期刊文献+

基于TCN-BiLSTM-AM的居民住宅短期电力负荷预测

Short-term residential power load prediction based on TCN-BiLSTM-AM
下载PDF
导出
摘要 针对当前住宅短期电力负荷预测模型存在预测精度低和特征提取困难等问题,提出一种基于TCN-BiLSTMAM的住宅电力负荷预测模型。该模型主要由TCN模型和引入注意力机制层改进的BiLSTM模型组成。首先,通过在历史数据中使用负荷曲线技术计算特征变量的输入权重,以提高数据输入的准确度和关联性;然后,采用权重匹配的方法将数据序列化输入到TCN模型进行采样训练,提取更多不同时间尺度的特征并加快训练速度,同时,构建改进的BiLSTM模型,引入AM层以提高BiLSTM网络结构的运算速度和处理长序列数据的能力,从而提高模型的泛化能力和运算速度;接着,通过对训练好的TCN模型和改进的BiLSTM模型进行加权输出初始预测值,并利用遗传算法对预测值与真实值的偏差进行偏置寻优,得到优化权重并输出最终预测结果。最后,在同一公开数据集上与RNN、LSTM、BiLSTM和TCN等模型进行对比验证,结果表明,相比较其中较好的模型,文中提出的TCN-BiLSTM-AM模型在MAE和RMSE上分别降低了40.43%和35.59%,同时R2指标为0.9957,具有更高的预测精度和更好的稳定性。 A short-term residential power load prediction model based on TCN-BiLSTM-AM is proposed to improve the prediction accuracy and cope with the difficulties in feature extraction in the current residential power load prediction models.This model mainly consists of a TCN(temporal convolutional network)model and a BiLSTM(bi-directional long short-term memory)model improved by an attention mechanism(AM)layer.The input weights of the feature variables are calculated in historical data with the load curve technology,so as to enhance the accuracy and correlation of data input.A weight matching method is employed to serialize the data into the TCN model for sampling training,so as to extract features of more time scales and accelerate the training speed.An improved BiLSTM model is constructed.An AM layer is introduced to enhance the computational speed of the BiLSTM network structure and its ability to process data with long sequence,thereby improving the generalization ability and computational speed of the model.The initially predicted values from the trained TCN model and the improved BiLSTM model are outputted after being weighted,and the genetic algorithm is used to minimize the deviation between the predicted values and the actual values,so as to obtain the optimized weights and the final prediction results.The proposed TCN-BiLSTM-AM model is validated by comparing with the models of RNN(recurrent neural network),LSTM(long short-term memory),BiLSTM and TCN on the same public dataset.The results indicate that,in comparison with the models with better performance,the MAE(mean absolute error)and RMSE(root mean squared error)of the TCN-BiLSTM-AM model is reduced by 40.43% and 35.59%,respectively,and its index R^(2)(coefficient of determination)is 0.9957.It can be seen that the proposed model demonstrates higher prediction accuracy and better stability.
作者 郭渊 张雪成 董振标 李俊杰 GUO Yuan;ZHANG Xuecheng;DONG Zhenbiao;LI Junjie(School of Mechanical Engineering,Shanghai Institute of Technology,Shanghai 201418,China;Network Security Laboratory,ZTE(Nanjing)Co.,Ltd.,Nanjing 211164,China)
出处 《现代电子技术》 北大核心 2024年第19期100-108,共9页 Modern Electronics Technique
基金 国家自然科学基金项目(52201268) 上海市晨光计划项目(20CG66) 上海市青年科技英才扬帆计划项目(21YF1446600)。
关键词 短期负荷 电力预测 TCN BiLSTM 注意力机制 权重匹配 short-term load electric power prediction TCN BiLSTM AM weight matching
  • 相关文献

参考文献7

二级参考文献76

共引文献110

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部