基于聚类分析方法的风电场日前功率预测研究

Research on Day-ahead Power Forecast of Wind Farm Based on Cluster Analysis

随着我国风电装机容量的逐步增加,风电在电源结构中的比例也进一步增大,而风电出力本身的随机性、波动性使得风电出力预测工作难度增加。鉴于不同特性的气象数据对应的出力差异性很大,因此采用模糊C均值聚类分析方法将风电场运行历史气象信息进行聚类并训练相应的BP神经网络,再将数值天气预报提供的待预测日的气象信息进行聚类,根据数据的聚类中心将待预测日气象信息与历史数据进行归类,最后将同一类风速、风向数据结合相应的BP神经网络进行预测。算例分析证明了此方法的有效性。

With the increase of installed capacity of wind power in China, the proportion of wind power in the energy mix further increases. Due to its randomness and fluctuation, it is difficult to forecast the wind power output. In the wind farm, different meteorological data result in differences in wind power. So historical data of wind farms are gathered by using cluster analysis of fuzzy C-means; then the BP neural network is trained by the clustered data. The next stop is clustering the meteorological data provided by the NWP of next day.Every clustered data has a cluster center. The clustered data of the two days are classified according to the cluster centers. Finally, the same type of clustered data of wind speed and direction are used to simulate by using the BP neural network. The calculation example verifies the effectiveness of the method.