引入BL-Seq2seq模型进行负荷预测

Introducing the BL-Seq2seq Model for Load Forecasting

短期负荷预测在电力系统运行和调度中起着重要作用,为了更好地提取数据中蕴含的有效信息,提升短期负荷预测精度,本文引入Seq2seq算法的注意力机制提出了多层Bi-LSTM的Seq2seq深度学习模型(BL-Seq2seq)实现短期用电负荷预测。其中Seq2seq的编码端由多层Bi-LSTM组成,将输入数据进行编码,并在网络末端输出编码后的最终状态;Seq2seq解码端为单层LSTM,它将编码端的最终状态作为初始输入状态,同时每一步的输出值作为下一步的输入值。利用用电负荷实测数据,基于Keras平台进行仿真,仿真结果表明,与多个经典的深度学习的短期用电负荷预测模型相比,所提BL-Seq2seq模型的预测误差明显降低,大大提升了短期用电负荷预测精度。

Short-term load forecasting plays an important role in power system operation and scheduling.In order to better extract the effective information contained in the data and improve the accuracy of short-term load forecasting,this paper introduces the attention mechanism of Seq2 seq algorithm and proposes a multi-layer Seq2 seq of Bi-LSTM deep learning model(BL-Seq2 seq)to achieve short-term load forecasting.The encoding end of seq2 seq consists of a multi-layer Bi-LSTM,which encodes the input data and outputs the final state after encoding at the end of the network;the decoding end of Seq2 seq is a single layer LSTM,which takes the final state of the encoding end as the initial input state and the output value of each step as the next input value.The simulation results show that compared with several classic deep learning short-term load forecasting models,the prediction error of BL-Seq2 seq model is significantly reduced and the accuracy of short-term load forecasting is greatly improved.