial colony is used to predict the wind power varying from 0 to 4 hour. Results indicate that the method is suitable for the farm in which the historical data is abundant and the demand for prediction accuracy is very high. Finally, because the BP neural network is usually need many data, easy to fall into local optimal and slow convergence speed, support vector machine (SVM) method based on improved artificial colony algorithm is applied to wind power prediction. It is found that the method has the advantages of not needing many data samples, higher accuracy of prediction and avoiding complex learning process. Comparing with other learning algorithms, the method is very suitable for the places where data samples are not many but require higher prediction accuracy.To meet the demand of states grid for developing dispersed wind power grid, a software system has also been developed for a dispersed wind power prediction. The developed software is based on VB. net language. During the treatment of complicated calculation process, it mainly relies on MATLAB dynamic link library files. Results show that the developed software could predict the wind power generation over a period of time on the basis of real-time numerical weather prediction and historical wind resource data. Moreover, it could achieve the function of the real-time monitoring and early warning according to the electrical dispatch data. At last, the proposed three methods are compared and verified through the developed software, which show that each kind of method has its own advantages and can’t be replaced by each other.