基于横纵向集成学习的短期负荷预测方法

Load Forecasting Method Based on Transverse and Longitudinal Integrated Learning

为进一步提高预测的准确度和普适能力,并降低组成算法的复杂度,对负荷的固有特性进行横纵向二维化分析,结合误差分布的特点,提出基于横纵向剖析负荷特性的集成预测方法。初级模型采用互信息提取横向特征,通过长短期记忆网络(LSTM)感知负荷波动;采用变分模态分解(VMD)提取纵向特征,通过Elman神经网络预知负荷趋势;然后基于改进的Stacking融合构建横纵向集成学习模型。最后,采用中国东部某地区的负荷数据验证模型的有效性,算例表明改进的Stacking充分融合了横纵向模型的优势并具备强大的学习小样本能力,横纵向集成预测方法有效提高了模型的预测精度和泛化能力。

For improving the accuracy and general applicability of the prediction and reducing the complexity of the algorithm,the inherent characteristics of the load is analyzed in two dimensions from transverse and longitudinal directions,combined with the error distribution,an integrated forecasting method based on learning load characteristics from transverse and longitudinal directions is proposed.For the primary model,mutual information is used to extract transverse features,then are learned by LSTM to perceive the load fluctuation.VMD is used to extract longitudinal features,then are learned by Elman to predict the load trend.Next,the improved Stacking integrated algorithm is used to fuse and construct the transverse and longitudinal integrated model.Finally,the validity of the model is verified by the load data of a certain area of eastern China.The results show that the improved Stacking fully integrates the advantages of transverse and longitudinal models and has strong learning ability of small samples,the transverse and longitudinal integrated forecasting method can effectively enhance the prediction accuracy and generalization ability of the model.