基于PCC提取负荷周期特征与时变趋势特征的精准短期负荷预测OA
Accurate Short-term Load Forecasting Based on PCC Extracted Load Cyclic Characteristics and Time-varying Trend Characteristics
为提升负荷预测精度和预测模型的泛化能力,提出融合负荷周期特性与负荷时变趋势特性的卷积神经网络-长短期记忆网络混合预测模型(C-T-CNN-LSTM).首先,使用皮尔逊相关系数量化各采样时刻负荷序列之间的相关性强弱,筛选出可以反映电力负荷固有周期演化规律的关键特征参量.然后,基于通过分析周期负荷数据时序变化特点,提出负荷时变趋势概念,并建立负荷时变趋势模型,提取负荷时变趋势特征.最后,将负荷周期特征变量与负荷时变趋势特征变量重构生成特征集,并输入到 CNN-LSTM 混合模型中进行预测.以美国新英格兰地区 2015 年电力负荷数据集为例,采用 9 种预测模型进行仿真验证与对比分析,结果表明,提出的 C-T-CNN-LSTM 模型拥有最高的预测精度.
To enhance the forecasting accuracy and generalization capability of prediction models,this paper proposes a hybrid forecasting model(C-T-CNN-LSTM)that integrates load cyclic characteristics and time-varying trend characteristics into a convolutional neural network and long short-term memory(CNN-LSTM)network.Firstly,the Pearson correlation coefficient(PCC)is employed to quantify the correlation strength among load sequences at different sampling instants,thereby selecting key feature vari-ables that can reflect the inherent periodic evolution patterns of electric power load.Subsequently,based on an analysis of the temporal variation characteristics of periodic load data,the concept of load time-varying trend is introduced,and a corresponding model is established to extract the time-varying trend features.Finally,the load cyclic feature variables and the time-varying trend feature variables are reconstructed to form a feature set,which is then fed into the CNN-LSTM hybrid model for prediction.Taking the 2015 e-lectric load dataset of the New England region of the United States as a case study,nine prediction models are employed for simulation validation and comparative analysis.The results demonstrate that the proposed C-T-CNN-LSTM model achieves the highest prediction accuracy among all compared models.
贾燕峰;俞飞;李飚;严亚帮;乔澳雪;季玉琦
国网河南省电力公司三门峡供电公司,河南 三门峡 472000国网河南省电力公司三门峡供电公司,河南 三门峡 472000国网河南省电力公司三门峡供电公司,河南 三门峡 472000郑州轻工业大学电气信息工程学院,河南 郑州 450002郑州轻工业大学电气信息工程学院,河南 郑州 450002郑州轻工业大学电气信息工程学院,河南 郑州 450002
信息技术与安全科学
时变趋势特征提取皮尔逊相关系数CNN-LSTM混合神经网络短期负荷预测
time-varying trend feature extractionPearson correlation coefficientCNN-LSTM hybrid neural networkshort-term load forecasting
《机械与电子》 2026 (8)
117-126,10
国网河南省电力公司科技项目资助(5217I0250004)
评论