基于MSTAGNN模型的可解释海上风电场多风机出力预测

Interpretable Power Output Prediction of Multiple Wind Turbines for Offshore Wind Farm Based on Multiple Spatio-temporal Attention Graph Neural Network Model

现有海上风电场出力预测研究对复杂时空关系考虑不足,且多为“黑盒”模型,缺乏可解释能力。为充分挖掘时空关联并实现模型可解释,提出一种基于多重时空注意力图神经网络(MSTAGNN)的海上风电场出力预测模型。首先,构建了一种考虑空间关联的海上风电场图拓扑,并引入空间注意力机制实现图拓扑的动态变化;其次,分别利用图卷积网络和时间门控卷积网络有效提取空间和时间特征;接着,对所提模型引入多维多头注意力机制,使其获得多重可解释能力;最后,基于中国东海大桥风电场真实数据进行仿真验证。结果表明,所提模型相比传统预测模型具有更高的预测精度,同时在空间、特征、时间多个维度具有合理的可解释性。

Existing research on the power output prediction of offshore wind farms does not take into account the complex spatiotemporal relationship,and the prediction models are mostly“black boxes”,lacking the interpretable ability.To fully exploit the spatio-temporal correlation and realize the interpretability of the model,this paper proposes a power output prediction model of offshore wind farms based on multiple spatio-temporal attention graph neural network(MSTAGNN).Firstly,a graph topology of the offshore wind farm considering the spatial correlation is proposed,and the spatial attention mechanism is introduced to realize the dynamic change of graph topology.Secondly,the graph convolutional network and temporal gated convolution network are separately used to extract spatial and temporal features.Then,the multi-dimensional and multi-head attention mechanism is introduced into the proposed model to obtain multiple interpretable abilities.Finally,based on the real data of 34 wind turbines in Donghai Bridge wind farm,China,this paper conducts the simulation verification.The results show that,compared with traditional prediction models,the proposed model has higher prediction accuracy,and has reasonable interpretability in multiple dimensions of space,feature and time.