“This is the 18th day of my participation in the Gwen Challenge in November. Check out the details: The Last Gwen Challenge in 2021”

preface

  1. Cheng Xiang, LI Linzhi, WU Hao, DING Yi, SONG Yonghua, SUN Weizhen
  2. Key words: intelligent electricity consumption; Non-invasive; Load monitoring; User energy management; Demand response
  3. Key words: Intelligent Power Utilization; non-intrusive; load monitoring; consumption management; demand response
  4. General content: This paper introduces the basic principle and typical framework of non-invasive load monitoring and decomposition: data measurement, data processing, event detection, feature extraction, load identification. In addition, event detection, feature extraction and load identification are introduced, and the algorithms used in each aspect are mainly introduced. Finally, two prospects for NILMD are proposed: the expansion of user types and the development of data sets; Improved performance of load decomposition.

The body of the

Load imprint (LS)

In layman’s terms, the characteristics of each appliance

This LS is defined as the unique information that can reflect the state of power consumption reflected in the operation of an electrical equipment, such as active power waveform. LS is classified into steady state, transient, and operating mode

NILM typical framework

Data measurement

  1. The steady-state and transient signals of total load are mainly obtained.
  2. Measuring error: one is the inconsistency of measuring device; But compression, transmission of original data and other reasons, resulting in data loss

The data processing

Denoising and standardization.

The purpose of standardization is to deal with the disturbance caused by power quality fluctuation easily

Event detection

Event detection is based on the change of load imprint in a certain period of time, and there are two specific methods: rule judgment and change point detection

Event change point detection method: Generalized Likelihood Ratio (GLR) test, sequential probability ratio test (SPRT) signal mutation detection, artificial immune algorithm considering Fisher’s criterion

Feature extraction

The following table summarizes the feature extraction techniques based on steady-state and transient features.

The feature extraction technology that comprehensively considers steady-state and transient characteristics of load has absorbed the advantages of both and proposed some new algorithms:

  • Select the current, voltage, active power, reactive power combination of 7 LS, the Fourier transform to achieve feature extraction, found that the combination of four LS can obtain the highest accuracy
  • Load decomposition is carried out by using current waveform, active power, reactive power and other steady-state characteristics and on active power transient waveform. The results show that this steady-state and transient integrated multi-feature extraction method
  • The two-layer feature extraction framework determines the type of electrical appliance through the steady-state feature of the first layer, and then distinguishes the electrical equipment through the steady-state feature, transient feature and user behavior mode of the second layer

Non-traditional characteristic indicators were used:

  • The open transient energy is used to distinguish different electrical appliances with the same active and reactive power
  • Select various marks of a single electrical equipment in the running period, including edge marks, sequence marks, trend marks, persistence marks, etc

Load identification

Other methods:

  1. Supervised self-organizing map and Bayesian hybrid recognizer are used
  2. Fuzzy C-means based on particle swarm optimization is combined with neural Fuzzy classification method to identify household appliances
  3. Improved implicit Markov model with differential observation

NILMD application

  1. Fault detection and diagnosis
  2. Optimization with energy strategy

The prospect of

  1. At present, there are few open data sets in China, and the foreign data sets are REDD BLUED AMPds UK-DALE
  2. Improved performance of load decomposition

Past wonderful articles

  • Cow guest latest front-end JS written test 100 questions
  • The latest front end of the interview questions summary (including analysis)
  • Grab the latest front end test five hundred data analysis JS interview hot spots
  • Happy cat stroking for VSCode and website adoptive cats
  • Native JavaScript soul Test (1), how much can you answer?
  • A thorough understanding of prototypes and prototype chains in JavaScript
  • Complete understanding of EventLoop in JavaScript
  • “2W word big chapter 38 interview questions” completely clear JS this pointing to the problem

After the language

Guys, if you find this article helpful, give a like to 👍 or follow ➕ to support me.

In addition, if this article has a question, or do not understand part of the article, you can reply to me in the comment section, we come to discuss, learn together, progress together!

If you feel that the comment section does not understand, you can also add my wechat (li444186976) or QQ (3315161861) for detailed communication, the name is battlefield small bag.