基于Seq2seq模型的多特征短期母线负荷预测

Short-term Bus Load Forecasting of Multi-feature Based on Seq2seq Model

母线负荷短期预测对电力系统调度运行及电力现货交易市场具有重要意义,由于母线负荷受到复杂多因素的影响,单独采用母线负荷历史数据进行预测不能取得很好的效果。为提升多因素影响下母线负荷预测的准确率,本文结合特征工程和深度学习法,对母线负荷的影响因素进行斯皮尔曼相关性分析,设计时间连续性周期编码;对Seq2seq模型的编码器和解码器设置不同的特征组进行消融实验;将实验结果与采用离散小波变换分解提取特征的方法进行对比,结果表明,本文提出的特征工程结合深度学习Seq2seq框架的母线负荷短期预测效果更佳。

Short-term bus load forecasting is of significance for power system dispatching operation and the power spot trading market. Since the bus load is affected by many complex factors,the forecasting with its historical data alone cannot achieve satisfying results. To improve the forecasting accuracy of bus load under the influence of many factors,feature engineering and deep learning methods are combined in this paper. First,Spearman correlation is used to calculate the related influencing factors of bus load. The time continuity periodic coding is designed,and different feature groups of the encoder and decoder in the Seq2seq model are set for ablation experiments. The experimental results are compared with the those obtained by the feature extraction method of discrete wavelet transform,indicating that the shortterm bus load forecasting based on feature engineering combined with the deep learning Seq2seq framework proposed in this paper is better.