基于启停状态识别改进因子隐马尔可夫模型的非侵入式负荷分解  

Non-intrusive Load Disaggregation With Improved Factorial Hidden Markov Model Considering ON-OFF Status Recognition

作　　者：于超 覃智君[1] 阳育德[1] YU Chao;QIN Zhijun;YANG Yude(School of Electrical Engineering,Guangxi University,Nanning 530004,Guangxi Zhuang Autonomous Region,China)

机构地区：[1]广西大学电气工程学院,广西壮族自治区南宁市530004

出　　处：《电网技术》2021年第11期4540-4550,共11页Power System Technology

基　　金：国家自然科学基金项目(51767001)。

摘　　要：非侵入式负荷分解(non-intrusiveloaddisaggregation,NILD)是通过对总电气量分析得到用户家中各电器设备的能耗数据,该数据为节能、需求响应及公共安全等方面提供重要依据。目前,NILD算法存在因设备启停状态识别准确率低导致负荷分解精度低的问题。为此,该文提出一种基于设备启停状态识别改进因子隐马尔可夫模型(factorialhidden Markov model,FHMM)的NILD方法。首先,通过人工少数类过采样法(syntheticminorityover-samplingtechnique,SMOTE)对训练数据做重采样处理、深度神经网络(deep neuralnetwork,DNN)模型提取启停状态特征以及双向长短时记忆网络及条件随机场(bidirectionallongshortterm memory-conditionalrandomfield,Bi LSTM-CRF)模型提升其对不平衡启停状态的识别能力;然后,将设备启停状态组合模块按照数理组合方法划分数据集,形成若干启停状态组合子数据集,并分别对各子数据集中处于启状态的设备建立FHMM进行负荷分解;最后,在公开数据集每分钟电力年鉴数据集(the almanac of minutely power dataset,AMPds)进行实验,该文算法得到的设备负荷分解平均精度比传统FHMM方法提升了3.8倍,验证了所提方法的有效性和准确性。Non-intrusive load disaggregation(NILD) is an approach to obtain the energy consumption data of each electrical appliances in the user’s house by the total electrical capacity, which can provide an important basis for energy conservation, demand response and public safety. At present,the NILD algorithm faces the problem of low load disaggregation accuracy due to the low accuracy of appliances ON-OFF state recognition In this paper, an NILD method with improved factorial hidden Markov model(FHMM) considering the ON-OFF state recognition is proposed. First of all, the training data are resampled by SMOTE, the ON-OFF state characteristics extracted by DNN model and ability to recognize unbalanced ON-OFF state is improved by BiLSTM-CRF model. Secondly, the appliances’ ON-OFF state combination module is split to form several ON-OFF state combination sub-datasets according to the mathematical combination method. An FHMM is established for load disaggregation of the appliances in the ON-OFF state of each sub-dataset. Finally, experiments were carried out on the open dataset AMPds. The average accuracy of the load disaggregation of each appliances obtained by the algorithm in this paper is 3.8 times higher than that by the traditional FHMM method, which verifies the effectiveness and accuracy of the proposed model.

关 键 词：非侵入式负荷分解 不平衡启停状态特征提取 启停状态识别 启停状态组合 因子隐马尔可夫模型 

分 类 号：TM721[电气工程—电力系统及自动化]