WO2022042070A1 - 一种非侵入式负荷检测方法 - Google Patents

一种非侵入式负荷检测方法 Download PDF

Info

Publication number
WO2022042070A1
WO2022042070A1 PCT/CN2021/105242 CN2021105242W WO2022042070A1 WO 2022042070 A1 WO2022042070 A1 WO 2022042070A1 CN 2021105242 W CN2021105242 W CN 2021105242W WO 2022042070 A1 WO2022042070 A1 WO 2022042070A1
Authority
WO
WIPO (PCT)
Prior art keywords
waveform data
data
sub
electrical equipment
equipment
Prior art date
Application number
PCT/CN2021/105242
Other languages
English (en)
French (fr)
Inventor
徐圣兵
王振友
杜钦涛
李培杰
杜青平
陈玮贤
陈玮霖
Original Assignee
广东工业大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 广东工业大学 filed Critical 广东工业大学
Priority to GB2207650.9A priority Critical patent/GB2606284A/en
Priority to US17/779,187 priority patent/US20230384355A1/en
Publication of WO2022042070A1 publication Critical patent/WO2022042070A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/25—Arrangements for measuring currents or voltages or for indicating presence or sign thereof using digital measurement techniques
    • G01R19/2506—Arrangements for conditioning or analysing measured signals, e.g. for indicating peak values ; Details concerning sampling, digitizing or waveform capturing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
    • G01R19/25—Arrangements for measuring currents or voltages or for indicating presence or sign thereof using digital measurement techniques
    • G01R19/2513—Arrangements for monitoring electric power systems, e.g. power lines or loads; Logging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
    • G06V10/431—Frequency domain transformation; Autocorrelation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Definitions

  • the technical field of the invention relates to the field of non-invasive load detection.
  • it relates to a non-invasive load detection method.
  • Non-intrusive load monitoring (NILM) technology only needs to install a sensor at the user entrance of the power grid, and monitors each or each type of user inside the user by collecting and analyzing the user's power consumption characteristics such as current and voltage.
  • NILM Non-intrusive load monitoring
  • the traditional non-invasive research framework mainly includes data collection, data preprocessing, feature extraction, classifier classification, and classification results.
  • This kind of system framework usually needs to establish a device library in advance, and perform data collection, data preprocessing, feature extraction, model training and other operations on the equipment in the device library to obtain a classifier model that can be used for prediction, and can only identify the components included in the The usage of the device in the device library.
  • the traditional non-intrusive load monitoring system architecture has limitations and cannot cope with complex equipment replacement and changes. For new equipment identified outside the equipment library, how to intercept the new equipment operation data and update it Equipment arsenal, retraining, no good way to deal with it.
  • the purpose of the present invention is to provide a new non-intrusive load detection system, when detecting a device other than the device library, the data of the device can be intercepted and stored in the device library, so as to achieve accurate A complete system that recognizes existing devices in the device library and automatically updates the device library when new devices outside the device library are discovered.
  • a non-invasive load detection method including a device classification prediction sub-process, a new device identification sub-process and a classifier self-training sub-process;
  • the new device identification sub-process includes the following steps:
  • the first step is to perform large dynamic marking on the transient event results to detect whether there is a periodic power transition device
  • the second step is to use the peak filter method to judge whether there is a periodic power transition, so as to mark whether the detected events have periodic changes and separate aperiodic large dynamic events from them;
  • the third step is to correct the event detection results, correct the data of the stable segment, obtain the periodic power transition equipment that may be included, and re-correct the prediction of the electrical equipment;
  • the fourth step is to intercept the waveform data of the stable operation section of the equipment as information input, and calculate the number of stable operation sections of the electrical equipment within the predetermined time according to the recorded time point and the end time point of each transient event, that is, this The number of segments that have not been turned on again for a period of time, and the segment numbers corresponding to the start time and end time are marked and recorded;
  • the fifth step is to perform feature extraction on the waveform data of the stable operation section of each of the electrical equipment
  • the sixth step is to use the feature similarity discrimination index to identify whether the unknown device is a new device
  • the seventh step is to subtract the waveform data of the stable operation section of the electrical equipment where it is judged that the new equipment exists and the waveform data of the previous stable operation section of the electrical equipment to separate the waveform data of the new equipment.
  • start the device classification prediction sub-process, and the device classification sub-process includes the following steps:
  • the first step is to collect the current and voltage data of the electrical equipment and preprocess it.
  • the installed data acquisition terminal collect the high-frequency current and high-frequency voltage data of a variety of electrical equipment within a specific period of time; and pre-process the data. Processing, including removing outliers and interpolation;
  • the second step detects the event behavior, detects the occurrence of the event, and distinguishes the transient event from the steady-state event; when the event detection is classified as a transient event, it enters the first step of the new device identification sub-process ;
  • the event detection zone is a steady-state event
  • enter the third step intercept the preprocessed data, obtain the stable segment waveform data, and perform feature extraction based on the stable segment waveform data to extract the operating state features of the electrical equipment.
  • the fourth step is to call the classifier model for prediction, and according to the operating state feature of the electrical equipment, use the operating state feature of the electrical equipment as the input of the classifier model, and call the classifier model parameters generated by the training to predict;
  • the fifth step is to analyze the classification result of the classifier model to obtain energy consumption information of the electrical equipment;
  • the energy consumption information includes operating state information and energy consumption information;
  • the classifier self-training sub-process includes the following steps:
  • the first step is to add and update the device library: the device library includes the device name, device number, steady-state waveform data and transient waveform data of each device.
  • the program will notify the user to the Send a request for inputting the device name, and finally automatically add the device name, device number, steady-state waveform data and transient waveform data information of the new device to the device library;
  • the second step is to generate comprehensive waveform data: call the device number in the device library, and use the calculation method of number arrangement and combination to generate a variety of arrangement combinations of the device numbers composed of different numbers; according to the obtained arrangement Combining the steady-state waveform data of each device in the device library, superimposing the corresponding device waveform data according to the arrangement and combination of different numbers, to obtain a comprehensive state waveform data superimposed by several sections of different device waveforms;
  • the third step is to perform feature extraction on the current waveform data of the combined operation of multiple electrical equipment to obtain feature set data, then divide the obtained feature set data, and divide the feature set data into training sets and test sets, and then use
  • the machine learning classifier model conducts parameter training and accurately predicts the behavior of electrical equipment
  • the fourth step, evaluation of model results Count the prediction results of each power-consuming device for n consecutive cycles in the time period of the stable wave-shaped data of the device intercepted each time, to judge the condition ratio of the device's opening and closing;
  • the new device identification sub-process Before the seventh step of the new device identification sub-process, it also includes the feature similarity comparison according to the sixth step of the new device identification sub-process, if there is a new device, then enter the classifier self-training sub-process, After completing the model training, perform the fifth step of the device classification and prediction sub-process; if there is no new device, directly enter the fifth step in the device classification and prediction sub-process to analyze which devices are running and when Turn it on and turn it off properly to get the energy consumption information of the electrical equipment.
  • the features in the feature extraction include current effective value, active power and reactive power; specifically:
  • I represents the effective value of the current
  • T represents a cycle
  • i represents the instantaneous current
  • P is the active power
  • U is the line voltage
  • I is the line current
  • the first step in the new device identification sub-process is to perform a large dynamic mark on the transient event result to detect whether there is a periodic power transition device, specifically:
  • the second step in the new device identification sub-process use the peak filter method to judge whether there is a periodic power transition, thereby marking whether the detected event has periodic changes and separating aperiodic large dynamic events therefrom;
  • the peak filter makes the non-maximum point and the maximum value of the sequence obtained by Discrete Fourier Transform (DFT) equal to 0 if the maximum value is less than the peak value ⁇ , and the peak filter effect is obtained,
  • DFT Discrete Fourier Transform
  • the purpose of separating aperiodic large dynamic events is to obtain the events that are actually caused by the opening of the device, and to obtain the stable operation key of the device, specifically:
  • the seventh step in the new device identification sub-process separates the waveform data of the new device, specifically:
  • the data of the waveform of the new device after the DFT change can be described as:
  • x j [n] represents the waveform data of the stable operation of the j-th equipment.
  • a j ⁇ I j,1 ,I j,2 ,...,I j,c ⁇
  • I[n] represents the current waveform data when multiple electrical appliances are operating in combination.
  • the fourth step in the classifier self-training sub-process is specifically:
  • the device status evaluation method is as follows:
  • the classifier is a neural network or a K-means clustering algorithm, a support vector machine or a random forest.
  • a non-invasive load detection system includes a device classification prediction sub-process, a new device identification sub-process and a classifier self-training sub-process;
  • the technology lacks the accuracy of identifying new equipment.
  • the data of the equipment can be intercepted and stored in the equipment library, which can not only achieve the function of accurately identifying the existing equipment in the equipment library, but also The device library can be automatically updated when a new device is outside the device library.
  • a method is proposed to use the peak filter to determine whether there is a periodic transition in the power sequence, and then determine whether there is a periodic power transition device. This method can correctly determine the behavior change of the device, so that the operation of the new device can be intercepted more accurately. data.
  • FIG. 1 is a working flowchart of a non-invasive load detection system provided by an embodiment of the present invention
  • FIG. 2 is a schematic diagram of the composition of a device library of a non-invasive load detection system provided by an embodiment of the present invention
  • FIG. 3 is an effect diagram of separating aperiodic large dynamic time of a non-intrusive load detection system according to an embodiment of the present invention.
  • the embodiment of the present invention provides a non-intrusive load detection system, which is used to automatically update the equipment library when a new equipment other than the equipment library is found, and when intercepting the operation data of the new equipment, a method for determining the power using a peak filter is proposed A method for determining whether there is a periodic transition in the sequence and then judging whether there is a periodic power transition device, so that the operation data of the new device can be intercepted more accurately.
  • FIG. 1 is a working flowchart of a non-intrusive load detection method according to an embodiment of the present invention.
  • a non-intrusive load detection method includes a device classification prediction sub-process, a new device identification sub-process, and a classifier Self-training sub-process;
  • the new device identification sub-process includes the following steps:
  • the first step is to perform large dynamic marking on the transient event results to detect whether there is a periodic power transition device
  • the second step is to use the peak filter method to judge whether there is a periodic power transition, so as to mark whether the detected events have periodic changes and separate aperiodic large dynamic events from them;
  • the third step is to correct the event detection results, correct the data of the stable segment, obtain the periodic power transition equipment that may be included, and re-correct the prediction of the electrical equipment;
  • the fourth step is to intercept the waveform data of the stable operation section of the equipment as information input, and calculate the number of stable operation sections of the electrical equipment within the predetermined time according to the recorded time point and the end time point of each transient event, that is, this The number of segments that have not been turned on again for a period of time, and the segment numbers corresponding to the start time and end time are marked and recorded;
  • the fifth step is to perform feature extraction on the waveform data of the stable operation section of each of the electrical equipment
  • the sixth step is to use the feature similarity discrimination index to identify whether the unknown device is a new device
  • the seventh step is to subtract the waveform data of the stable operation section of the electrical equipment where it is judged that the new equipment exists and the waveform data of the previous stable operation section of the electrical equipment to separate the waveform data of the new equipment.
  • start the device classification prediction sub-process, and the device classification sub-process includes the following steps:
  • the first step is to collect the current and voltage data of the electrical equipment and preprocess it.
  • the installed data acquisition terminal collect the high-frequency current and high-frequency voltage data of a variety of electrical equipment within a specific period of time; and pre-process the data. Processing, including removing outliers and interpolation;
  • the second step detects the event behavior, detects the occurrence of the event, and distinguishes the transient event from the steady-state event; when the event detection is classified as a transient event, it enters the first step of the new device identification sub-process ;
  • the event detection zone is a steady-state event
  • enter the third step intercept the preprocessed data, obtain the stable segment waveform data, and perform feature extraction based on the stable segment waveform data to extract the operating state features of the electrical equipment.
  • the fourth step is to call the classifier model for prediction, and according to the operating state feature of the electrical equipment, use the operating state feature of the electrical equipment as the input of the classifier model, and call the classifier model parameters generated by the training to predict;
  • the fifth step is to analyze the classification result of the classifier model to obtain energy consumption information of the electrical equipment;
  • the energy consumption information includes operating state information and energy consumption information;
  • the classifier self-training sub-process includes the following steps:
  • the first step is to add and update the device library: the device library includes the device name, device number, steady-state waveform data and transient waveform data of each device.
  • the program will notify the user to the Send a request for inputting the device name, and finally automatically add the device name, device number, steady-state waveform data and transient waveform data information of the new device to the device library;
  • the second step is to generate comprehensive waveform data: call the device number in the device library, and use the calculation method of number permutation and combination to generate a variety of permutations and combinations of the device numbers composed of different numbers; according to the obtained arrangement Combining and combining the steady-state waveform data of each device in the device library, superimposing the corresponding device waveform data according to the arrangement and combination of different numbers, to obtain a comprehensive state waveform data superimposed by several sections of different device waveforms;
  • the third step is to perform feature extraction on the current waveform data of the combined operation of multiple electrical equipment to obtain feature set data, then divide the obtained feature set data, and divide the feature set data into training sets and test sets, and then use
  • the machine learning classifier model conducts parameter training and accurately predicts the behavior of electrical equipment
  • the fourth step, evaluation of model results Count the prediction results of each power-consuming device for n consecutive cycles in the time period of the stable wave-shaped data of the device intercepted each time, to judge the condition ratio of the device's opening and closing;
  • the new device identification sub-process Before the seventh step of the new device identification sub-process, it also includes the feature similarity comparison according to the sixth step of the new device identification sub-process, if there is a new device, then enter the classifier self-training sub-process, After completing the model training, perform the fifth step of the device classification and prediction sub-process; if there is no new device, directly enter the fifth step in the device classification and prediction sub-process to analyze which devices are running and when Turn it on and turn it off properly to get the energy consumption information of the electrical equipment.
  • processing is performed according to specific steps, and the device classification prediction sub-flow includes the following steps:
  • the first step is to collect current and voltage data and perform preprocessing.
  • the data acquisition terminal installed at the entrance of the home, collect the high-frequency current and high-frequency voltage data of various devices within Time; and preprocess the data, Including removing outliers and interpolation processing, the data table is as follows:
  • the second step detects event behavior, detects the occurrence of events, and distinguishes transient events from steady-state events;
  • the third step is to intercept the preprocessed data to obtain stable segment waveform data, and based on the stable segment waveform data, extract the operating state characteristics of the electrical equipment;
  • the features in the feature extraction include current RMS, active power and reactive power; specifically: Feature 1: Current RMS
  • I represents the rms value of the current
  • T represents a cycle
  • i represents the instantaneous current
  • P is the active power
  • U is the line voltage
  • I is the line current
  • the fourth step is to call the classifier model for prediction, extract the operating state features of the electrical equipment according to the third step, use the operating state features of the electrical equipment as the input of the classifier model, and call the training generated data.
  • the classifier model parameters are predicted;
  • the fifth step is to analyze the classification results of the classifier, analyze which equipment is running, when to open it and close it properly, and obtain the operation status and energy consumption information of the electric equipment;
  • the new device identification sub-process includes the following steps:
  • the first step is to perform large dynamic marking on the transient and steady-state event results input by the device prediction sub-process to detect whether there is a periodic power transition device;
  • the detection of event behavior is carried out, the occurrence of the event is detected, and if it is classified as a steady-state event, then it enters the third step of the device classification prediction sub-process; If it is classified as a transient event, then enter the first large dynamic mark in the new device identification sub-process;
  • Periodic power transition equipment means that its power will undergo periodic sudden changes during the operation of the equipment. This kind of equipment is called periodic power transition equipment, specifically:
  • the second step is to use the peak filter method to judge whether there is a periodic power transition, so as to mark whether the detected events have periodic changes and separate aperiodic large dynamic events from them;
  • the peak filter makes the non-maximum point of the sequence obtained by the Discrete Fourier Transform (DFT) and the value of the maximum value less than the peak value ⁇ equal to 0, so as to obtain the peak filter effect.
  • DFT Discrete Fourier Transform
  • the purpose of separating aperiodic large dynamic events is to obtain the events that are really caused by the opening of the device, and to obtain the steady state key of the device operation. specific:
  • the purpose of separating aperiodic large dynamic events is to obtain the events that are really caused by the opening of the device, and to obtain the steady state key of the device operation. Specifically:
  • 0.15 is a threshold in the program.
  • the third step is to correct the event detection results, correct the data of the stable segment, and obtain the periodic power transition equipment that may be contained therefrom, and re-correct the prediction of the equipment;
  • the periodic power transition equipment may not be identified, resulting in a big difference between the combination predicted by the model and the actual electrical combination, and finally the trained model will be invalid.
  • the data at this time is that all devices are in a normal operation state. In this case, the interference of the current waveform change caused by the periodic power transition device not being powered on for a period of time is excluded, and the prediction is enhanced. accuracy.
  • the data is intercepted for each stable segment, and the steady state waveform of each device is obtained.
  • the fourth step is to intercept the waveform data of the stable operation of the equipment.
  • calculate the number of the stable operation period of the electrical equipment in a period of time that is, this
  • the number of segments that have not been restarted for a period of time, and the start time and end time are marked with the corresponding segment number and recorded; by intercepting the nodes where each device is turned on and ended, multiple segments containing the corresponding start time and end time are obtained. And the device running stable segment waveform data of segment number.
  • the fifth step is to perform feature extraction on the waveform data of the stable operation section of each equipment; perform feature extraction on the waveform data of the stable operation section of each equipment, and the feature calculation method is equivalent to the third step in the equipment classification prediction process. extract.
  • the various characteristics of the stable section of different equipment such as current RMS, active power, reactive power, etc. are obtained by calculation.
  • the sixth step is to use the feature similarity discrimination index to identify whether the unknown device is a new device
  • the seventh step is to subtract the waveform data of the stable operation segment of the equipment where it is judged that there is a new device from the waveform data of the previous stable operation segment of the equipment, so as to separate the waveform data of the new device; the details are as follows:
  • the data of the waveform of the new device after the DFT change can be described as:
  • x j [n] represents the waveform data of the stable operation of the j-th equipment.
  • the classifier self-training sub-process includes the following steps:
  • Step 1 Add and update the device library: In the device library, the device name, device number, steady-state waveform data and transient waveform data of each device are included in the device library. For the identified new device, the program will send a message to the user. Enter the request for the device name, and finally the device name, device number, steady-state waveform data and transient waveform data of the new device are automatically added to the device library, as shown in Figure 2.
  • the second step is to generate comprehensive waveform data: call the device number in the device library, and use the calculation method of number arrangement and combination to generate a variety of arrangement combinations of the device numbers composed of different numbers; according to the obtained arrangement Combining the steady-state waveform data of each device in the device library, superimposing the corresponding device waveform data according to the arrangement and combination of different numbers, to obtain a comprehensive state waveform data superimposed by several sections of different device waveforms;
  • a j ⁇ I j,1 ,I j,2 ,...,I j,c ⁇
  • I[n] represents the current waveform data when multiple electrical appliances are operating in combination.
  • the third step is to perform feature extraction on the current waveform data when multiple electrical appliance combinations are running to obtain feature set data, then divide the obtained feature set data into training sets and test sets, and then use machine learning classifiers
  • the model performs parameter training for accurate prediction of equipment behavior
  • the fourth step, evaluation of model results Count the prediction results of each power-consuming device for N consecutive cycles in the time period of the stable wave-shaped data of the device intercepted each time, to determine the ratio of the conditions for turning on and off the device;
  • the new device identification sub-process Before the seventh step of the new device identification sub-process, it also includes the feature similarity comparison according to the sixth step of the new device identification sub-process, if there is a new device, then enter the classifier self-training sub-process, After completing the model training, perform the fifth step of the device classification and prediction sub-process; if there is no new device, directly enter the fifth step in the device classification and prediction sub-process to analyze which devices are running and when Turn it on properly and turn it off to get the energy consumption information such as the running status and energy consumption of the electrical equipment.
  • a non-intrusive load detection system includes a device classification prediction sub-process, a new device identification sub-process, and a classifier self-training sub-process; the setting of these methods and processes overcomes the prior art
  • the data of the equipment can be intercepted and stored in the equipment library.
  • the equipment library can be automatically updated when the new equipment is outside the library; when the operation data of the new equipment is intercepted, a method is proposed to use the peak filter to determine whether there is a periodic transition in the power sequence, and then determine whether there is a periodic power transition device. The method can correctly judge the behavior change of the device, so that the operation data of the new device can be intercepted more accurately.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Computing Systems (AREA)
  • Multimedia (AREA)
  • Mathematical Physics (AREA)
  • Power Engineering (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种非侵入式负荷检测系统，包括设备分类预测子流程、新设备识别子流程和分类器自我训练子流程；通过这些方法流程的设置克服了现有技术对新设备识别准确度的不足，当检测出设备库以外的设备时，能够截取该设备的数据并存入到设备库，不仅可以实现准确识别出设备库已有设备的功能，而且当发现设备库之外的新设备时能够自动更新设备库。

Description

一种非侵入式负荷检测方法 技术领域
本发明的技术领域涉及非侵入式负荷检测领域。尤其涉及一种非侵入式负荷检测方法。
背景技术
目前国内电力行业所研究与使用的多数为侵入式负载监测系统，但侵入式负载监测系统存在应用成本高、部署难度大、适应系统变化能力弱的问题。而非侵入式负荷检测系统(non-intrusive load monitoring,NILM)技术仅需在电网的用户入口处安装一个传感器，通过采集和分析电流、电压等用户用电特征来监测用户内部每个或每类用电设备的工作状态，得到广泛的研究。
传统的非侵入式研究的框架主要包括数据采集、数据预处理、特征提取、分类器分类、得到分类结果。这种系统框架通常需要事先建立设备库，对设备库中的设备进行数据采集、数据预处理、特征提取、模型训练等操作，得到能够用于预测的分类器模型，且只能识别出包括在设备库里的设备的使用情况。但在实际应用中，传统的非侵入式负荷监测系统架构存在局限性，无法应对复杂的设备更换、变动等情况，对于识别到设备库之外的新设备，如何截取新设备运行数据，并更新设备库、重新训练，没有很好的应对方法。
发明内容
针对现有技术的缺陷，本发明的目的是提供一种新的非侵入式负荷检测系统，当检测出设备库以外的设备时，能够截取该设备的数据并存入到设备库，实现能够准确识别出设备库已有设备，以及发现设备库之外的新设备时能够自动更新设备库的一整套完整的系统。
为达此目的，本发明采用以下技术方案：一种非侵入式负荷检测方法，包括设备分类预测子流程、新设备识别子流程和分类器自我训练子流程；
所述新设备识别子流程包括如下步骤：
第一步、对暂态事件结果进行大动态标记，用以检测是否存在周期性功率跃迁设备；
第二步、利用峰值滤波器方法判断是否存在周期性功率跃迁，从而标记探测到的事件是否有周期性变化并从中分离非周期性大动态事件；
第三步、修正事件探测结果，修正稳定段的数据，从中得到可能含有的周期性功率跃迁设备，并重新对用电设备的预测进行修正；
第四步、截取设备运行稳定段波形数据，作为信息输入，根据记录的每一个暂态事件发生的时间点和结束的时间点，计算出预定时间内用电设备运行平稳段的数量，即这段时间没有用电设备再开启的段数，以及标出起始时间及终止时间相应的段号并记录；
第五步、对各所述用电设备运行稳定段波形数据，进行特征提取；
第六步、利用特征相似度判别指标来识别未知设备是否为新设备；
第七步、将判断存在新设备的用电设备运行稳定段波形数据与前一段用电设备运行稳定段波形数据做减，从而分离出新设备的波形数据。
进一步的，在所述新设备识别子流程的所述第一步运作之前，启动所述设备分类预测子流程，所述设备分类子流程包括如下步骤：
第一步、采集用电设备的电流电压数据并进行预处理，通过安装好的数据采集终端，采集特定时间长度内多种用电设备的高频电流、高频电压数据；并对数据进行预处理，包括剔除异常值和插值处理；
第二步、事件探测，进行事件行为的检测，检测事件的发生，区分暂态事件和稳态事件；当事件探测区分为暂态事件时，则进入所述新设备识别子流程的第一步；
当事件探测区分为稳态事件时，则进入第三步、对预处理过的数据进行截取，获得稳定段波形数据，基于所述稳定段波形数据，进行特征提取，提取用电设备运行状态特征；
第四步、调用分类器模型进行预测，根据所述用电设备运行状态特征，以所述用电设备运行状态特征作为分类器模型输入，调用训练生成的分类器模型参数进行预测；
第五步、对分类器模型的分类结果进行分析，得到用电设备的用能信息；所述用能信息包括运行状态信息和能耗信息；
在所述新设备识别子流程的第七步分离出新设备的波形数据之后，进入到 所述分类器自我训练子流程；
所述分类器自我训练子流程包括如下步骤：
第一步、加入并更新设备库：在所述设备库中，包括每个设备的设备名称、设备编号、以及稳态波形数据与暂态波形数据，对于识别到的新设备，程序会向用户发出输入设备名称的请求，最终将新设备的设备名称、设备编号、稳态波形数据与暂态波形数据信息自动加入到设备库中；
第二步、生成综合态波形数据：调用所述设备库中的所述设备编号，使用编号排列组合的计算方法，生成由不同编号组成的多种所述设备编号的排列组合；根据得到的排列组合和所述设备库中各设备的稳态波形数据，依据不同编号的排列组合，将相对应的设备波形数据进行叠加，得到若干段不同设备波形叠加的综合态波形数据；
第三步、对多个用电设备组合运行时的电流波形数据进行特征提取得到特征集数据，然后对得到的特征集数据进行划分，将特征集数据划分为训练集和测试集，之后，使用机器学习分类器模型进行参数训练，并进行用电设备行为的准确预测；
第四步、模型结果评价：统计每次截取到的设备稳定波段形数据时间段内，连续n个周期，每个周期各个用电设备的预测结果，来判断设备的开启、关闭的条件比例；
所述新设备识别子流程的第七步之前，还包括根据所述新设备识别子流程的第六步进行特征相似度对比，如果存在新设备，则进入到所述分类器自我训练子流程，完成模型训练后再执行所述设备分类预测子流程的第五步；如果不存在新设备，则直接进入到所述设备分类预测子流程中的第五步，分析出哪些设备正在运行，何时开启合适关闭，得到用电设备的用能信息。
进一步的，所述特征提取中的特征包括电流有效值、有功功率和无功功率；具体为：
特征1：电流有效值
计算用电设备运行状态下的电流有效值，具体地：
Figure PCTCN2021105242-appb-000001
其中，I表示电流有效值，T表示一个周期，i表示瞬时电流；
特征2：有功功率
计算用电设备运行状态下的有功功率，具体地：
Figure PCTCN2021105242-appb-000002
其中，P表示有功功率，U为线电压，I为线电流，
Figure PCTCN2021105242-appb-000003
为U和I之间的相位差；
特征3：无功功率
计算用电设备运行状态下的无功功率，具体地：
Figure PCTCN2021105242-appb-000004
进一步的，所述新设备识别子流程中的第一步、对暂态事件结果进行大动态标记，用以检测是否存在周期性功率跃迁设备，具体为：
针对上述功率序列(P 1，P 2，…，P N)，用1标记探测到事件，用0表示没有事件发生，得到c点序列x[n]，n∈{0，1，2，3...，N-1}，表示为：
Figure PCTCN2021105242-appb-000005
进一步的，所述新设备识别子流程中的第二步：利用峰值滤波器方法判断是否存在周期性功率跃迁，从而标记探测到的事件是否有周期性变化并从中分离非周期性大动态事件；其中的峰值滤波器令对离散傅里叶变换(Discrete Fourier Transform，DFT)得到的序列的非极大值点和极大值小于峰值θ的值等于0，得到峰值滤波效果，
具体的：
对于大动态标记序列x[n]，进行离散傅里叶变换为：
Figure PCTCN2021105242-appb-000006
对DFT得到的序列进行峰值滤波，滤波公式表示为：
Figure PCTCN2021105242-appb-000007
其中
Figure PCTCN2021105242-appb-000008
Figure PCTCN2021105242-appb-000009
这里的截止峰值θ，在程序里采用
Figure PCTCN2021105242-appb-000010
分离非周期性大动态事件的目的是得到真正是设备开启引起的事件，获得设备运行稳态取键，具体为：
对
Figure PCTCN2021105242-appb-000011
进行IDFT为：
Figure PCTCN2021105242-appb-000012
分离后的序列表示为：
Figure PCTCN2021105242-appb-000013
非周期性大动态事件满足条件：
η[n]＞η[n] max。
进一步的，所述新设备识别子流程中的第七步分离出新设备的波形数据，具体为：
新设备的波形经过DFT变化后的数据可描述为：
Figure PCTCN2021105242-appb-000014
对X[k]进行IDFT可得到新设备波形为：
Figure PCTCN2021105242-appb-000015
式中，x j[n]表示第j段设备运行稳定段的波形数据。
进一步的，所述分类器自我训练子流程的第二步，
具体的：
设已知家用电器数据库有r种家用电器，每种电器每个周期的采样点数为c，则对于电器A j：
A j＝{I j，1，I j，2，…，I j，c}
当采集到所有电器单独运行时的电流数据时，我们可以通过以下操作获得多个电器组合运行的电流数据，
对于电器波形数据I j，经过DTFT分解后信号波形的模型可描述为：
Figure PCTCN2021105242-appb-000016
多个电器同时运行时，其信号波形经过DTFT分解后信号波形的模型可描述为：
Figure PCTCN2021105242-appb-000017
其中s表示同时组合运行的电器总数
对X[k]进行IDFT为：
Figure PCTCN2021105242-appb-000018
其中I[n]表示多个电器组合运行时的电流波形数据。
进一步的，所述分类器自我训练子流程中的第四步，具体为：
统计每次截取到的设备稳定段波形数据时间段内，连续n个周期，每个周期各个设备的预测结果，设备状态评价方式如下：
对于某一次识别结果，我们对各个设备的行为标记如下：
Figure PCTCN2021105242-appb-000019
统计各个设备n个连续周期的预测结果，则
Figure PCTCN2021105242-appb-000020
故在这段时间内，各个设备的真实状态为：
Figure PCTCN2021105242-appb-000021
其实Status等于1时表示这段时间内设备开启，等于None时表示这段时间内设备的状态未知，等于0表示这段时间内设备关闭，p1表示判定设备开启的条件比例，p2表示判定设备一定未开启的预测结果比例。
进一步的，所述分类器是神经网络或者K均值聚类算法或者支持向量机或者随机森林。
本发明的有益效果：本发明实施例的一种非侵入式负荷检测系统，包括设备分类预测子流程、新设备识别子流程和分类器自我训练子流程；通过这些方法流程的设置克服了现有技术对新设备识别准确度的不足，当检测出设备库以外的设备时，能够截取该设备的数据并存入到设备库，不仅可以实现准确识别出设备库已有设备的功能，而且当发现设备库之外的新设备时能够自动更新设备库。
具体的具备以下有益效果：
1)能应对复杂的实际应用情况，能够识别出周期性功率跃迁设备；
2)利用大动态标记，提高设备行为变化判断的准确度，能够正确判断设备的行为变化，从而可以更准确得截取新设备运行数据；
3)利用分离出设备稳定运行的数据，对其进行特征提取和相似度比较，将新设备的数据加入到设备库并对新的设备库进行重新训练；
4)提出一种利用峰值滤波器判断功率序列是否存在周期性跃变、进而判断是否存在周期性功率跃迁设备的方法，该方法能够正确判断设备的行为变 化，从而可以更准确地截取新设备运行数据。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其它的附图。
图1为本发明实施例提供的一种非侵入式负荷检测系统的工作流程图；
图2为本发明实施例提供的一种非侵入式负荷检测系统设备库组成示意图；
图3为本发明实施例提供的一种非侵入式负荷检测系统的分离非周期性大动态时间的效果图。
具体实施方式
本发明实施例提供了一种非侵入式负荷检测系统，用于当发现设备库之外的新设备时能够自动更新设备库，在截取新设备运行数据时，提出一种利用峰值滤波器判断功率序列是否存在周期性跃变、进而判断是否存在周期性功率跃迁设备的方法，从而可以更准确地截取新设备运行数据。
为使得本发明的发明目的、特征、优点能够更加的明显和易懂，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，下面所描述的实施例仅是本发明一部分实施例，而非全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
下面结合附图并通过具体实施方式来进一步说明本发明的技术方案。
请参考图1，图1为本发明实施例的一种非侵入式负荷检测方法的工作流程图，一种非侵入式负荷检测方法，包括设备分类预测子流程、新设备识别子流程和分类器自我训练子流程；
所述新设备识别子流程包括如下步骤：
第一步、对暂态事件结果进行大动态标记，用以检测是否存在周期性功率跃迁设备；
第二步、利用峰值滤波器方法判断是否存在周期性功率跃迁，从而标记探测到的事件是否有周期性变化并从中分离非周期性大动态事件；
第三步、修正事件探测结果，修正稳定段的数据，从中得到可能含有的周期性功率跃迁设备，并重新对用电设备的预测进行修正；
第四步、截取设备运行稳定段波形数据，作为信息输入，根据记录的每一个暂态事件发生的时间点和结束的时间点，计算出预定时间内用电设备运行平稳段的数量，即这段时间没有用电设备再开启的段数，以及标出起始时间及终止时间相应的段号并记录；
第五步、对各所述用电设备运行稳定段波形数据，进行特征提取；
第六步、利用特征相似度判别指标来识别未知设备是否为新设备；
第七步、将判断存在新设备的用电设备运行稳定段波形数据与前一段用电设备运行稳定段波形数据做减，从而分离出新设备的波形数据。
进一步的，在所述新设备识别子流程的所述第一步运作之前，启动所述设备分类预测子流程，所述设备分类子流程包括如下步骤：
第一步、采集用电设备的电流电压数据并进行预处理，通过安装好的数据采集终端，采集特定时间长度内多种用电设备的高频电流、高频电压数据；并对数据进行预处理，包括剔除异常值和插值处理；
第二步、事件探测，进行事件行为的检测，检测事件的发生，区分暂态事件和稳态事件；当事件探测区分为暂态事件时，则进入所述新设备识别子流程的第一步；
当事件探测区分为稳态事件时，则进入第三步、对预处理过的数据进行截取，获得稳定段波形数据，基于所述稳定段波形数据，进行特征提取，提取用电设备运行状态特征；
第四步、调用分类器模型进行预测，根据所述用电设备运行状态特征，以所述用电设备运行状态特征作为分类器模型输入，调用训练生成的分类器模型参数进行预测；
第五步、对分类器模型的分类结果进行分析，得到用电设备的用能信息；所述用能信息包括运行状态信息和能耗信息；
在所述新设备识别子流程的第七步分离出新设备的波形数据之后，进入到 所述分类器自我训练子流程；
所述分类器自我训练子流程包括如下步骤：
第一步、加入并更新设备库：在所述设备库中，包括每个设备的设备名称、设备编号、以及稳态波形数据与暂态波形数据，对于识别到的新设备，程序会向用户发出输入设备名称的请求，最终将新设备的设备名称、设备编号、稳态波形数据与暂态波形数据信息自动加入到设备库中；
第二步、生成综合态波形数据：调用所述设备库中的所述设备编号，使用编号排列组合的计算方法，生成由不同编号组成的多种所述设备编号的排列组合；根据得到的排列组合和所述设备库中各设备的稳态波形数据，依据不同编号的排列组合，将相对应的设备波形数据进行叠加，得到若干段不同设备波形叠加的综合态波形数据；
第三步、对多个用电设备组合运行时的电流波形数据进行特征提取得到特征集数据，然后对得到的特征集数据进行划分，将特征集数据划分为训练集和测试集，之后，使用机器学习分类器模型进行参数训练，并进行用电设备行为的准确预测；
第四步、模型结果评价：统计每次截取到的设备稳定波段形数据时间段内，连续n个周期，每个周期各个用电设备的预测结果，来判断设备的开启、关闭的条件比例；
所述新设备识别子流程的第七步之前，还包括根据所述新设备识别子流程的第六步进行特征相似度对比，如果存在新设备，则进入到所述分类器自我训练子流程，完成模型训练后再执行所述设备分类预测子流程的第五步；如果不存在新设备，则直接进入到所述设备分类预测子流程中的第五步，分析出哪些设备正在运行，何时开启合适关闭，得到用电设备的用能信息。
下面介绍本发明的一个具体应用场景：
如图1所示，按照具体步骤进行处理，所述设备分类预测子流程包括如下步骤：
第一步、采集电流电压数据并进行预处理，通过安装在家庭入口处的数据采集终端，采集时间长度为Time内多种设备的高频电流、高频电压数据；并对数据进行预处理，包括剔除异常值和插值处理，数据表如下：
Figure PCTCN2021105242-appb-000022
第二步、事件探测，进行事件行为的检测，检测事件的发生，区分暂态事件和稳态事件；
第三步、对预处理过的数据进行截取，获得稳定段波形数据，基于所述稳定段波形数据，提取用电设备运行状态特征；
所述特征提取中的特征包括电流有效值、有功功率和无功功率；具体为：特征1：电流有效值
计算用电设备运行状态下的电流有效值，具体地：
Figure PCTCN2021105242-appb-000023
其中，I表示电流有效值，T表示一个周期，i表示瞬时电流。
特征2：有功功率
计算用电设备运行状态下的有功功率，具体地：
Figure PCTCN2021105242-appb-000024
其中，P表示有功功率，U为线电压，I为线电流，
Figure PCTCN2021105242-appb-000025
为U和I之间的相位差；
特征3：无功功率
计算用电设备运行状态下的无功功率，具体地：
Figure PCTCN2021105242-appb-000026
第四步、调用分类器模型进行预测，根据所述第三步中的所述提取所述用电设备运行状态特征，以所述用电设备运行状态特征作为分类器模型输入，调用训练生成的分类器模型参数进行预测；
第五步、对分类器的分类结果进行分析，分析出哪些设备正在运行，何时开启合适关闭，得到用电设备的运行状态、能耗用能信息；
所述新设备识别子流程包括如下步骤：
第一步、对设备预测子流程输入的暂态和稳态事件结果进行大动态标记，用以检测是否存在周期性功率跃迁设备；
在所述设备分类预测子流程的第二步中，进行事件行为的检测，检测事件的发生，如果区分为稳态事件，则进入到所述设备分类预测子流程的所述第三步中；如果区分为暂态事件，则进入到所述新设备识别子流程中的第一步大动态标记；
周期性功率跃迁设备是指在设备运行过程中，其功率会会进行周期性的突变，这种设备称为周期性功率跃迁设备，具体为：
针对上述功率序列(P 1，P 2，…，P N)，用1标记探测到事件，用0表示没有事件发生，得到c点序列x[n]，n∈{0，1，2，3...，N-1}，表示为：
Figure PCTCN2021105242-appb-000027
第二步、利用峰值滤波器方法判断是否存在周期性功率跃迁，从而标记探测到的事件是否有周期性变化并从中分离非周期性大动态事件；
峰值滤波器令对离散傅里叶变换(Discrete Fourier Transform，DFT)得到的序列的非极大值点和极大值小于峰值θ的值等于0，得到峰值滤波效果。分离非周期性大动态事件的目的是得到真正是设备开启引起的事件，获得设备运行稳态取键。具体的：
对于大动态标记序列x[n]，进行离散傅里叶变换为：
Figure PCTCN2021105242-appb-000028
对DFT得到的序列进行峰值滤波，滤波公式表示为：
Figure PCTCN2021105242-appb-000029
其中
Figure PCTCN2021105242-appb-000030
Figure PCTCN2021105242-appb-000031
这里的截止峰值θ，在程序里采用
Figure PCTCN2021105242-appb-000032
分离非周期性大动态事件的目的是得到真正是设备开启引起的事件，获得设备运行稳态取键。具体为：
对
Figure PCTCN2021105242-appb-000033
进行IDFT为：
Figure PCTCN2021105242-appb-000034
分离后的序列表示为：
Figure PCTCN2021105242-appb-000035
非周期性大动态事件满足条件：
η[n]＞η[n] max
注：0.15为程序里一个阈值。
以有功功率作为研究对象，分离非周期性大动态时间的效果如图3：
Figure PCTCN2021105242-appb-000036
在识别效果子图中可以发现有几个明显的突出点，这些就是分离出来的非周期性大动态事件。然后将这些非周期性大动态事件结果与原先的暂态识别结果进行比较，就能够得出设备是否为周期性设备。若与暂态识别结果相吻合，则说明此非周期性大动态事件为现有设备的暂态过程；否则，我们可以认为该设备是一个非周期性设备。
第三步、修正事件探测结果，修正稳定段的数据，从中得到可能含有的周期性功率跃迁设备，并重新对设备的预测进行修正；
若不进行修正，则可能会导致周期性功率跃迁设备无法被识别出，造成模型预测的组合与实际的电器组合出现较大差别，最终造成训练出来的模型无效。在修正稳定段数据之后，此时的数据为所有设备都处于正常运行的状态，在这种情况下，排除了周期性功率跃迁设备在一段时间内没有通电导致电流波形变化的干扰，增强了预测的准确性。对每一个稳定段进行数据截取，得到的是每一个设备的稳定态波形。
第四步、截取设备运行稳定段波形数据，根据记录的每一个暂态事件发生的时间点和结束的时间点，作为信息输入，计算出一段时间内用电设备运行 平稳段的数量，即这段时间没有设备再开启的段数，以及标出起始时间及终止时间标出相应的段号并记录；通过截取每一个设备开启和结束的节点，得到多段包含相对应的起始时间、终止时间及段号的设备运行稳定段波形数据。
第五步、对所述的各个设备运行稳定段波形数据，进行特征提取；对所述的各个设备运行稳定段波形数据，进行特征提取，特征计算方法等同于设备分类预测流程中第三步特征提取。计算得到电流有效值、有功功率、无功功率等不同设备稳定段的各个特征。
第六步、利用特征相似度判别指标来识别未知设备是否为新设备；
第七步、将判断存在新设备的设备运行稳定段波形数据与前一段设备运行稳定段波形数据做减，从而分离出新设备的波形数据；具体为：
新设备的波形经过DFT变化后的数据可描述为：
Figure PCTCN2021105242-appb-000037
对X[k]进行IDFT可得到新设备波形为：
Figure PCTCN2021105242-appb-000038
式中，x j[n]表示第j段设备运行稳定段的波形数据。
所述分类器自我训练子流程包括如下步骤：
第一步、加入并更新设备库：在所述设备库中，包括每个设备的设备名称，设备编号，稳态波形数据与暂态波形数据，对于识别到的新设备，程序会向用户发出输入设备名称的请求，最终将新设备的设备名称，设备编号，稳态波形数据与暂态波形数据等信息自动加入到设备库中，如图2所示，
Figure PCTCN2021105242-appb-000039
第二步、生成综合态波形数据：调用所述设备库中的所述设备编号，使用编号排列组合的计算方法，生成由不同编号组成的多种所述设备编号的排列组合；根据得到的排列组合和所述设备库中各设备的稳态波形数据，依据不同编号的排列组合，将相对应的设备波形数据进行叠加，得到若干段不同设备波形叠加的综合态波形数据；
具体的：
设已知家用电器数据库有r种家用电器，每种电器每个周期的采样点数为c，则对于电器A j：
A j＝{I j，1，I j，2，…，I j，c}
当采集到所有电器单独运行时的电流数据时，我们可以通过以下操作获得多个电器组合运行的电流数据。
1.对于电器波形数据I j，经过DTFT分解后信号波形的模型可描述为：
Figure PCTCN2021105242-appb-000040
2.多个电器同时运行时，其信号波形经过DTFT分解后信号波形的模型可描述为：
Figure PCTCN2021105242-appb-000041
其中s表示同时组合运行的电器总数
3.对X[k]进行IDFT为：
Figure PCTCN2021105242-appb-000042
其中I[n]表示多个电器组合运行时的电流波形数据。
第三步、对多个电器组合运行时的电流波形数据进行特征提取得到特征集数据，然后对得到的特征集数据进行划分，将数据划分为训练集和测试集，之后，使用机器学习分类器模型进行参数训练进行设备行为的准确预测；
第四步、模型结果评价：统计每次截取到的设备稳定波段形数据时间段内，连续N个周期，每个周期各个用电设备的预测结果，来判断设备的开启、关闭的条件比例；
对于某一次识别结果，我们对各个设备的行为标记如下：
Figure PCTCN2021105242-appb-000043
统计各个设备n个连续周期的预测结果，则
Figure PCTCN2021105242-appb-000044
故在这段时间内，各个设备的真实状态为：
Figure PCTCN2021105242-appb-000045
其实Status等于1时表示这段时间内设备开启，等于None时表示这段时间内设备的状态未知，等于0表示这段时间内设备关闭，p1表示判定设备开启的条件比例，p2表示判定设备一定未开启的预测结果比例。
所述新设备识别子流程的第七步之前，还包括根据所述新设备识别子流程的第六步进行特征相似度对比，如果存在新设备，则进入到所述分类器自我训练子流程，完成模型训练后再执行所述设备分类预测子流程的第五步；如果不存在新设备，则直接进入到所述设备分类预测子流程中的第五步，分析出哪些设备正在运行，何时开启合适关闭，得到用电设备的运行状态、能耗 等用能信息。
综上所述，本发明实施例的一种非侵入式负荷检测系统，包括设备分类预测子流程、新设备识别子流程和分类器自我训练子流程；通过这些方法流程的设置克服了现有技术对新设备识别准确度的不足，当检测出设备库以外的设备时，能够截取该设备的数据并存入到设备库，不仅可以实现准确识别出设备库已有设备的功能，而且当发现设备库之外的新设备时能够自动更新设备库；在截取新设备运行数据时，提出一种利用峰值滤波器判断功率序列是否存在周期性跃变、进而判断是否存在周期性功率跃迁设备的方法，该方法能够正确判断设备的行为变化，从而可以更准确地截取新设备运行数据。
以上所述，以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (9)

  1. 一种非侵入式负荷检测方法，其特征在于，包括设备分类预测子流程、新设备识别子流程和分类器自我训练子流程；
    所述新设备识别子流程包括如下步骤：
    第一步、对暂态事件结果进行大动态标记，用以检测是否存在周期性功率跃迁设备；
    第二步、利用峰值滤波器方法判断是否存在周期性功率跃迁，从而标记探测到的事件是否有周期性变化并从中分离非周期性大动态事件；
    第三步、修正事件探测结果，修正稳定段的数据，从中得到可能含有的周期性功率跃迁设备，并重新对用电设备的预测进行修正；
    第四步、截取设备运行稳定段波形数据，作为信息输入，根据记录的每一个暂态事件发生的时间点和结束的时间点，计算出预定时间内用电设备运行平稳段的数量，即这段时间没有用电设备再开启的段数，以及标出起始时间及终止时间相应的段号并记录；
    第五步、对各所述用电设备运行稳定段波形数据，进行特征提取；
    第六步、利用特征相似度判别指标来识别未知设备是否为新设备；
    第七步、将判断存在新设备的用电设备运行稳定段波形数据与前一段用电设备运行稳定段波形数据做减，从而分离出新设备的波形数据。
  2. 根据权利要求1所述的非侵入式负荷检测方法，其特征在于，在所述新设备识别子流程的所述第一步运作之前，启动所述设备分类预测子流程，所述设备分类子流程包括如下步骤：
    第一步、采集用电设备的电流电压数据并进行预处理，通过安装好的数据采集终端，采集特定时间长度内多种用电设备的高频电流、高频电压数据；并对数据进行预处理，包括剔除异常值和插值处理；
    第二步、事件探测，进行事件行为的检测，检测事件的发生，区分暂态事件和稳态事件；当事件探测区分为暂态事件时，则进入所述新设备识别子流程的第一步；
    当事件探测区分为稳态事件时，则进入第三步、对预处理过的数据进行截取，获得稳定段波形数据，基于所述稳定段波形数据，进行特征提取，提取用电设备运行状态特征；
    第四步、调用分类器模型进行预测，根据所述用电设备运行状态特征，以所述用电设备运行状态特征作为分类器模型输入，调用训练生成的分类器模型参数进行预测；
    第五步、对分类器模型的分类结果进行分析，得到用电设备的用能信息；所述用能信息包括运行状态信息和能耗信息；
    在所述新设备识别子流程的第七步分离出新设备的波形数据之后，进入到所述分类器自我训练子流程；
    所述分类器自我训练子流程包括如下步骤：
    第一步、加入并更新设备库：在所述设备库中，包括每个设备的设备名称、设备编号、以及稳态波形数据与暂态波形数据，对于识别到的新设备，程序会向用户发出输入设备名称的请求，最终将新设备的设备名称、设备编号、稳态波形数据与暂态波形数据信息自动加入到设备库中；
    第二步、生成综合态波形数据：调用所述设备库中的所述设备编号，使用编号排列组合的计算方法，生成由不同编号组成的多种所述设备编号的排列组合；根据得到的排列组合和所述设备库中各设备的稳态波形数据，依据不同编号的排列组合，将相对应的设备波形数据进行叠加，得到若干段不同设备波形叠加的综合态波形数据；
    第三步、对多个用电设备组合运行时的电流波形数据进行特征提取得到特征集数据，然后对得到的特征集数据进行划分，将特征集数据划分为训练集和测试集，之后，使用机器学习分类器模型进行参数训练，并进行用电设备行为的准确预测；
    第四步、模型结果评价：统计每次截取到的设备稳定波段形数据时间段内，连续n个周期，每个周期各个用电设备的预测结果，来判断设备的开启、关闭的条件比例；
    所述新设备识别子流程的第七步之前，还包括根据所述新设备识别子流程的第六步进行特征相似度对比，如果存在新设备，则进入到所述分类器自我训练子流程，完成模型训练后再执行所述设备分类预测子流程的第五步；如果不存在新设备，则直接进入到所述设备分类预测子流程中的第五步，分析出哪些设备正在运行，何时开启合适关闭，得到用电设备的用能信息。
  3. 根据权利要求1或2所述的非侵入式负荷检测系统，其特征在于，所述特征提取中的特征包括电流有效值、有功功率和无功功率；具体为：
    特征1：电流有效值
    计算用电设备运行状态下的电流有效值，具体地：
    Figure PCTCN2021105242-appb-100001
    其中，I表示电流有效值，T表示一个周期，i表示瞬时电流；
    特征2：有功功率
    计算用电设备运行状态下的有功功率，具体地：
    Figure PCTCN2021105242-appb-100002
    其中，P表示有功功率，U为线电压，I为线电流，
    Figure PCTCN2021105242-appb-100003
    为U和I之间的相位差；
    特征3：无功功率
    计算用电设备运行状态下的无功功率，具体地：
    Figure PCTCN2021105242-appb-100004
  4. 根据权利要求1所述的非侵入式负荷检测方法，其特征在于，所述新设备识别子流程中的第一步、对暂态事件结果进行大动态标记，用以检测是否存在周期性功率跃迁设备，具体为：
    针对上述功率序列(P 1，P 2，...，P N)，用1标记探测到事件，用0表示没有事件发生，得到c点序列x[n]，n∈{0，1，2，3...，N-1}，表示为：
    Figure PCTCN2021105242-appb-100005
  5. 根据权利要求1所述的非侵入式负荷检测方法，其特征在于，所述新设备识别子流程中的第二步：利用峰值滤波器方法判断是否存在周期性功率跃迁，从而标记探测到的事件是否有周期性变化并从中分离非周期性大动态事件；其中的峰值滤波器令对离散傅里叶变换(Discrete Fourier Transform，DFT)得到的序列的非极大值点和极大值小于峰值θ的值等于0，得到峰值滤 波效果，
    具体的：
    对于大动态标记序列x[n]，进行离散傅里叶变换为：
    Figure PCTCN2021105242-appb-100006
    对DFT得到的序列进行峰值滤波，滤波公式表示为：
    Figure PCTCN2021105242-appb-100007
    其中
    Figure PCTCN2021105242-appb-100008
    Figure PCTCN2021105242-appb-100009
    这里的截止峰值θ，在程序里采用
    Figure PCTCN2021105242-appb-100010
    分离非周期性大动态事件的目的是得到真正是设备开启引起的事件，获得设备运行稳态取键，具体为：
    对
    Figure PCTCN2021105242-appb-100011
    进行IDFT为：
    Figure PCTCN2021105242-appb-100012
    分离后的序列表示为：
    Figure PCTCN2021105242-appb-100013
    非周期性大动态事件满足条件：
    η[n]＞η[n] max。
  6. 根据权利要求1所述的非侵入式负荷检测方法，其特征在于，所述新 设备识别子流程中的第七步分离出新设备的波形数据，具体为：
    新设备的波形经过DFT变化后的数据可描述为：
    Figure PCTCN2021105242-appb-100014
    对X[k]进行IDFT可得到新设备波形为：
    Figure PCTCN2021105242-appb-100015
    式中，x j[n]表示第j段设备运行稳定段的波形数据。
  7. 根据权利要求2所述的非侵入式负荷检测方法，其特征在于，所述分类器自我训练子流程的第二步，
    具体的：
    设已知家用电器数据库有r种家用电器，每种电器每个周期的采样点数为c，则对于电器A j：
    A j＝{I j，1，I j，2，…，I j，c}
    当采集到所有电器单独运行时的电流数据时，我们可以通过以下操作获得多个电器组合运行的电流数据，
    对于电器波形数据I j，经过DTFT分解后信号波形的模型可描述为：
    Figure PCTCN2021105242-appb-100016
    多个电器同时运行时，其信号波形经过DTFT分解后信号波形的模型可描述为：
    Figure PCTCN2021105242-appb-100017
    其中s表示同时组合运行的电器总数
    对X[k]进行IDFT为：
    Figure PCTCN2021105242-appb-100018
    其中I[n]表示多个电器组合运行时的电流波形数据。
  8. 根据权利要求2所述的非侵入式负荷检测方法，其特征在于，所述分类器自我训练子流程中的第四步，具体为：
    统计每次截取到的设备稳定段波形数据时间段内，连续n个周期，每个周期各个设备的预测结果，设备状态评价方式如下：
    对于某一次识别结果，我们对各个设备的行为标记如下：
    Figure PCTCN2021105242-appb-100019
    统计各个设备n个连续周期的预测结果，则
    Figure PCTCN2021105242-appb-100020
    故在这段时间内，各个设备的真实状态为：
    Figure PCTCN2021105242-appb-100021
    其实Status等于1时表示这段时间内设备开启，等于None时表示这段时间内设备的状态未知，等于0表示这段时间内设备关闭，p1表示判定设备开启的条件比例，p2表示判定设备一定未开启的预测结果比例。
  9. 根据权利要求1所述的所述的非侵入式负荷检测系统，其特征在于，所述分类器是神经网络或者K均值聚类算法或者支持向量机或者随机森林。
PCT/CN2021/105242 2020-08-27 2021-07-08 一种非侵入式负荷检测方法 WO2022042070A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
GB2207650.9A GB2606284A (en) 2020-08-27 2021-07-08 Non-intrusive load monitoring method
US17/779,187 US20230384355A1 (en) 2020-08-27 2021-07-08 Non-intrusive load monitoring method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010877531.4 2020-08-27
CN202010877531.4A CN112149510B (zh) 2020-08-27 2020-08-27 一种非侵入式负荷检测方法

Publications (1)

Publication Number Publication Date
WO2022042070A1 true WO2022042070A1 (zh) 2022-03-03

Family

ID=73888591

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/105242 WO2022042070A1 (zh) 2020-08-27 2021-07-08 一种非侵入式负荷检测方法

Country Status (4)

Country Link
US (1) US20230384355A1 (zh)
CN (1) CN112149510B (zh)
GB (1) GB2606284A (zh)
WO (1) WO2022042070A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114629242A (zh) * 2022-03-17 2022-06-14 广东泰迪智能科技股份有限公司 一种基于非侵入式电力负荷检测与分解方法
CN115201615A (zh) * 2022-09-15 2022-10-18 之江实验室 基于物理约束神经网络的非侵入式负荷监测方法及装置
CN115412450A (zh) * 2022-10-31 2022-11-29 南京南瑞信息通信科技有限公司 一种面向溯源图的多电力终端协同行为检测方法及装置
CN115952447A (zh) * 2023-03-14 2023-04-11 广东云山能源科技有限公司 电器负载类型智能识别系统及方法
CN116610922A (zh) * 2023-07-13 2023-08-18 浙江大学滨江研究院 基于多策略学习的非侵入式负荷识别方法和系统
CN116861318A (zh) * 2023-09-05 2023-10-10 国网浙江省电力有限公司余姚市供电公司 一种用户用电负荷分类方法、装置、设备及存储介质
CN118033208A (zh) * 2024-04-12 2024-05-14 江苏尚研电力科技有限公司 智能化的空气开关

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112149510B (zh) * 2020-08-27 2022-05-10 广东工业大学 一种非侵入式负荷检测方法
CN113077138A (zh) * 2021-03-30 2021-07-06 国家电网有限公司 一种基于波形数据特征分析的用电负荷识别方法
CN113378655B (zh) * 2021-05-24 2022-04-19 电子科技大学 一种基于深度神经网络的对抗性能量分解方法
US20230075475A1 (en) * 2021-09-07 2023-03-09 PeerNova, Inc. Outlier Determination Among Models of Future Event Prediction
CN114884075A (zh) * 2022-07-11 2022-08-09 华南理工大学 一种基于监督式学习的事件型非侵入式负荷监测方法
CN117333724B (zh) * 2023-11-28 2024-02-27 天津滨电电力工程有限公司 一种基于多特征融合图像的非侵入式负荷辨识方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140336831A1 (en) * 2013-05-08 2014-11-13 Samsung Electronics Co., Ltd. Non-intrusive load monitoring apparatus and method
CN108021736A (zh) * 2017-11-08 2018-05-11 天津大学 一种基于滑动窗残差模型的负荷投切动作监测方法
CN110514889A (zh) * 2019-07-19 2019-11-29 浙江万胜智能科技股份有限公司 一种非侵入式家庭用电负载识别的方法及系统
CN110768234A (zh) * 2019-11-13 2020-02-07 南京邮电大学 一种具有不确定性馈电恒功率负载直流微电网系统的峰值滤波方法
CN110956220A (zh) * 2019-12-11 2020-04-03 深圳市活力天汇科技股份有限公司 一种非侵入式家用电器负荷识别方法
CN111428755A (zh) * 2020-02-29 2020-07-17 国网(苏州)城市能源研究院有限责任公司 非侵入式负荷监测方法
CN112149510A (zh) * 2020-08-27 2020-12-29 广东工业大学 一种非侵入式负荷检测方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6917203B1 (en) * 2001-09-07 2005-07-12 The United States Of America As Represented By The Administrator Of The National Aeronautics And Space Administration Current signature sensor
US20130307524A1 (en) * 2012-05-02 2013-11-21 Ramot At Tel-Aviv University Ltd. Inferring the periodicity of discrete signals
FR3018058B1 (fr) * 2014-03-03 2017-08-11 Jtekt Europe Sas Procede de filtrage pour la detection des transitions d’un signal de direction assistee
GB201509209D0 (en) * 2015-05-28 2015-07-15 Micromass Ltd Echo cancellation for time of flight analogue to digital converter
CN106597277B (zh) * 2016-11-21 2019-03-15 河南理工大学 变频调速笼型异步电机转子断条故障动态诊断方法及装置
WO2018171875A1 (en) * 2017-03-21 2018-09-27 Toyota Motor Europe Nv/Sa Control device, system and method for determining the perceptual load of a visual and dynamic driving scene
CN111461189B (zh) * 2020-03-23 2022-11-15 东南大学 基于非侵入式量测技术的居民节律检测方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140336831A1 (en) * 2013-05-08 2014-11-13 Samsung Electronics Co., Ltd. Non-intrusive load monitoring apparatus and method
CN108021736A (zh) * 2017-11-08 2018-05-11 天津大学 一种基于滑动窗残差模型的负荷投切动作监测方法
CN110514889A (zh) * 2019-07-19 2019-11-29 浙江万胜智能科技股份有限公司 一种非侵入式家庭用电负载识别的方法及系统
CN110768234A (zh) * 2019-11-13 2020-02-07 南京邮电大学 一种具有不确定性馈电恒功率负载直流微电网系统的峰值滤波方法
CN110956220A (zh) * 2019-12-11 2020-04-03 深圳市活力天汇科技股份有限公司 一种非侵入式家用电器负荷识别方法
CN111428755A (zh) * 2020-02-29 2020-07-17 国网(苏州)城市能源研究院有限责任公司 非侵入式负荷监测方法
CN112149510A (zh) * 2020-08-27 2020-12-29 广东工业大学 一种非侵入式负荷检测方法

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114629242A (zh) * 2022-03-17 2022-06-14 广东泰迪智能科技股份有限公司 一种基于非侵入式电力负荷检测与分解方法
CN115201615A (zh) * 2022-09-15 2022-10-18 之江实验室 基于物理约束神经网络的非侵入式负荷监测方法及装置
CN115201615B (zh) * 2022-09-15 2022-12-20 之江实验室 基于物理约束神经网络的非侵入式负荷监测方法及装置
CN115412450A (zh) * 2022-10-31 2022-11-29 南京南瑞信息通信科技有限公司 一种面向溯源图的多电力终端协同行为检测方法及装置
CN115412450B (zh) * 2022-10-31 2023-02-14 南京南瑞信息通信科技有限公司 一种面向溯源图的多电力终端协同行为检测方法及装置
CN115952447A (zh) * 2023-03-14 2023-04-11 广东云山能源科技有限公司 电器负载类型智能识别系统及方法
CN115952447B (zh) * 2023-03-14 2023-06-02 广东云山能源科技有限公司 电器负载类型智能识别系统及方法
CN116610922A (zh) * 2023-07-13 2023-08-18 浙江大学滨江研究院 基于多策略学习的非侵入式负荷识别方法和系统
CN116861318A (zh) * 2023-09-05 2023-10-10 国网浙江省电力有限公司余姚市供电公司 一种用户用电负荷分类方法、装置、设备及存储介质
CN116861318B (zh) * 2023-09-05 2023-11-21 国网浙江省电力有限公司余姚市供电公司 一种用户用电负荷分类方法、装置、设备及存储介质
CN118033208A (zh) * 2024-04-12 2024-05-14 江苏尚研电力科技有限公司 智能化的空气开关

Also Published As

Publication number Publication date
CN112149510B (zh) 2022-05-10
US20230384355A1 (en) 2023-11-30
GB202207650D0 (en) 2022-07-06
CN112149510A (zh) 2020-12-29
GB2606284A (en) 2022-11-02

Similar Documents

Publication Publication Date Title
WO2022042070A1 (zh) 一种非侵入式负荷检测方法
CN108021736B (zh) 一种基于滑动窗残差模型的负荷投切动作监测方法
CN109387712B (zh) 基于状态矩阵决策树的非侵入式负荷检测与分解方法
CN111669375B (zh) 一种电力工控终端在线安全态势评估方法及系统
Wild et al. A new unsupervised event detector for non-intrusive load monitoring
De Baets et al. VI-based appliance classification using aggregated power consumption data
CN111428755B (zh) 非侵入式负荷监测方法
CN111027408A (zh) 一种基于支持向量机和v-i曲线特征的负荷识别方法
CN112198385A (zh) 一种非侵入式负荷监测方法
CN116559598A (zh) 一种智慧配电网故障定位方法及系统
CN110569876A (zh) 非侵入式负荷识别方法、装置及计算设备
CN110222765A (zh) 一种永磁同步电机健康状态监测方法及系统
Zhang et al. Pca-svm-based approach of detecting low-rate dos attack
CN113533910A (zh) 一种适用于换流变压器局部放电预警的方法及系统
CN114236234A (zh) 一种基于基波和谐波混合判据的电器特征辨识方法
CN115578217A (zh) 基于决策树模型的非侵入式负荷监测方法及监测系统
CN115049410A (zh) 窃电行为识别方法、装置、电子设备及计算机可读存储介质
Yuan et al. A novel current signal feature and its application in noninvasive load monitoring
CN118015812A (zh) 一种设备异常报警方法
CN111506636A (zh) 一种基于自回归和近邻算法的居民用电行为分析的系统及方法
Rizvi et al. Real-time ZIP load parameter tracking using adaptive window and variable elimination with realistic synthetic synchrophasor data
WO2023030838A1 (en) A non-intrusive load monitoring device for a site and method therefore
CN114662576A (zh) 基于有监督分类的非侵入式电瓶车充电检测方法及系统
CN115112989A (zh) 基于低频数据的非侵入式负荷监测方法
CN114912070A (zh) 一种基于多维特征分析的电机类负载非侵入式监测方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21859910

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 202207650

Country of ref document: GB

Kind code of ref document: A

Free format text: PCT FILING DATE = 20210708

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21859910

Country of ref document: EP

Kind code of ref document: A1