基于一种新技术的风电功率短期预测

Short Term Prediction of Wind Power Based ona New Technology

近年来,循环神经网络(Recurrent Neural Network, RNN)等序列化模型越来越多地应用到风电功率预测任务中,鉴于序列化模型在处理长距离依赖关系时的固有缺陷,提出一种结合Transformer模型和端到端记忆网络(End-To-End Memory Networks, MemN2N)的预测方法。利用Transformer挖掘历史数据中的长距离依赖性信息,并将编码结果引入到MenN2N网络的记忆池中。为了进一步增强模型在多步预测中的稳定性,基于注意力机制对MenN2N网络的输入组件和输出组件加以改进,对连续多步天气数据之间的不确定性进行建模,实现对风电功率在短期内的多步预测。通过风场中实际数据进行测试,实验结果表明,与其它预测方法相比,所提方法在多步预测中具有更高的预测精度和稳定性,具有很大的工程价值。

In recent years, serialization models such as Recurrent Neural Network(RNN) are more and more used in wind power prediction tasks. In view of the inherent defects of serialization model in dealing with long-distance dependence, this paper proposes a new prediction method combining Transformer model and End-To-End Memory Networks(MemN2 N). We use Transformer model to mine the long-distance dependency information in historical data, and introduce the coding results into the memory pool of MemN2 N network. In order to further enhance the stability of the model in the multi-step prediction, the input component and output component of the MemN2 N network are improved based on the attention mechanism, and the uncertainty between the continuous multi-step weather data are modeled to realize the multi-step prediction of wind power in the short term. Through the test of the actual data in the wind field, the simulation results show that, compared with other prediction methods, the method proposed in this paper has higher prediction accuracy and stability in multi-step prediction, and has great engineering value.