基于空间深度置信网络的风速预测优化方法

The improvement of wind speed prediction using spatial deep belief network

风能是目前应用最为广泛、技术最为成熟的可再生能源。为了保证风电场的稳定和安全运行,风速的准确预测至关重要。除传统的数值天气预报以外,机器学习技术已经广泛应用于不同时间尺度的风速预测。然而这些工作大多局限于单一地点的风速序列分析,没有考虑和利用风速的空间相关性。对此,使用深度置信网络(Deep Belief Network,DBN)对同一区域内多个地点的风速序列进行空间相关性特征识别。在训练过程中,深度置信网络充分挖掘了该区域内历史风速的联合分布,借此改善未来的风速预测。多组风速预测实验表明,空间深度置信网络能够有效降低风速的预测误差,经过空间深度置信网络重构后的风速预测误差平均降低了0.4 m/s。

Wind energy is the most widely used renewable energy.Accurate wind speed prediction is critical for the safety and stability of wind power system.Besides traditional numerical weather prediction,the machine learning technique has been used in wind speed prediction of different time scales.However,most previous studies focused on the wind speed sequence of single station and ignored the spatial dependency and correlation of wind.To improve the prediction with spatial information,this paper tries to extract the wind spatial correlation features in one region area and reconstruct the wind speed using deep belief network(DBN).The experiment results of different regions prove that the spatial deep belief network can reduce the prediction error significantly and increase the accuracy of wind speed prediction by 0.4 m/s on average.