【作者】 刘博；

【导师】 余贻鑫； 张群华；

【作者基本信息】 天津大学 ， 电气工程， 2014， 硕士

【摘要】 电力负荷用电细节监测在优化电力系统运行、规划和管理，指导用户智能用电以节省电费和耗电量，加速能效技术革新和诱发能效市场变革，以及促进全社会把提高生态文明意识落到实处等方面均有重要意义。本文着眼于一种新颖的电力负荷用电细节监测方法——非侵入式电力负荷监测与分解（NILMD）技术的研究，目标是获取总负荷内部每个/类用电设备的工作状态和用电功率这两种分量信息。该技术能够突破目前智能电表仅收集和上送用电总量信息的瓶颈，而且与侵入式电力负荷监测技术相比，具有易操作、低成本、高可靠、数据完整性好和易于迅速推广等优势。在综述课题研究现状和发展趋势的基础上，归纳出了如下基本认识：①负荷印记（LS）的存在和具有重复性是实现NILMD的基本假设和必要条件；②将NILMD问题归结为模式识别问题是合理的，从而可通过类比模式识别系统建立NILMD系统，并确定系统必需的功能模块；③NILMD的基本任务是基于用电设备的每个工作状态都有一个与之对应的LS集合的事实，寻找能够依据用户入口的总电流和/或电压辨识负荷内部各用电设备（类）的工作状态和功率分量的数学模型及求解方法。对于NILMD问题的求解，总结出了仍未完满解决的关键技术问题。基于总负荷工作状态的向量表示，提出了一种非侵入式电力负荷监测与分解的电流模式匹配方法。为完成电流模式（电流谐波特征）最优匹配，该法并未采用优化算法，而是利用基于实例的机器学习方法的优势建立了一种新颖的查表法。大量算例证明该法能够准确监测单个用电设备的工作状态和用电功率，进而可以准确得到分类负荷的用电信息。解决了负荷事件混叠会导致辨识错误的问题，能够有效监测有限多状态设备（FSM）、连续变状态设备（CVSM）和“恒运行”用电设备，提高了辩别相似用电设备的准确率；而且有效地降低了监测系统对微处理器（CPU）计算性能的要求，将有助于合理提高计算速度和削减非侵入式监测系统的经济成本。更多还原

【Abstract】 Electricity consumption details monitoring is of great significance for utilitiesoptimizing the operation, planning and management of the power grid, guiding theintelligent usage of electricity for consumers to save electricity consumption and fee,accelerating energy-efficiency technology innovation and inducing energy-efficiencymarket transformation, and improving the whole society of consciousness ofecological civilization and fulfilling it into practical action.A novel electricity consumption details monitoring technique, Non-IntrusivePower Load Monitoring and Decomposition (NILMD), is focused on with theobjective of obtaining the power consumption and operating state of each (or class of)electrical equipment within a total power load. The NILMD technology can breakthrough the bottleneck facing current smart meters that only the total powerconsumption information is collected and sent above to the data centers. Comparedwith intrusive power load monitoring technology, many advantages are provided bythe non-intrusive one, such as easy operation, low cost, high reliability, good dataintegrity, ease of rapid promotion and so on.Based on the review of the research status and development tend of the subject,some basic understanding is generalized：①The existence and repeatability of theLoad Signature (LS) are the fundamental assumption and necessary condition forimplementing NILMD.②Since it is reasonabe that the NILMD problem isformulated as the pattern recognition problem, the NILMD System can be establishedby the analogy with the pattern recognition system, and the necessary functionalmodules are further determined.③On the basis of the fact that every operating stateof electrical equipment corresponds to a set of LS, the basic task of NILMD is to seekthe mathematical model and solving method, based on which the power component ofeach (class of) electrical equipment can be estimated by the voltage and/or aggregatedcurrent at the customer entry of electrticity. Finally, several key technical problemswhich have not been successfully addressed are summarized.Based on the vectorized representation of the operating state of total load, aNILMD-method built on electric current pattern matching is proposed. In order tocomplete the optimal matching between the measured current pattern (harmoniccharacteristics) and the estimated one, the commom optimization algorithms aren’t utilized, but a novel table-looking method is established in view of the advantages ofthe instance-based learning method. A large number of cases prove that the proposedmethod can accurately monitor the operating state and power consumption of eachelectrical equipment, from which the electricity usage information of every class ofequipment can be easily derived. The proposed method solves the problem thatsimultaneous or aliasing load events would cause false identification, enables theeffective monitoring of finite state machine (FSM), continuous variable state machine(CVSM) and the "always-on" or long-period electrical equipment, and also improvesthe accuracy of discriminating the similar electrical equipment. Furthermore, therequirement of NILMD system for the microprocessor computational performance iseffectively lowered by the suggested method, which is beneficial for reasonablesystem cost reduction.更多还原