基于时空神经网络的风电场超短期风速预测模型

Ultra-short-term Wind Speed Prediction Model for Wind Farms Based on Spatiotemporal Neural Network

随着风电场的大规模接入,提高风电场风速的预测精度对于促进可再生能源的消纳具有重大意义。传统的预测方法通常根据风电场单一高度的历史风速进行预测,当预测的时间尺度达到三四小时的时候,预测误差较大。不同高度的风速、风向数据蕴含了风电场内部的时空相关性,数值天气预报数据也体现了风电场周边的大气运动对风速发展规律的影响。文中在输入数据层面,同时引入了不同高度的风速、风向数据和数值天气预报数据。为了充分挖掘数据中的规律,提出了一种新的时空神经网络,采用深度卷积神经网络和双向门控循环单元,分别提取风速、风向等历史数据以及数值天气预报的时空特征,并利用融合后的特征进行风速预测。最后,利用中国东北某风电场的实际测量数据,验证了算法的有效性。

With the large-scale integration of wind farms, improving the prediction accuracy of wind speed in wind farms is of great significance to promote the consumption of renewable energy. Traditional prediction methods are usually based on the historical wind speed of a single altitude in the wind farm. When the prediction horizon reaches about three or four hours, the prediction error becomes relatively large. Wind speed and direction data at different altitudes contain the spatiotemporal correlation and the numerical weather prediction data reflects the influence of atmospheric motion around the wind farm on the variation of wind speed.In this paper, wind speed and direction data at different altitudes and numerical weather prediction data are introduced at the input data level. In order to fully exploit the rules of data, a new spatiotemporal neural network(STNN) is proposed. The deep convolutional network and the bidirectional gated recurrent unit are used to extract the spatiotemporal features of historical wind speed, wind direction and numerical weather prediction, respectively. The fused features are used to predict the wind speed.Finally, the actual measurement data of a wind farm in northeast China is used to verify the effectiveness of the algorithm.