纸质出版:2026
移动端阅览
张建文,努尔麦麦提·如则,山宪武,邹晶梅,徐瑞翔,万 升,袁培森.基于ConvBERT-BiLSTM模型的非侵入式负荷分析研究[J].智慧电力,2026,54(2):98-105.
doi:10.20204/j.sp.2026.02012
张建文,努尔麦麦提·如则,山宪武,邹晶梅,徐瑞翔,万 升,袁培森.基于ConvBERT-BiLSTM模型的非侵入式负荷分析研究[J].智慧电力,2026,54(2):98-105. DOI: 10.20204/j.sp.2026.02012.
doi:10.20204/j.sp.2026.02012 DOI:
在电力系统加速数智化转型的背景下,非侵入式负荷监测因其低成本、易部署的优势,成为提升能源管理精细化水平的关键技术。然而,现有方法多依赖人工特征或启发式规则,难以在复杂用电场景下兼顾局部特征提取与全局时序依赖建模,导致负荷分解精度受限。为此,提出一种融合卷积通道注意力、双向Transformer编码器(BERT)与改进双向长短期记忆网络(BiLSTM)的非侵入式负荷分析模型ConvBERT-BiLSTM。该模型首先通过卷积层提取功率序列的局部特征,并引入通道注意力机制强化关键特征通道;继而利用BERT的多头自注意力机制捕捉长距离全局依赖,再结合BiLSTM增强对短时上下文动态的感知能力;最后通过转置卷积层实现特征上采样与功率序列重构。在公开UK-DALE数据集上的实验结果表明,所提方法在负荷分类任务中的准确率、精确率、召回率和F1分数分别达到98.45%,89.73%,88.61%和88.83%,显著优于对比模型,同时有效提升了复杂用电场景下负荷分解的准确性与鲁棒性,为电力系统数智化监测提供了可靠的技术支撑。
Against the backdrop of accelerating digital and intelligent transformation in power systems
non-intrusive load monitoring (NILM) has emerged as a key technology for enhancing the granularity of energy management
owing to its advantages of low cost and ease of deployment. However
existing methods often rely on manual feature extraction or heuristic rules
making it difficult to simultaneously capture local feature extraction and global temporal dependency modeling in complex electricity consumption scenarios
which limits load decomposition accuracy.To address this
a novel non-intrusive load analysis model named ConvBERT-BiLSTM is proposed
integrating convolutional channel attention
a bidirectional Transformer encoder (BERT)
and an improved bidirectional long short-term memory network (BiLSTM). The model first extracts local features from power sequences via convolutional layers and incorporates a channel attention mechanism to enhance key feature channels.It then utilizes the multi-head self-attention mechanism of BERT to capture long-range global dependencies
followed by BiLSTM to strengthen the perception of short-term contextual dynamics. Finally
feature upsampling and power sequence reconstruction are achieved through a transposed convolutional layer. Experimental results on the public UK-DALE dataset show that the proposed method achieves accuracy
precision
recall
and F1-score of 98.45%
89.73%
88.61%
and 88.83%
respectively
in load classification tasks
significantly outperforming comparative models. The proposed model effectively improves the accuracy and robustness of load decomposition in complex electricity consumption scenarios
providing reliable technical support for the digital and intelligent monitoring of power systems.
华昊辰, 张洲赫,邹奕群,等. 新型电力系统需求侧灵活性资源低碳协同优化研究综述 [J ] .电力建设,2025,46(6):60-75.
张帅,程志友,田甜,等. 基于马尔可夫转移场和轻量级网络的非侵入式负荷识别 [J ] .电力系统保护与控制,2024,52(17):51-61.
沈鑫,王钢,赵毅涛,等. 融合SENet注意力机制和GA-CNN的非侵入式负荷识别方法 [J ] .中国电力,2025,58(5):33-42.
刘航,刘春阳,赵浩然,等. 基于深度学习的公共楼宇电-气负荷非侵入式分解 [J ] .电网技术,2023,47(3):1188-1197.
白宏坤,卜飞飞,李文峰,等. 基于复合分析算法的工业用户非侵入监测研究 [J ] .河南电力,2021(S1):43-47.
李辉,高嘉颉,席荣军,等. 基于改进GAF-inception网络的非侵入式工业负荷识别算法 [J ] .电力科学与技术报,2025,40(4):103-112.
LIN S F,ZHAO B,ZHAN Y F,et al. Non-intrusive residential load identification based on load feature matrix and CBAM-BiLSTM algorithm [J ] .Frontiers in Energy Research.2024,12:1443700
何健明,李梦诗,张禄亮,等. 基于Attention和残差网络的非侵入式负荷监测 [J ] .电测与仪表,2024,61(6):173-180.
XU W Y,JIANG C Q,ZHANG Q H,et al. An intelligent non-intrusive load monitoring model based on power encoding and convolutional state modules [J ] .Measurement Science and Technology,2024,35(8):085801.
李新国,杨轩,程少靖,等. 面向多类型用户负荷的需求响应潜力量化评估 [J ] .智慧电力,2024,52(9):56-64.
张晋波,张昭维,武建文,等. 基于VMD-BiLSTM的高压断路器故障多信号联合辨识与智能诊断 [J ] .高压电器,2025,61(7):34-43.
杨勇,陈永辉,郝一浩,等. 非侵入式负荷识别研究综述 [J ] .电子设计工程,2025,33(14):34-41.
廖耀华,常艳平,王恩,等. 基于线性复杂度自注意力机制的非侵入式负荷监测方法 [J ] .电测与仪表,2025,62(8):197-205.
孙睿晨,董坤,赵剑锋,等. 基于递进式模型结构和时间信息嵌入的非侵入式负荷分解 [J ] .智慧电力,2024,52(2):55-62,70.
李利刚,刘浩,陈建强,等. 基于最优特征和改进随机森林的非侵入式负荷辨识方法 [J ] .电力需求侧管理,2024,26(3):55-61.
赵学明,王宏伟,吴宇彤,等. 基于时域分析特征的非侵入式负荷事件检测方法 [J ] .浙江电力,2025,44(3):62-70.
杨赟辉,程虎,魏敬和,等. 面向Transformer模型边缘端部署的常用激活函数高精度轻量级量化推理方法 [J ] .电子学报.2024,52(10):3301-3311.
TAO Y C,QIU J,LAI S Y,et al. Reserve evaluation and energy management of micro-grids in joint electricity markets based on non-intrusive load monitoring [J ] .IEEE Transactions on Industry Applications,2023,59( 1):207-219.
GUO W,LIU S,WENG L,et al. Power grid load forecasting using a CNN-LSTM network based on a multi-modal attention mechanism [J ] .Applied Sciences,2025,15(5):2435.
YUAN J,JIN R,WANG L D,et al. A nonintrusive load identification method based on dual-branch attention gru fusion network [J ] .IEEE Transactions on Instrumentation and Measurement,2024,73:1-10.
0
浏览量
4
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621