Nonintrusive Appliance Load Monitoring 组长:辛美惠 组员:赵蓓,姚宁,刘铸 2012-6-18.

Slides:



Advertisements
Similar presentations
Applications of one-class classification
Advertisements

WATERSENSE: WATER FLOW DISAGGREGATION USING MOTION SENSORS Vijay Srinivasan, John Stankovic, Kamin Whitehouse Department of Computer Science University.
Management and Control of Domestic Smart Grid Technology IEEE Transactions on Smart Grid, Sep Albert Molderink, Vincent Bakker Yong Zhou
QR Code Recognition Based On Image Processing
Marković Miljan 3139/2011
Data Mining Methodology 1. Why have a Methodology  Don’t want to learn things that aren’t true May not represent any underlying reality ○ Spurious correlation.
Home Area Networks …Expect More Mohan Wanchoo Jasmine Systems, Inc.
FixtureFinder: Discovering the Existence of Electrical and Water Fixtures Vijay Srinivasan*, John Stankovic, Kamin Whitehouse University of Virginia *(Currently.
Digital Signal Processing
Introduction The aim the project is to analyse non real time EEG (Electroencephalogram) signal using different mathematical models in Matlab to predict.
ISSBN, Electronic faculty of Niš, November Use of distortion power for side identification of the harmonic polution Dejan Stevanović, Electronic.
Classification of Music According to Genres Using Neural Networks, Genetic Algorithms and Fuzzy Systems.
Report on Intrusion Detection and Data Fusion By Ganesh Godavari.
A Study of Approaches for Object Recognition
Basic Concepts and Definitions Vector and Function Space. A finite or an infinite dimensional linear vector/function space described with set of non-unique.
Smart Traveller with Visual Translator for OCR and Face Recognition LYU0203 FYP.
Per Unit Representation Load Flow Analysis Power System Stability Power Factor Improvement Ashfaq Hussain.
Laurent Itti: CS599 – Computational Architectures in Biological Vision, USC Lecture 7: Coding and Representation 1 Computational Architectures in.
KING FAHAD UNIVERSITY OF PETROLUEM & MINERALS DEPARTMENT OF ELECICAL ENGINEERING EE-306 PROJECT REACTIVE POWR PREPARE BY Yasre Ahmed Saleh ID#
Chapter 27 Lecture 12: Circuits.
IIS for Image Processing Michael J. Watts
1 Template-Based Classification Method for Chinese Character Recognition Presenter: Tienwei Tsai Department of Informaiton Management, Chihlee Institute.
Presented by Tienwei Tsai July, 2005
Multiuser Detection (MUD) Combined with array signal processing in current wireless communication environments Wed. 박사 3학기 구 정 회.
1 Novel Capabilities of Power Quality Monitoring at the Smart Grid Netzah Calamaro, Yuval Beck, Doron Shmilovitz Israel Electric Company, TAU energy conversion.
Report on Intrusion Detection and Data Fusion By Ganesh Godavari.
Image Classification 영상분류
Basics of Neural Networks Neural Network Topologies.
Phase Congruency Detects Corners and Edges Peter Kovesi School of Computer Science & Software Engineering The University of Western Australia.
11/25/2015 Wireless Sensor Networks COE 499 Localization Tarek Sheltami KFUPM CCSE COE 1.
A NOVEL METHOD FOR COLOR FACE RECOGNITION USING KNN CLASSIFIER
Z bigniew Leonowicz, Wroclaw University of Technology Z bigniew Leonowicz, Wroclaw University of Technology, Poland XXIX  IC-SPETO.
Power Quality: A Nonlinear Adaptive Filter for Improved Power System Operation and Protection Research Overview: Focuses on the application of a new algorithm.
1 An Efficient Classification Approach Based on Grid Code Transformation and Mask-Matching Method Presenter: Yo-Ping Huang.
DETECTION OF HIGH IMPEDANCE FAULTS USING ARTIFICIAL NEURAL NETWORKS
Data Mining and Decision Support
Turning a Mobile Device into a Mouse in the Air
Analysis of Traction System Time-Varying Signals using ESPRIT Subspace Spectrum Estimation Method Z. Leonowicz, T. Lobos
Comparison of filters for burst detection M.-A. Bizouard on behalf of the LAL-Orsay group GWDAW 7 th IIAS-Kyoto 2002/12/19.
IEEE International Conference on Fuzzy Systems p.p , June 2011, Taipei, Taiwan Short-Term Load Forecasting Via Fuzzy Neural Network With Varied.
1 A Statistical Matching Method in Wavelet Domain for Handwritten Character Recognition Presented by Te-Wei Chiang July, 2005.
ADAPTIVE BABY MONITORING SYSTEM Team 56 Michael Qiu, Luis Ramirez, Yueyang Lin ECE 445 Senior Design May 3, 2016.
WELCOME.
CLASSIFICATION OF ECG SIGNAL USING WAVELET ANALYSIS
J.PRAKASH.  The term power quality means different things to different people.  Power quality is the interaction of electronic equipment within the.
EDGE DETECTION USING EVOLUTIONARY ALGORITHMS. INTRODUCTION What is edge detection? Edge detection refers to the process of identifying and locating sharp.
LIGO-G Z The Q Pipeline search for gravitational-wave bursts with LIGO Shourov K. Chatterji for the LIGO Scientific Collaboration APS Meeting.
P.1 Book 4 Section 3.2 Mains electricity and household wiring Electricity: friend or foe? Alternating current and mains circuit Safety designs in domestic.
Introduction to Machine Learning, its potential usage in network area,
Content: Distortion at electronic loads
Automated power Factor Correction and Energy Monitoring System
HARMONIC MITIGATION USING PASSIVE FILTERS
A 2 veto for Continuous Wave Searches
Fundamentals of Harmonics
ARTIFICIAL NEURAL NETWORKS
P. Janik, Z. Leonowicz, T. Lobos, Z. Waclawek
QRS Detection Linda Henriksson 1.
IIS for Image Processing
System Control based Renewable Energy Resources in Smart Grid Consumer
Enhancing User identification during Reading by Applying Content-Based Text Analysis to Eye- Movement Patterns Akram Bayat Amir Hossein Bayat Marc.
Baselining PMU Data to Find Patterns and Anomalies
UNIT-8 INVERTERS 11/27/2018.
Spike Sorting for Extracellular Recordings
Where did we stop? The Bayes decision rule guarantees an optimal classification… … But it requires the knowledge of P(ci|x) (or p(x|ci) and P(ci)) We.
An Improved Neural Network Algorithm for Classifying the Transmission Line Faults Slavko Vasilic Dr Mladen Kezunovic Texas A&M University.
Department of Electrical Engineering
Govt. Polytechnic Dhangar(Fatehabad)
Presenter: Shih-Hsiang(士翔)
Introduction to Artificial Intelligence Lecture 22: Computer Vision II
Presentation transcript:

Nonintrusive Appliance Load Monitoring 组长:辛美惠 组员:赵蓓,姚宁,刘铸

Structure of this paper 2 I.Introduction II.Low-Frequency Hardware Installation Features: Changes of real power and reactive power Features: Changes of real power and reactive power and additional “macroscopic” signatures Features: Changes of real power only III.Higher-Frequency Sampling Hardware Features: Harmonics and Fourier Transform Features: beyond FFT transform IV.Algorithms V.Conclusions

I. Introduction  Building account for a larger portion of U.S energy 3 SectorPercentageRemark Building 40% (with 73% electrical) 2-3 times of similar Chinese buildings Transportation40% Others20%  Building electricity consumption can be reduced by up to 10 to 15% using better energy management  The Smart Grid and home automation networks have the potential to become the main energy management tools

I. Introduction  Two problems impede the deployment: Existing appliances need to be modified for a two-way communication. Acceptance is still low.  In NIALM, the appliance use information obtained at the main breaker level and the energy management is realized manually  NIALM can present a viable alternative to the home automation network  Many electronics companies have initiated aggressive research and development efforts in NIALM. 4

I. Introduction  NIALM have several common principles: ① Specific appliance features, or signatures, need to be selected and mathematically characterized. ② A hardware installation that can detect the selected features. ③ A mathematical algorithm detects the features in the overall signal. 5

II. Low-Frequency Hardware Installation This Hardwareinexpensive Record real power and voltage More advanced systems can record the reactive power Typical sampling rate of 1 Hz 6

II. A. Features: Changes of real power and reactive power  Original NIALM method by MIT (1980s) Four categories of appliances 7 On-off appliances. Finite state machines(FSM). Be detected Permanent consumer devices. Continuously variable consumer devices. Not be detected

II. A. Features: Changes of real power and reactive power  The established version of the MIT method includes five steps. 1.an edge detector identifies changes in the steady-state power draw levels. 2.a cluster analysis algorithm locates these changes in a two- dimensional signature space of real and reactive power. 3.positive and negative clusters of similar magnitude are paired or matched. 4.unmatched clusters and events are paired or associated with existing or new clusters according to a best likelihood algorithm. 5.pairs of clusters are associated with known load power consumption levels to determine the operating schedule of individual loads. 8

II. A. Features: Changes of real power and reactive power  Advantages: can relatively easy detect and track the on-off appliances  Disadvantages: 1.Have problem in detecting multi-state and variable- load appliances; 2.Can not separate the similar power of appliances e.g. a computer and an incandescent bulb; 3.Mismatch of the positive change and the negative change of power. 9

II. B. Features: additional “macroscopic” signatures  1. MIT Method with filtering and transient shapes(1996) Extended the original method by filtering the overall electric signal of an industrial building. A median filter removes meaningless abrupt peaks in the raw signal. It considers shapes of the transient events as an additional feature. 10

II. C. Features: Changes of real power only  A cost-effective NIALM solution can be based solely on the real power data instead.  1) Heuristic End-Use Load Profiler (HELP)(1991)  Disaggregate only relatively large loads  Sparsely sampled (15min) data 11

II. C. Features: Changes of real power only For a given data set scans the whole home-level power draw profile records the occurrence, timing, and magnitude of all large changes determines what spikes correspond to the end-use considered adjusts them according to consistency checks  Not be widely disseminated  range of appliances could be detected is limited 12

II. C. Features: Changes of real power only  2) Concordia University (CU) data disaggregation(1999) real power data sampled at a 16 s period uses the changes in real power along with appliance- specific decision rules achieved an estimated detection accuracy of about 80% it requires excessive training 13

II. C. Features: Changes of real power only  3) Extension of the CU method(2000) it uses changes in real power and signal filtering/smoothing to detect on-off events. The detected appliances are checked for consistency using duration statistics accumulated over the training period  In this way, an accidently missing “on” signal can be reconstructed, or two appliances with similar power consumptions but different durations can be separated. 14

II. C. Features: Changes of real power only  4) Baranski’s method(2004) measurement system is based on the conventional power meter that continuously monitors overall real power with a time resolution of 1 s does not require training it creates a frequency analysis(histogram) based on the historical data, and only frequent power changes are considered further 15

III. Higher-Frequency Sampling Hardware 16 Higher sampling rate In order to reach high accuracy of detection and monitoring of appliances MicroscopicHarmonics Signal waveforms

III. A. Features: Harmonics and Fourier Transform Harmonics as complementary features in addition to the changes of real and reactive power. A set of harmonics, obtained by a Fourier transform, can characterize the signal better than a single harmonic  1) Harmonics for transients (MIT method extended)(2003) incorporation of harmonics only use transient signals for harmonic analysis A vector of the first several coefficients of the short- time FFT of the signal as the spectral envelope 17

III. A. Features: Harmonics and Fourier Transform  Advantages can detect numerous appliances including the variable loads  Disadvantages  excessive training is required for each particular appliance  the performance accuracy has not been characterized for many practical scenarios  Robust is unknown 18

III. A. Features: Harmonics and Fourier Transform 2) Harmonics for both transient and steady signals(2006) monitoring of both transient and steady signals in NIALM a monitoring technique based on continuous calculation of signal harmonics Maybe use a neural network for appliance detection the detection system must be trained for all possible combinations of the appliances being on and off detection accuracy was about 80 to 90 %  the implementation is straightforward  Not practical 19

III. A. Features: Harmonics and Fourier Transform 3) Harmonics for nonlinear devices(2007)  use of harmonics specifically for non-linear devices, such as power-electronics appliances (CFL, lamp dimmer, fan).  The method is based on a short-time FFT of a transient signal  No new development 20

III. A. Features: Harmonics and Fourier Transform 4) Noise FFT as a feature(2007,2010) A new way of NIALM By monitoring electric noise in a socket for transient signals, detect most appliances connected to other sockets The FFT of the noise was used as a feature Required training for each appliance and their combinations utilizes Fourier features of the EMI signals to detect SMPS(switch mode power supplies) Accuracy is up to 93.8 % 21

III. A. Features: Harmonics and Fourier Transform  questions remain  depend on the household electrical wiring  the EMI from neighboring will influence the detection  overlaps between the EMI signatures  Some appliances are not equipped with SMPS.  Estimate of energy consumption is disable 22

III. B. Features: beyond FFT transform  signal sampled at a high rate  Not by harmonics / Fourier transform  Proposed features range from wavelet transform to geometrical shape of the waveform to transient energy. 23

III. B. Features: beyond FFT transform 2) Wavelet transform features(2000) using a wavelet transform instead of the FFT for feature calculation 3) Geometrical properties of I-V curves as features(2004,2007) The main novelty is to use an I-V curve without time reference instead of the waveform which is a function of time I(t). 24

III. B. Features: beyond FFT transform 4) Numerous features combined(2010) Using several different features simultaneously to increase algorithm accuracy include waveform, changes of real and reactive power, harmonics admittance waveform (ratio I/V vs. time) power waveform (product I*V vs. time) eigenvalues of waveform transient waveform (power over half-cycle) The latter four features are fairly new so that they may provide significant improvement in accuracy 25

III. B. Features: beyond FFT transform 6) CMU method(2008,2009)  The features used are Real power reactive power the transient features in terms of the regression coefficients of a non-linear fit to FFT transform  matching algorithms considered are standard pattern- recognition classifiers  accuracy ranges from 67% to 100% 26

IV. Algorithms  the main research effort in NIALM has been focused on signature exploration rather than on algorithm development. 27 seeks a combination of appliances that the resultant aggregate signal is as close to the observed signal as possible numerous appliance presences are matched simultaneously to the detected features over a prolonged period of time utilizes more information; expected to provide better disaggregation performance; changes of appliance state are matched to the detected features one-by-one pattern recognition approach; Bayes classifier; neural networks be more robust

V. Conclutions  No complete NIALM solution suitable for all types of household appliances is available.  No complete set of robust, widely accepted appliance features has been identified.  Using more mutually-independent features improves accuracy, albeit with higher false positive rates.  Using several “orthogonal” disaggregation algorithms may improve accuracy, but optimal fusion needs to be implemented. 28

the End Thank you ! 29