基于维特比算法改进的稳暂态混合非侵入式负荷识别方法  

An Improved Steady-and Transient-State Mixed Non-Intrusive Load Monitoring Using Viterbi Algorithm

作　　者：刘宇[1] 刘丛笑 赵欣[1] 高山[1] 黄学良[1] Liu Yu;Liu Congxiao;Zhao Xin;Gao Shan;Huang Xueliang(School of Electrical Engineering Southeast University,Nanjing 210018 China)

机构地区：[1]东南大学电气工程学院,南京210018

出　　处：《电工技术学报》2023年第19期5241-5255,共15页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金(51907024);东南大学“至善青年学者”支持计划和中央高校基本科研业务费专项资金(2242022R40050)资助项目。

摘　　要：非侵入式负荷识别技术因其具有诸多优势,成为目前负荷运行状态监测的主流方法。通常可将其分为基于稳态特征的负荷识别和基于暂态特征的负荷识别。考虑相邻稳态过程和暂态过程之间的负荷运行状态开关逻辑,该文构建了时序校核模型和策略,以此排除不可信的结果以提高识别精度。首先,分别采用离散粒子群优化算法和动态时间规整进行基于稳态特征和暂态特征的负荷识别;然后,基于概率评价选取多个识别结果构建负荷识别结果候选集,联立多个连续的稳态和暂态过程的候选集,基于维特比算法构建时序识别结果的概率序列和评价方法,并进行优选以确定最终的识别结果;最后,分别在仿真数据集和实测数据集上对该方法进行验证分析。实验结果表明,该方法能有效提高整体的负荷识别精度,并明显改善大功率负荷的识别效果,同时保持小功率负荷的识别准确率。Non-intrusive load monitoring(NILM)is the most commonly used method to achieve load state identification,which is an important technology to realize power grid panoramic perception and support the carbon peaking and carbon neutrality goals.Depending on the detection object,NILM can be divided into two categories,i.e.,steady-state signature based NILM and transient-state signature based NILM.Various studies have been conducted on both fields respectively,but the internal relationships between different states are rarely discussed.To explore the potential of associating the sequential states in load disaggregation problem,this paper thoroughly investigates the sequential logics between the load states in adjacent steady-state process and transient-state process,and makes use of them to improve the NILM performance.Firstly,considering the switching states of appliances,steady-state signature based NILM is solved by discrete particle swarm optimization(DPSO)algorithm.Secondly,transient-state signature based NILM is addressed by dynamic time warping(DTW)approach,to deal with the complex event characteristics.Then,a probability evaluation system is applied for decision-making,where multiple identification results with high confidence are selected to construct the candidate sets of independent identification results.Lastly,the candidate sets of adjacent steady-state and transient-state are associated together,and Viterbi algorithm is utilized to establish the probabilistic sequential model and optimize the load identification results.The proposed method is analyzed and validated on load consumption data from both low voltage network simulator(LVNS)and UK-DALE dataset.Four metrics,including F1 measure(F1),root mean square error(RMSE),mean absolute error(MAE),and normalized mean square error(NMSE),are utilized to evaluate the load disaggregation performance.The results show that the proposed method can effectively improve the overall load identification accuracy,especially for these appliances with high rated power.In the

关 键 词：非侵入式负荷识别 离散粒子群优化算法 动态时间规整 维特比算法 稳暂态 

分 类 号：TM714[电气工程—电力系统及自动化]