CN115563511B - Edge side power fingerprint identification method and device based on machine learning - Google Patents

Edge side power fingerprint identification method and device based on machine learning Download PDF

Info

Publication number
CN115563511B
CN115563511B CN202211545503.8A CN202211545503A CN115563511B CN 115563511 B CN115563511 B CN 115563511B CN 202211545503 A CN202211545503 A CN 202211545503A CN 115563511 B CN115563511 B CN 115563511B
Authority
CN
China
Prior art keywords
load switching
switching event
active power
power
value
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN202211545503.8A
Other languages
Chinese (zh)
Other versions
CN115563511A (en
Inventor
孙立明
余涛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou Shuimu Qinghua Technology Co ltd
Original Assignee
Guangzhou Shuimu Qinghua Technology Co ltd
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 Guangzhou Shuimu Qinghua Technology Co ltd filed Critical Guangzhou Shuimu Qinghua Technology Co ltd
Priority to CN202211545503.8A priority Critical patent/CN115563511B/en
Publication of CN115563511A publication Critical patent/CN115563511A/en
Application granted granted Critical
Publication of CN115563511B publication Critical patent/CN115563511B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01RMEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R21/00Arrangements for measuring electric power or power factor
    • G01R21/001Measuring real or reactive component; Measuring apparent energy
    • G01R21/002Measuring real component
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01RMEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R21/00Arrangements for measuring electric power or power factor
    • G01R21/06Arrangements for measuring electric power or power factor by measuring current and voltage
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01RMEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R29/00Arrangements for measuring or indicating electric quantities not covered by groups G01R19/00 - G01R27/00
    • G01R29/04Measuring form factor, i.e. quotient of root-mean-square value and arithmetic mean of instantaneous value; Measuring peak factor, i.e. quotient of maximum value and root-mean-square value

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Power Engineering (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a machine learning-based edge side power fingerprint identification method and device, wherein the method comprises the following steps: acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value; detecting and filtering the load switching event according to the active power value by using an improved accumulation sum algorithm to obtain a time point of the load switching event; intercepting steady-state waveforms which are a preset number of cycles away from the front and back of a time point, and obtaining single waveform data of the electric equipment with the load switching event according to the steady-state waveforms; obtaining an event waveform of a load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate power fingerprint data; and inputting the electric power fingerprint data into the trained electric power fingerprint identification model to obtain a corresponding electric equipment type identification result. The method can efficiently and accurately identify the electric power fingerprint data locally, and improves the robustness in a real and complex electricity utilization environment.

Description

Edge side power fingerprint identification method and device based on machine learning
Technical Field
The invention relates to the technical field of power data processing, in particular to a marginal side power fingerprint identification method and device based on machine learning.
Background
With the advance of "dual carbon" and new power systems, low voltage users are required to be able to participate in demand response, enabling source-grid interaction. The electric power fingerprint technique comes to life at the end, through accurate discernment, the accurate control to the load, makes the low pressure user can initiatively participate in the demand response for the power consumption is more wisdom.
The electric power fingerprint identification technology is developed from an electric power load identification technology and is an important component of the electric power fingerprint technology. The electric power fingerprint identification technology utilizes an artificial intelligence technology and a big data technology to excavate 'electric power fingerprint' characteristics capable of representing equipment characteristics by collecting electric data of electric equipment.
In the prior art, however, the electric power fingerprint identification scheme aiming at the edge intelligent terminal is less, and the electric power fingerprint characteristics are difficult to be dug out locally and efficiently; meanwhile, the existing electric power fingerprint identification scheme usually ignores interference in a real complex power utilization environment, so that the accuracy of an identification result is not high.
Disclosure of Invention
The invention aims to provide a machine learning-based edge side power fingerprint identification method and device, and aims to solve the technical problem that power fingerprint data are difficult to identify efficiently and accurately on an edge intelligent terminal with limited computing power in the prior art.
The purpose of the invention can be realized by the following technical scheme:
an edge side power fingerprint identification method based on machine learning comprises the following steps:
acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value;
detecting and filtering load switching events according to the active power values by utilizing an improved accumulation sum algorithm to obtain time points of the load switching events;
intercepting steady-state waveforms which are a preset number of cycles away from the front and back of the time point, and obtaining monomer waveform data of the electric equipment with the load switching event according to the steady-state waveforms;
obtaining an event waveform of the load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate electric power fingerprint data;
and inputting the electric power fingerprint data into a trained electric power fingerprint identification model to obtain a corresponding electric equipment type identification result, wherein the trained electric power fingerprint identification model is obtained by utilizing electric power fingerprint data and a corresponding load label to train based on a LightGBM algorithm.
Optionally, the improved accumulation and algorithm includes power value filtering and dual time window filtering, and the detecting and filtering of the load switching event according to the active power value by using the improved accumulation and algorithm includes:
screening the active power value through power value screening by utilizing an improved accumulation sum algorithm to obtain an active power average value, and obtaining a load switching event detection result according to the active power average value and a preset threshold value;
and filtering the load switching events in the load switching event detection result through double-time window filtering to obtain a final load switching event detection result.
Optionally, the screening the active power value through power value screening by using an improved accumulation sum algorithm to obtain an active power average value includes:
the improved accumulation sum algorithm adopts two adjacent active power sliding windows with the widths respectively being a first width and a second width, the active power values in the active power sliding windows are sorted respectively, the active power values with the sizes ranging from a first preset interval to a second preset interval are screened out, and then the average value is calculated to obtain the average value of the active power.
Optionally, obtaining a load switching event detection result according to the active power average value and a preset threshold includes:
according to the active power average value and the preset threshold value
Figure SMS_1
Obtaining the detection result of the load switching event, if so
Figure SMS_2
If yes, a load switching event occurs, otherwise, no load switching event occurs;
in the formula (I), the compound is shown in the specification,
Figure SMS_3
Figure SMS_4
respectively refers to the average value of the active power in the active power sliding window with the width of a first width and a second width,
Figure SMS_5
is the variance of an active power sliding window having a width of a first width,
Figure SMS_6
is a preset threshold.
Optionally, filtering the load switching event in the load switching event detection result through double-time window filtering, and obtaining a final load switching event detection result includes:
setting two time windows with widths respectively being a first number and a second number of cycles, starting to wait and count a third number of cycles after detecting that a load switching event occurs, and if the load switching event is detected again in the first number of cycles, determining that the two time windows still belong to the same load switching event;
if the load switching events are detected again within the first number of cycles to the third number of cycles, deleting all unidentified load switching events to obtain a final load switching event detection result; the third number is the sum of the first number and the second number.
Optionally, obtaining the single waveform data of the electrical equipment with the load switching event according to the steady-state waveform includes:
and after the steady-state waveforms are aligned by using the positive zero point of the voltage, monomer waveform data of the electric equipment with the load switching event are obtained through differential calculation.
Optionally, obtaining the event waveform of the load switching event according to the monomer waveform data includes:
and carrying out mean value filtering on the monomer waveform data, dividing each cycle by a positive zero point of voltage, and taking the mean value of each cycle as an event waveform of the load switching event.
Optionally, the time domain features include: current effective value, current peak-to-peak value, current peak-to-average ratio, active power, reactive power, apparent power and power factor.
Optionally, the frequency domain feature further includes three combined features of a current total harmonic factor, a current odd-even harmonic ratio, and a current harmonic spectrum centroid, in addition to the current harmonic amplitude and the phase of 20 to 30 times.
The invention also provides a device for recognizing the edge side electric power fingerprint based on machine learning, which comprises:
the electric data acquisition module is used for acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value;
the event time point determining module is used for detecting and filtering a load switching event according to the active power value by utilizing improved accumulation and algorithm to obtain a time point of the load switching event;
the waveform data acquisition module is used for intercepting steady-state waveforms which are a preset number of cycles away from the time point and obtaining single waveform data of the electric equipment with the load switching event according to the steady-state waveforms;
the power fingerprint data generation module is used for obtaining an event waveform of the load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate power fingerprint data;
and the electric equipment type identification module is used for inputting the electric fingerprint data into a trained electric fingerprint identification model to obtain a corresponding electric equipment type identification result, and the trained electric fingerprint identification model is obtained by training the electric fingerprint data and a corresponding load label based on a LightGBM algorithm.
The invention provides a machine learning-based edge side power fingerprint identification method and device, wherein the method comprises the following steps: acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value; detecting and filtering load switching events according to the active power values by utilizing an improved accumulation sum algorithm to obtain time points of the load switching events; intercepting steady-state waveforms which are a preset number of cycles away from the front and back of the time point, and obtaining monomer waveform data of the electric equipment with the load switching event according to the steady-state waveforms; obtaining an event waveform of the load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate electric power fingerprint data; and inputting the electric power fingerprint data into a trained electric power fingerprint identification model to obtain a corresponding electric equipment type identification result, wherein the trained electric power fingerprint identification model is obtained by utilizing electric power fingerprint data and a corresponding load label to train based on a LightGBM algorithm.
Therefore, the invention has the beneficial effects that:
the method collects and processes high-frequency electrical quantity in real time, detects load switching events by utilizing improved robust accumulation and algorithm, improves event waveform extraction, extracts multi-dimensional distinctive features from a time domain and a frequency domain respectively, generates high-dimensional power fingerprint data, can fully represent fine-grained characteristics of power loads, and realizes efficient and accurate power fingerprint data identification based on LightGBM which is adaptive to edge side high-instantaneity, low-computing power and easy-to-deploy requirements. The method can give consideration to the richness, identification accuracy and edge deployment easiness of the electric power fingerprint, can locally and efficiently complete the whole process of event detection, event waveform extraction and electric power fingerprint data generation and identification, and can avoid the problems of network congestion and privacy leakage of household power consumption data transmission. Meanwhile, the invention considers the interference of various electric appliance running states and improves the robustness and the accuracy of the algorithm in the real and complex power utilization environment.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the method of the present invention;
FIG. 2 is a schematic flow chart of another embodiment of the method of the present invention;
FIG. 3 is a schematic flow chart of an embodiment of the apparatus of the present invention.
Detailed Description
Interpretation of terms:
light Gradient Boosting Machine (LightGBM) is a Decision Tree algorithm based distributed Gradient Boosting (GBDT) framework sourced by the microsoft asia institute distributed Machine learning toolkit (DMTK) team.
LightGBM optimization:
if the GBDT is to be optimized, there are two aspects:
(1) The scale of the training set is reduced, obviously, the calculation amount can be reduced, and the calculation efficiency of the algorithm is improved.
(2) And the characteristic dimension is reduced, so that the calculated amount can be reduced when the split point is selected, and the performance of the algorithm is improved.
But directly reducing the training set size or reducing the feature dimensions obviously sacrifices the accuracy of the model. The LightGBM algorithm improves the training speed and the reasoning speed of model edge deployment by optimizing the sampling of sample points during model training and optimizing the feature dimension during selecting split points without sacrificing precision.
The embodiment of the invention provides a machine learning-based edge side power fingerprint identification method and device, and aims to solve the technical problem that power fingerprint data are difficult to identify efficiently and accurately on an edge intelligent terminal with limited computing power in the prior art.
To facilitate an understanding of the invention, the invention will now be described more fully with reference to the accompanying drawings. Preferred embodiments of the present invention are shown in the drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
With the construction of advanced measurement systems, load identification technology for sensing energy consumption information of electric equipment level becomes a research trend. The power fingerprint identification technology is developed from a load identification technology, power fingerprint data highly representing the fine granularity characteristic of the power load is constructed by calculating load characteristics of time domain and frequency domain dimensions, and accurate identification is realized by combining an advanced machine learning technology. However, a few of methods can realize real-time monitoring of household electricity consumption only based on an edge intelligent terminal with limited computing power, namely, the whole process of event detection, event waveform extraction, electricity fingerprint data generation and identification is completed locally and efficiently, so that the problems of network congestion and privacy disclosure of household electricity consumption data transmission are avoided. Meanwhile, the interference in the real power utilization environment is usually ignored by the existing power fingerprint identification method, which causes the difficulty in popularization and application of the technology.
The following two problems still exist at present:
(1) The electric power fingerprint identification full-flow implementation scheme aiming at the edge intelligent terminal is less, and the richness and identification accuracy of the electric power fingerprint and the usability of edge deployment cannot be considered, so that the load identification technology is difficult to fall to the ground.
(2) Most researches lack consideration on application of the recognition model in a real and complex power utilization environment, and influence of long transient, spike noise, fluctuation of the running state of a base electrical appliance and the like is difficult to deal with, so that the robustness of the recognition model is low.
The invention provides a machine learning-based edge side power fingerprint identification method and device, which can generate high-dimensional power fingerprint data, can represent fine-grained characteristics of power loads, considers the identification accuracy and the edge deployment usability, considers the interference of various electrical appliance running states, can realize real-time accurate monitoring on household electrical appliances only by using an edge intelligent terminal deploying a machine learning model in a complex real environment, and is a key technology for practical power fingerprint identification.
Referring to fig. 1, an embodiment of a method for recognizing edge power fingerprint based on machine learning according to the present invention includes:
s100: acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value;
s200: detecting and filtering load switching events according to the active power values by utilizing an improved accumulation sum algorithm to obtain time points of the load switching events;
s300: intercepting steady-state waveforms which are a preset number of cycles away from the front and back of the time point, and obtaining monomer waveform data of the electric equipment with the load switching event according to the steady-state waveforms;
s400: obtaining an event waveform of the load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate electric power fingerprint data;
s500: and inputting the electric power fingerprint data into a trained electric power fingerprint identification model to obtain a corresponding electric equipment type identification result, wherein the trained electric power fingerprint identification model is obtained by training by using the electric power fingerprint data and a corresponding load label based on a LightGBM algorithm.
In step S100, voltage and current data of various electric devices are collected in real time according to a preset sampling frequency, and an active power value is obtained through calculation. Specifically, the edge intelligent terminal collects voltage and current data at the electrical interface of the house in real time, can obtain the voltage and current data of various electric equipment in real time, and calculates the active power value according to the obtained voltage and current data. According to the obtained real-time voltage and current waveforms, the active power value can be calculated by using the formula (1):
Figure SMS_7
; (1)
in the formula (I), the compound is shown in the specification,
Figure SMS_8
the instantaneous value of the active power of a single cycle is represented as the active power value;
Figure SMS_9
the number of sampling points of a single cycle,
Figure SMS_10
and
Figure SMS_11
respectively mean the first of a single cycle
Figure SMS_12
Instantaneous current and voltage values of sampling points.
In this embodiment, the edge intelligent terminal can be an integrated intelligent socket, an intelligent air switch, and an intelligent electric meter which have real-time high-frequency acquisition and calculation of electric quantity data and carry on a machine learning model, and the full process of electric power fingerprint data identification is realized locally at the terminal without uploading the electric quantity data to a server for calculation. The edge intelligent terminal can avoid network congestion caused by communication with a server in practical application, the response speed of the terminal is improved, and privacy of a user can be guaranteed through local operation.
Specifically, the real-time acquisition means that the sampling frequency of the electrical quantity data is not lower than 1kHz, and the time from the occurrence of a load switching event to the output of an identification result is less than 5 seconds; the house electrical interface is a socket end and an access bus end. In a preferred embodiment, the preset sampling frequency of the electrical quantity data may be 7.5kHz.
In step S200, a load switching event is detected and filtered according to the active power value by using an improved accumulation sum algorithm, so as to obtain a time point at which the load switching event occurs.
In this embodiment, the improved accumulation SUM algorithm is also referred to as an improved CUSUM (Cumulative SUM) algorithm, the improved CUSUM algorithm adds a power value screening function and a double-time-window filtering function on the basis of the conventional CUSUM algorithm, and detects and filters the load switching event according to the active power value by using the improved accumulation SUM algorithm to obtain a time point at which the load switching event occurs.
Specifically, an active power value is screened through power value screening by utilizing an improved accumulation sum algorithm to obtain an active power average value, and a load switching event detection result is obtained according to the active power average value and a preset threshold value; and filtering the load switching events in the load switching event detection result through double-time-window filtering to obtain a final load switching event detection result.
In this embodiment, screening the active power values through power value screening by using an improved accumulation sum algorithm to obtain an active power average value includes:
the improved accumulation sum algorithm adopts two adjacent active power sliding windows with the widths respectively being a first width and a second width, the active power values in the active power sliding windows are sorted respectively, the active power values with the sizes ranging from a first preset interval to a second preset interval are screened out, and then the average value is calculated to obtain the active power average value.
Specifically, when the improved CUSUM algorithm is used for screening the power value, the widths are respectively
Figure SMS_13
Figure SMS_14
The two adjacent active power sliding windows are used for sorting active power values in the two active power sliding windows respectively, screening the active power values with the size within the interval of 5% -95%, and then calculating the average value to obtain the active power average value. The active power average value can reduce the influence of the load switching transient process and the spike noise of the operation of the electric appliance (electric equipment) on the detection of the load switching event.
In this embodiment, after the average value of the active power is obtained, the detection result of the load switching event is obtained according to the average value of the active power and the preset threshold, and the detection result of the load switching event is obtained by specifically performing judgment by using the following formula (2):
Figure SMS_15
;(2)
if it is
Figure SMS_16
If yes, a load switching event occurs; otherwise, no load switching event occurs.
In the formula (I), the compound is shown in the specification,
Figure SMS_17
Figure SMS_18
respectively refers to the average value of the active power in the active power sliding window with the width of a first width and a second width,
Figure SMS_19
is of a first widthThe variance of the active power sliding window is,
Figure SMS_20
is a preset threshold.
In particular, variance
Figure SMS_21
The formula (3) is shown as follows:
Figure SMS_22
;(3)
in the formula (I), the compound is shown in the specification,
Figure SMS_23
means the active power sliding window with the first width
Figure SMS_24
An active power value; if it is
Figure SMS_25
Then it is a load input event, if
Figure SMS_26
Then a load shedding event.
In this embodiment, filtering the load switching event in the load switching event detection result by the double-time window filtering to obtain a final load switching event detection result includes:
setting two time windows with widths respectively being a first number of cycles and a second number of cycles, starting to wait and counting a third number of cycles after detecting that a load switching event occurs, and if the load switching event is detected again in the first number of cycles, determining that the two time windows still belong to the same load switching event;
if the load switching events are detected again in the cycles from the first number to the third number, deleting all unidentified load switching events to obtain a final load switching event detection result; the third quantity is the sum of the first quantity and the second quantity.
Specifically, in the modified CUSUM algorithm, twoThe two widths are set to be respectively
Figure SMS_29
Figure SMS_31
The time window of each cycle begins to wait and count after detecting the occurrence of the load switching event
Figure SMS_33
+
Figure SMS_27
A cycle of waves in
Figure SMS_30
If the load switching event is detected again in each cycle, the same event is considered to belong to; if at
Figure SMS_32
To
Figure SMS_34
+
Figure SMS_28
And if the load switching event is detected again in each cycle, deleting all unidentified events so as to filter out repeated load switching events caused by the long transient state of the switching of the electrical appliance and obtain a final load switching event judgment result.
The double-time window filtering is added to further filter the detection result of the load switching event, so that the influence of the long transient electric appliance such as high-power heating equipment and the fluctuation of the running state of the base electric appliance on the event detection can be effectively reduced. Compared with the original CUSUM algorithm, the improved CUSUM algorithm in the embodiment improves the anti-interference performance applied in the real complex power utilization environment, and balances the real-time performance and the robustness. In the course of the specific implementation process,
Figure SMS_35
is a total of 150 a of the total weight of the composition,
Figure SMS_36
Figure SMS_37
can be set to 30, preset threshold
Figure SMS_38
It is possible to set the number to 200,
Figure SMS_39
Figure SMS_40
which may be set to 100 and 20, respectively.
In step S300, a steady waveform that is a preset number of cycles away from the time point is intercepted, and monomer waveform data of the electrical equipment in which the load switching event occurs is obtained according to the steady waveform.
In this embodiment, steady-state waveforms before and after the occurrence time point of the load switching event in step 200 are taken, and after the positive zero points of the voltage are aligned, monomer waveform data of the switching appliance is obtained through differential calculation. Specifically, the time point of the event occurrence in step S200 is taken to be spaced from the front and back
Figure SMS_41
And (3) aligning the steady-state waveforms of the cycles by using the positive zero points of the voltage, and then obtaining monomer waveform data of the switching electrical appliance through differential calculation. Before and after the event occurrence time point obtained in step S200
Figure SMS_42
At each cycle, take forward and backward
Figure SMS_43
The current and voltage waveform data of each cycle is used as the steady state waveform before and after the event occurrence time point.
The voltage zero-positive alignment means that two sections of steady-state voltage waveforms are respectively subjected to zero-positive searching, and the width is used
Figure SMS_44
The sliding window slides from head to tail, when the sliding window is in front
Figure SMS_45
All voltage instantaneous values are less than zero
Figure SMS_46
The middle point of the sliding window when the instantaneous values of the voltages are all larger than zero is the positive zero point, the first positive zero point of the two sections of steady-state voltage waveforms is taken as the alignment point of the two sections of steady-state current waveforms, and aliasing waveform separation distortion caused by phase difference can be avoided.
The difference calculation means that the aligned current instantaneous values are directly subtracted, and if the current instantaneous values are load input events, the current instantaneous values are obtained by subtracting the current waveform before the events occur from the current waveform after the events occur; and if the event is a load shedding event, subtracting the steady-state current waveform after the event from the steady-state current waveform before the event occurs.
In a preferred embodiment of the present invention,
Figure SMS_47
in general it may be of the order of 100,
Figure SMS_48
it may be in the form of a 120,
Figure SMS_49
it is possible to set the number to 20,
Figure SMS_50
may be set to 9.
In step S400, an event waveform of the load switching event is obtained according to the monomer waveform data, and a time domain feature and a frequency domain feature are extracted from the event waveform to generate power fingerprint data.
In step S400, the single waveform data obtained in step S300 is subjected to mean filtering, and then each cycle is divided by a positive voltage zero point, and the mean value of each cycle is taken as an event waveform.
Mean filtering refers to using a width of
Figure SMS_51
The sliding window of (2) slides from the beginning to the end of the steady-state current waveform obtained in step (300) after the event occurs, and the current value at the midpoint of the sliding window is replaced byAnd (3) averaging the original instantaneous values of the steady-state current in the sliding window to filter the peak noise of the operation of the electric appliance. The mean value of each cycle is divided into
Figure SMS_52
After each cycle, in the same order
Figure SMS_53
The sampling points of the voltage and the current are averaged to reduce the influence of the fluctuation of the running state of the substrate electric appliance, and the average value is obtained
Figure SMS_54
The event waveform is composed of voltage and current average value points. In the course of the specific implementation,
Figure SMS_55
may be set to 5.
In step S400, time domain features and frequency domain features are extracted from the event waveform, and high-dimensional power fingerprint data is generated. The time domain features in the power fingerprint data may include: current effective value, current peak-to-peak value, current peak-to-average ratio, active power, reactive power, apparent power and power factor. In addition to the current harmonic amplitude and the phase thereof of 2 to 30 times, three combined characteristics of a current total harmonic factor, a current odd-even harmonic ratio and a current harmonic spectrum centroid are added to the frequency domain characteristics in the power fingerprint data so as to fully exploit the distinguishability of each power load on the frequency domain.
Wherein the effective value of the current
Figure SMS_56
Is calculated as shown in equation (4):
Figure SMS_57
; (4)
in the formula (I), the compound is shown in the specification,
Figure SMS_58
the number of sampling points of a single cycle,
Figure SMS_59
mean the first of a single cycle
Figure SMS_60
Instantaneous current values of the sampling points.
Peak to peak current value
Figure SMS_61
Is calculated as shown in equation (5):
Figure SMS_62
;(5)
peak to average ratio of current
Figure SMS_63
Is calculated as shown in equation (6):
Figure SMS_64
;(6)
apparent power
Figure SMS_65
Is calculated as shown in equation (7):
Figure SMS_66
;(7)
in the formula (I), the compound is shown in the specification,
Figure SMS_67
mean the first of a single cycle
Figure SMS_68
The instantaneous voltage values of the sampling points.
Reactive power
Figure SMS_69
Is calculated as shown in equation (8):
Figure SMS_70
;(8)
power factor
Figure SMS_71
Is calculated as shown in equation (9):
Figure SMS_72
;(9)
obtaining the amplitude of the first 30 current harmonics, the phase thereof and the total harmonic factor of the current based on Fourier decomposition
Figure SMS_73
Is calculated as shown in equation (10):
Figure SMS_74
;(10)
in the formula (I), the compound is shown in the specification,
Figure SMS_75
is the effective value of the current fundamental wave;
current odd-even harmonic ratio
Figure SMS_76
Is represented by equation (11):
Figure SMS_77
;(11)
in the formula (I), the compound is shown in the specification,
Figure SMS_78
is as follows
Figure SMS_79
The amplitude of the sub-current harmonics; centroid of current harmonic spectrum
Figure SMS_80
The formula (2) is shown in formula (12):
Figure SMS_81
;(12)
in the formula (I), the compound is shown in the specification,
Figure SMS_82
the frequency of the fundamental wave of the current.
Preferred embodiments, in practice,
Figure SMS_83
the power fingerprint data is 50, has 68-dimensional characteristics, and can fully characterize the fine-grained characteristics of the household appliance.
In step S500, the power fingerprint data is input into a trained power fingerprint identification model to obtain a corresponding electric device type identification result, where the trained power fingerprint identification model is obtained by training with the power fingerprint data and a corresponding load tag based on a LightGBM algorithm.
Firstly, the electric power fingerprint data obtained in step S400 and the corresponding load label are used to train the electric power fingerprint identification model based on the LightGBM algorithm to convergence, the algorithm judgment logic can be converted into C language, and after deployment and use, the electric power fingerprint data obtained in steps S100 to S400 are directly input into the electric power fingerprint identification model based on the LightGBM to obtain the electric equipment type identification result corresponding to the electric power fingerprint data.
It should be noted that the present embodiment is mainly used for sensing the household electrical equipment, and the load label refers to the type of the household electrical equipment or the type of the electrical appliance, such as a hair dryer, an electric cooker, a computer, and the like.
It should be noted that LightGBM belongs to an ensemble learning model using a tree model as a base learner. Compared with a neural network model, the method has the advantages that the LightGBM is high in operation speed, low in consumed computing resource and high in identification accuracy, is particularly suitable for application scenes of hardware deployment, and can realize full-flow power fingerprint identification at local edge intelligent terminals with limited performance.
The LightGBM performs unilateral gradient sampling on the samples in the training process of the subtrees, and the variance gain adopted by leaf node feature splitting is as shown in formula (13):
Figure SMS_84
;(13)
in the formula (I), the compound is shown in the specification,
Figure SMS_104
mean the first
Figure SMS_108
Characteristic value at the splitting threshold of
Figure SMS_112
The gain of the variance of the time-of-day,
Figure SMS_87
mean the first
Figure SMS_93
The power fingerprint data of the bar input is,
Figure SMS_96
for the number of samples to be trained,
Figure SMS_100
is a first
Figure SMS_88
Characteristic value less than split threshold
Figure SMS_91
The number of samples of (a) to (b),
Figure SMS_95
is as follows
Figure SMS_99
The characteristic value is greater than the splitting threshold
Figure SMS_92
The number of samples of (a) to (b),
Figure SMS_98
is the first in the corresponding set
Figure SMS_102
The negative gradient of individual samples with respect to the load signature,
Figure SMS_105
Figure SMS_101
is a coefficient of a gradient sampling,
Figure SMS_106
before the gradient absolute value is ranked from large to small
Figure SMS_110
In the sample set of (1)
Figure SMS_114
A characteristic value less than
Figure SMS_86
The set of samples of (a) is,
Figure SMS_90
is ranked before the absolute value of the gradient is from big to small
Figure SMS_94
In the sample set of (1)
Figure SMS_97
A characteristic value is greater than
Figure SMS_103
Set of samples of (1), setBIs a setARandom sampling of non-samples
Figure SMS_107
The composition of the sample is determined by the composition of the sample,
Figure SMS_109
as a set of samplesBTo middle
Figure SMS_115
A characteristic value less than
Figure SMS_111
The set of samples of (a) is,
Figure SMS_113
as a set of samplesBTo middle
Figure SMS_116
A characteristic value is greater than
Figure SMS_117
The set of samples of (a); random stay in training to convergence
Figure SMS_85
The training sample is used as a verification set, and the LightGBM electric power fingerprint identification model has identification accuracy in the verification set
Figure SMS_89
Not rising within one round.
In the course of the specific implementation,
Figure SMS_118
Figure SMS_119
can be respectively set to be 0.2 and 0.1,
Figure SMS_120
the content was set to 20%,
Figure SMS_121
may be set to 30.
It should be noted that the epoch parameter is used to control whether the electric power fingerprint identification model based on the LightGBM algorithm stops training, and when the identification accuracy of the electric power fingerprint identification model in the verification set is in the verification set
Figure SMS_122
And when the number of rounds does not rise, stopping training the electric power fingerprint identification model based on the LightGBM algorithm, and obtaining the trained electric power fingerprint identification model based on the LightGBM algorithm.
In addition, in the specific implementation process, steps S100 to S500 in this embodiment are all executed by an edge intelligent terminal, and all the calculations of the present invention may be implemented based on C language in the edge intelligent terminal taking an STM32H750 chip as a core.
The trained electric power fingerprint identification model based on the LightGBM algorithm is deployed on the edge intelligent terminal, and the electric power fingerprint identification model is obtained based on LightGBM algorithm training, so that the method has the advantages of high operation speed, less consumption of computing resources, high identification accuracy and the like, is particularly suitable for application scenes of hardware edge deployment, can realize full-flow electric power fingerprint data identification locally on the edge intelligent terminal with limited performance, can identify the electric power fingerprint data locally and efficiently and accurately on the edge intelligent terminal, and improves robustness, accuracy and safety in real and complex power utilization environments.
According to the edge side power fingerprint identification method based on machine learning, high-frequency electrical quantity is collected and processed in real time, load switching events are detected through improved robust accumulation and algorithm, event waveform extraction is improved, multidimensional distinguishing characteristics are extracted from time domains and frequency domains respectively, high-dimensional power fingerprint data are generated, fine-grained characteristics of power loads can be represented fully, and efficient and accurate power fingerprint data identification is achieved based on LightGBM which is adaptive to high instantaneity, low computing power and easy deployment requirements of the edge side. The invention can give consideration to the richness of the electric power fingerprint, the identification accuracy and the edge deployment easiness, can locally and efficiently complete the whole processes of event detection, event waveform extraction and electric power fingerprint data generation and identification, and can avoid the problems of network congestion and privacy leakage of household electric power data transmission. Meanwhile, the invention considers the interference of various electric appliance running states and improves the robustness and the accuracy of the algorithm in the real and complex power utilization environment.
Referring to fig. 2, another embodiment of a method for edge power fingerprint identification based on machine learning according to the present invention includes:
1) Collecting high-frequency voltage and current data in real time through an edge intelligent terminal;
2) Calculating active power according to the obtained voltage and current data, and detecting a load switching event based on an improved CUSUM algorithm of adding power value screening and double-time-window filtering to obtain an event occurrence time point;
3) Taking the steady-state waveforms before and after the event in the step 2), aligning by using positive zero points of voltage, and obtaining monomer waveform data of the switching electrical appliance through differential calculation;
4) Carrying out mean value filtering on the monomer waveform data, dividing each cycle by a positive zero point of voltage, and taking the mean value of each cycle as an event waveform;
5) Extracting time domain and frequency domain characteristics from the event waveform to generate high-dimensional power fingerprint data;
6) The method comprises the steps of training a LightGBM-based power fingerprint identification model to be convergent by using power fingerprint data and a corresponding load label, converting algorithm judgment logic into C language, and directly identifying the power fingerprint data obtained in the steps 1) to 5) after deployment and use to obtain an identification result.
Referring to fig. 3, the present invention further provides an embodiment of an edge side power fingerprint identification apparatus based on machine learning, including:
the electric data acquisition module 11 is used for acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value;
the event time point determining module 22 is configured to detect and filter a load switching event according to the active power value by using an improved accumulation sum algorithm, so as to obtain a time point at which the load switching event occurs;
the waveform data acquisition module 33 is configured to intercept a steady-state waveform having a preset number of cycles before and after the time point, and obtain single waveform data of the electrical equipment in which the load switching event occurs according to the steady-state waveform;
the electric power fingerprint data generation module 44 is configured to obtain an event waveform of the load switching event according to the monomer waveform data, and extract a time domain feature and a frequency domain feature from the event waveform to generate electric power fingerprint data;
and the electric equipment type identification module 55 is used for inputting the electric fingerprint data into a trained electric fingerprint identification model to obtain a corresponding electric equipment type identification result, wherein the trained electric fingerprint identification model is obtained by training electric fingerprint data and a corresponding load label based on a LightGBM algorithm.
Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) The invention provides a set of full-flow practicability scheme, which comprises real-time high-frequency electric quantity acquisition and processing, robust event detection and event waveform extraction, high-dimensional feature extraction, accurate electric power fingerprint identification and practical and feasible edge identification model deployment, wherein all the steps in the invention can be realized only in an edge intelligent terminal taking an STM32H750 chip as a core based on C language.
(2) The method improves event detection and event waveform extraction aiming at the interference caused by the operation states of various electric appliances, such as long transient state of electric appliance switching, spike noise of electric appliance operation, fluctuation of the operation state of a base electric appliance and the like, and improves the robustness of an algorithm in a real and complex power utilization environment.
(3) The invention extracts multidimensional distinguishing characteristics from time domain and frequency domain respectively, especially constructs combined frequency domain characteristics, generates rich electric power fingerprint data, and realizes accurate electric power fingerprint identification based on LightGBM which is adaptive to edge side and has high real-time performance, low computational power and easy deployment requirement.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one type of logical functional division, and other divisions may be realized in practice, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or may also be implemented in the form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
The above examples are only intended to illustrate the technical solution of the present invention, and not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (8)

1. An edge side power fingerprint identification method based on machine learning is characterized by comprising the following steps:
acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value;
detecting and filtering load switching events according to the active power values by utilizing an improved accumulation sum algorithm to obtain time points of the load switching events;
the improved accumulation and algorithm comprises power value screening and double-time window filtering, and the detection and filtering of the load switching event according to the active power value by using the improved accumulation and algorithm comprises the following steps:
screening the active power value through power value screening by utilizing an improved accumulation sum algorithm to obtain an active power average value, and obtaining a load switching event detection result according to the active power average value and a preset threshold value;
filtering the load switching events in the load switching event detection results through double-time window filtering to obtain final load switching event detection results;
the method for obtaining the load switching event detection result according to the active power average value and the preset threshold value comprises the following steps:
according to the active power average value and the preset threshold value
Figure QLYQS_1
Obtaining the detection result of the load switching event, if so
Figure QLYQS_2
If yes, a load switching event occurs, otherwise, no load switching event occurs;
in the formula (I), the compound is shown in the specification,
Figure QLYQS_3
Figure QLYQS_4
respectively refers to the average value of the active power in the active power sliding window with the width of a first width and a second width,
Figure QLYQS_5
is the variance of an active power sliding window having a width of a first width,
Figure QLYQS_6
is a preset threshold value;
intercepting steady-state waveforms which are a preset number of cycles away from the front and back of the time point, and obtaining monomer waveform data of the electric equipment with the load switching event according to the steady-state waveforms;
obtaining an event waveform of the load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate electric power fingerprint data;
and inputting the electric power fingerprint data into a trained electric power fingerprint identification model to obtain a corresponding electric equipment type identification result, wherein the trained electric power fingerprint identification model is obtained by utilizing electric power fingerprint data and a corresponding load label to train based on a LightGBM algorithm.
2. The machine learning-based edge side power fingerprinting method of claim 1, characterized in that the filtering of the active power values by power value filtering with improved accumulation sum algorithm to get the active power average value comprises:
the improved accumulation sum algorithm adopts two adjacent active power sliding windows with the widths respectively being a first width and a second width, the active power values in the active power sliding windows are sorted respectively, the active power values with the sizes ranging from a first preset interval to a second preset interval are screened out, and then the average value is calculated to obtain the average value of the active power.
3. The machine learning-based edge-side power fingerprint identification method according to claim 1, wherein filtering the load switching events in the load switching event detection results through double-time window filtering to obtain final load switching event detection results comprises:
setting two time windows with widths respectively being a first number of cycles and a second number of cycles, starting to wait and counting a third number of cycles after detecting that a load switching event occurs, and if the load switching event is detected again in the first number of cycles, determining that the two time windows still belong to the same load switching event;
if the load switching events are detected again within the first number of cycles to the third number of cycles, deleting all unidentified load switching events to obtain a final load switching event detection result; the third number is the sum of the first number and the second number.
4. The machine learning-based edge-side power fingerprint identification method according to claim 1, wherein obtaining the single waveform data of the power-consuming equipment with the load switching event according to the steady-state waveform comprises:
and after the steady-state waveforms are aligned by using the positive zero point of the voltage, monomer waveform data of the electric equipment with the load switching event are obtained through differential calculation.
5. The machine learning-based edge-side power fingerprint identification method according to claim 1, wherein obtaining the event waveform of the load switching event according to the cell waveform data comprises:
and carrying out mean value filtering on the monomer waveform data, dividing each cycle by a positive zero point of voltage, and taking the mean value of each cycle as an event waveform of the load switching event.
6. The machine learning-based edge-side power fingerprinting method of claim 1, characterized in that the time-domain features comprise: current effective value, current peak-to-peak value, current peak-to-average ratio, active power, reactive power, apparent power and power factor.
7. An edge side power fingerprint identification method based on machine learning as claimed in claim 1 wherein three combined features of current total harmonic factor, current odd-even harmonic ratio and current harmonic spectrum centroid are added in addition to the current harmonic amplitude and phase of 20 to 30 times of frequency domain feature.
8. An edge side electric power fingerprint identification device based on machine learning, comprising:
the electric data acquisition module is used for acquiring voltage and current data of various electric equipment in real time according to a preset sampling frequency and calculating to obtain an active power value;
the event time point determining module is used for detecting and filtering a load switching event according to the active power value by utilizing an improved accumulation sum algorithm to obtain a time point of the load switching event;
the improved accumulation and algorithm comprises power value screening and double-time window filtering, and the detection and filtering of the load switching event according to the active power value by using the improved accumulation and algorithm comprises the following steps:
screening the active power value through power value screening by utilizing an improved accumulation sum algorithm to obtain an active power average value, and obtaining a load switching event detection result according to the active power average value and a preset threshold value;
filtering the load switching events in the load switching event detection results through double-time window filtering to obtain final load switching event detection results;
the method for obtaining the load switching event detection result according to the active power average value and the preset threshold value comprises the following steps:
according to the active power average value and the preset threshold value
Figure QLYQS_7
Obtaining the detection result of the load switching event, if so
Figure QLYQS_8
If it is true, then a negativity occursA load switching event is generated, otherwise, the load switching event does not occur;
in the formula (I), the compound is shown in the specification,
Figure QLYQS_9
Figure QLYQS_10
respectively refers to the average value of the active power in the active power sliding window with the width of a first width and a second width,
Figure QLYQS_11
is the variance of an active power sliding window having a width of a first width,
Figure QLYQS_12
is a preset threshold;
the waveform data acquisition module is used for intercepting steady-state waveforms which are a preset number of cycles away from the time point and obtaining monomer waveform data of the electric equipment with the load switching event according to the steady-state waveforms;
the power fingerprint data generation module is used for obtaining an event waveform of the load switching event according to the monomer waveform data, and extracting time domain characteristics and frequency domain characteristics from the event waveform to generate power fingerprint data;
and the electric equipment type identification module is used for inputting the electric fingerprint data into a trained electric fingerprint identification model to obtain a corresponding electric equipment type identification result, and the trained electric fingerprint identification model is obtained by training electric fingerprint data and a corresponding load label based on a LightGBM algorithm.
CN202211545503.8A 2022-12-05 2022-12-05 Edge side power fingerprint identification method and device based on machine learning Active CN115563511B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211545503.8A CN115563511B (en) 2022-12-05 2022-12-05 Edge side power fingerprint identification method and device based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211545503.8A CN115563511B (en) 2022-12-05 2022-12-05 Edge side power fingerprint identification method and device based on machine learning

Publications (2)

Publication Number Publication Date
CN115563511A CN115563511A (en) 2023-01-03
CN115563511B true CN115563511B (en) 2023-03-10

Family

ID=84770102

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211545503.8A Active CN115563511B (en) 2022-12-05 2022-12-05 Edge side power fingerprint identification method and device based on machine learning

Country Status (1)

Country Link
CN (1) CN115563511B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110544177A (en) * 2019-08-15 2019-12-06 广州水沐青华科技有限公司 Load identification method based on power fingerprint and computer readable storage medium
CN111368904A (en) * 2020-02-29 2020-07-03 华南理工大学 Electrical equipment identification method based on electric power fingerprint
CN112989133A (en) * 2021-03-29 2021-06-18 广州水沐青华科技有限公司 Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment
CN113191253A (en) * 2021-04-28 2021-07-30 浙江大学 Non-invasive load identification method based on feature fusion under edge machine learning
CN115001796A (en) * 2022-05-27 2022-09-02 浙江工业大学 Non-invasive load identification method based on character message queue

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2571119B (en) * 2018-02-18 2020-04-29 Faraday Predictive Ltd System and method for monitoring an operating condition of an electrical device when in operation
WO2022141330A1 (en) * 2020-12-31 2022-07-07 Guizhou Power Grid Company Limited A non-intrusive load identification method based on fingerprint characteristics of load power
CN114862488A (en) * 2021-02-04 2022-08-05 腾讯科技(深圳)有限公司 Identification method of resource consumption abnormal object and related device

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110544177A (en) * 2019-08-15 2019-12-06 广州水沐青华科技有限公司 Load identification method based on power fingerprint and computer readable storage medium
CN111368904A (en) * 2020-02-29 2020-07-03 华南理工大学 Electrical equipment identification method based on electric power fingerprint
CN112989133A (en) * 2021-03-29 2021-06-18 广州水沐青华科技有限公司 Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment
CN113191253A (en) * 2021-04-28 2021-07-30 浙江大学 Non-invasive load identification method based on feature fusion under edge machine learning
CN115001796A (en) * 2022-05-27 2022-09-02 浙江工业大学 Non-invasive load identification method based on character message queue

Also Published As

Publication number Publication date
CN115563511A (en) 2023-01-03

Similar Documents

Publication Publication Date Title
CN110956220B (en) Non-invasive household appliance load identification method
Heo et al. Toward load identification based on the Hilbert transform and sequence to sequence long short-term memory
CN111830347B (en) Two-stage non-invasive load monitoring method based on event
CN109975673B (en) Method for identifying fault arc at direct current side of photovoltaic microgrid
CN111209976B (en) Non-invasive load monitoring method and system
CN110119545B (en) Non-invasive power load identification method based on stack type self-encoder
Himeur et al. Efficient multi-descriptor fusion for non-intrusive appliance recognition
CN112580471A (en) Non-invasive load identification method based on AdaBoost feature extraction and RNN model
CN113361831B (en) Non-invasive load identification electric quantity decomposition method and system based on priority distribution
CN111242161A (en) Non-invasive non-resident user load identification method based on intelligent learning
Hamdi et al. A new approach of electrical appliance identification in residential buildings
CN113887912A (en) Non-invasive load identification method for deeply learning downward embedded equipment
CN115758188A (en) Non-invasive load identification method, device, equipment and medium
Akarslan et al. A novel approach based on a feature selection procedure for residential load identification
Luo et al. A generalizable method for practical non-intrusive load monitoring via metric-based meta-learning
CN111898694B (en) Non-invasive load identification method and device based on random tree classification
CN115563511B (en) Edge side power fingerprint identification method and device based on machine learning
Jiang et al. A new method for recognition and classification of power quality disturbances based on IAST and RF
Kumar et al. Data-driven electrical load disaggregation using graph signal processing
Mei et al. Random forest based adaptive non-intrusive load identification
CN115112989B (en) Non-invasive load monitoring method based on low-frequency data
Li et al. A training-free non-intrusive load monitoring approach for high-frequency measurements based on graph signal processing
CN116628620A (en) Non-invasive load identification calculation method
CN115828091A (en) Non-invasive load identification method and system based on end-cloud cooperation
Zhou et al. Feature extraction for non-intrusive load monitoring system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant