CN110991818B - Load identification method integrating event detection and neural network - Google Patents

Load identification method integrating event detection and neural network Download PDF

Info

Publication number
CN110991818B
CN110991818B CN201911115166.7A CN201911115166A CN110991818B CN 110991818 B CN110991818 B CN 110991818B CN 201911115166 A CN201911115166 A CN 201911115166A CN 110991818 B CN110991818 B CN 110991818B
Authority
CN
China
Prior art keywords
current
state
steady
data
neural network
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
CN201911115166.7A
Other languages
Chinese (zh)
Other versions
CN110991818A (en
Inventor
蒋雯倩
杨舟
李刚
韦杏秋
梁捷
李金瑾
陈珏羽
林秀清
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.)
Electric Power Research Institute of Guangxi Power Grid Co Ltd
Original Assignee
Electric Power Research Institute of Guangxi Power Grid 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 Electric Power Research Institute of Guangxi Power Grid Co Ltd filed Critical Electric Power Research Institute of Guangxi Power Grid Co Ltd
Priority to CN201911115166.7A priority Critical patent/CN110991818B/en
Publication of CN110991818A publication Critical patent/CN110991818A/en
Application granted granted Critical
Publication of CN110991818B publication Critical patent/CN110991818B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06QDATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/06Resources, workflows, human or project management, e.g. organising, planning, scheduling or allocating time, human or machine resources; Enterprise planning; Organisational models
    • G06Q10/063Operations research or analysis
    • G06Q10/0639Performance analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6256Obtaining sets of training patterns; Bootstrap methods, e.g. bagging, boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computer systems based on biological models
    • G06N3/02Computer systems based on biological models using neural network models
    • G06N3/04Architectures, e.g. interconnection topology
    • G06N3/0454Architectures, e.g. interconnection topology using a combination of multiple neural nets
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computer systems based on biological models
    • G06N3/02Computer systems based on biological models using neural network models
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06QDATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06Electricity, gas or water supply

Abstract

The invention discloses a load identification method integrating event detection and a neural network, which relates to the technical field of load identification, and is characterized in that historical operation data of a total load of a user is obtained, and time is marked; performing event detection on active power in the marked data to obtain a steady-state current waveform of each working state of each piece of electric equipment; randomly combining the obtained steady-state current waveform and the current waveform of the section without the electric appliance to obtain a multi-device current random combination; and extracting combined current characteristics from the random combination of the currents of the multiple devices as training samples, and outputting the state of each device through a trained neural network model. According to the invention, by combining the event method and the neural network mode identification method, the power consumption data of each piece of power consumption equipment of a user is not required to be measured in advance, a large amount of manual marking cost is saved, a large amount of samples can be provided for training of a neural network model only by marking short-time power consumption information, and the accuracy of the data can be ensured.

Description

Load identification method integrating event detection and neural network
Technical Field
The invention belongs to the technical field of load identification, and particularly relates to a load identification method integrating event detection and a neural network.
Background
As an important research direction in the field of smart power grids, the power load electricity consumption detail monitoring research has strong economic and social significance, and provides technical support and information guarantee for application of improving the interaction depth of power companies and users, realizing demand side response and the like. The technical means for realizing the power consumption detail monitoring of the power load can be classified into an invasive type and a non-invasive type. In a non-invasive mode, a sensor is not required to be arranged for each device by invading the interior of a user, and the power consumption information of each electric device switched in the total load can be obtained only based on detailed analysis of load electricity consumption total data measured at an inlet of a power supply. By virtue of the advantages of low cost, strong expansibility, convenience in maintenance, easiness in acceptance by users and the like, the non-invasive method becomes a mainstream technical means for realizing power load electricity consumption detail monitoring.
The non-invasive power load monitoring specifically comprises the step of completing load identification based on load electricity consumption total data, namely the problem can be treated as pattern identification, so that artificial intelligence algorithms such as neural networks and the like commonly used in the field of pattern identification are researched and applied to solve the problem. The neural network model has strong nonlinear fitting capability and has many advantages which cannot be compared with other algorithms in the load identification field, but the following difficulties and challenges still exist:
(1) to build a neural network model, a large amount of labeled sample data is often required for training and learning. Some researches need to collect the operation data of a single device, but in practical application, all the electric devices of a user cannot be measured in advance independently, and even if the user permits, a certain part of the electric devices cannot collect data in an plug-in manner; however, in some studies, only the load total data is used as a sample, but in order to ensure the generalization capability of the model, a large number of labeled samples are often required to be constructed, and a user needs to manually label a large number of data as training samples, which lacks the possibility of practical implementation.
(2) The neural network model trained for the single equipment is utilized to extract the operation characteristics of the single equipment in the actual non-invasive load identification process, often through event detection. Therefore, when event aliasing occurs, the load recognition effect is poor at this time.
Therefore, in order to solve the above problems, a non-invasive load recognition method that organically combines an event method and neural network pattern recognition with reduced labor labeling cost is needed.
Disclosure of Invention
The invention aims to provide a load identification method fusing event detection and a neural network, so as to overcome the defect that the load identification effect of non-invasive power load monitoring by adopting neural network training sample data is poor.
In order to achieve the above object, the present invention provides a load identification method integrating event detection and a neural network, comprising the following steps:
s1, acquiring historical operation data of the total load of the user, and marking time on the historical operation data of the total load of the user according to the state of the equipment;
s2, performing event detection on active power in the historical operation data of the total load of the user marked by the S1, determining transient state and steady state operation sections of the electric equipment, and acquiring a steady state current waveform of each working state of each electric equipment through a vector difference between steady state currents of adjacent steady state operation sections;
s3, randomly combining the marked equipment steady-state current waveform and the current waveform of the no-electric-appliance operation section obtained in the step S2 to obtain a multi-equipment current random combination;
s4, extracting combined current characteristics from the multi-device current random combination, namely performing fast Fourier transform on the combined current of the simulation scene, and obtaining the amplitude and phase angle of the harmonic current for multiple times;
and S5, taking the data obtained by the processing of S1-S4 as training samples, and outputting the state of each device through the trained neural network model.
Furthermore, the data types of the historical operation data of the total load of the user comprise original voltage waveform data, current waveform data and active power data.
Further, the S2 includes the following steps:
s21, carrying out event detection on active power in the historical operation data of the total load of the user marked by the S1, and dividing the historical data of the user marked by the user into a transient section and a steady-state section which are alternately distributed; the event detection method adopts a variable point detection method:
the active power of a user with marked historical data is represented by a formula (1), P is an active power sequence, and the P is detected point by point;
P=[P1,P2,P3,...Pi,...Pm] (1)
in the formula (1), PiThe total load active power value of the ith time point is represented, and m represents the total time length of the historical operating data of the total load of the user;
ΔPi=|Pi+1-Pi| (2)
when the absolute value of the difference between the active power value at a certain time point and the active power value at the later time point shown in formula (2) is greater than a certain threshold value, Δ PthIf the time point is regarded as a change point, the continuous occurrence frequency of the non-change point exceeds a certain threshold value tthThen, the section containing the non-variable point is regarded as a steady-state section, and the rest sections are event transient sections;
s22, selecting voltage and current with a certain time length in a steady-state section before and after each device is started to reconstruct voltage and current analog signals to obtain reconstructed voltage and current analog signals;
the voltage and current for a certain time are shown in formula (3):
in the formula (3), Vi,IiIs the instantaneous value of the voltage and the current at the ith sampling point in the selected time period, and n represents the total number of the sampling points contained in the selected time period in the steady-state section;
fitting the formula (3) by a cubic spline interpolation function V (t), I (t) and V (t), I (t) are functions defined on the interval [1, n ] and meet the limiting conditions of the cubic spline interpolation function;
s23, selecting a zero crossing point of the voltage signal as a starting point of the selected current waveform, namely, meeting the limiting condition of the formula (5); the zero-crossing point t of the first satisfied equation (5) in the voltage analog signal reconstructed from S2210At the beginning, the fourth zero-crossing point t satisfying equation (5)40Stopping, and taking the current analog signal waveform under the corresponding time scale as a steady-state current waveform;
s24, subtracting the steady-state current waveform of the steady-state section after the event is started from the steady-state current waveform before the event is started according to the steady-state current waveform obtained in the step S23 to obtain the steady-state current waveform difference of the single device;
and S25, repeating S21-S24 to obtain the steady-state current waveforms of all the electric equipment in each working state.
Further, a constraint of the cubic spline interpolation function;
a. a form of a cubic polynomial is satisfied over each interval [ i, i +1] (i ═ 1, 2.. n-1);
b.V (t), I (t) is continuously derivable of the second order over the interval [1, n ], i.e. satisfying formula (4):
in formula (4), k represents the order of derivation.
Further, before S24, the time length shown in formula (6) needs to be expanded and contracted for the current analog signal waveform, so as to complete the acquisition of the steady-state current waveform;
in the formula (6), Itr(t) an expression of a current analog signal after time-length expansion and contraction, f0Representing a standard frequency of voltage 50Hz, fsRepresenting the voltage, current sampling frequency.
Further, fsIs 3200 Hz.
Further, the current waveform of the section without electric appliance operation is a section without electric appliance operation taken from the user history data marked by S1, and the background current waveform is obtained through S22-S23 for the voltage and current signals of the section; each device current is then randomly combined with the background current waveform.
Further, in S4, the amplitude and phase angle of the first 5 th harmonic current are obtained.
Further, the neural network model is an RBF neural network.
And further comprising S6, collecting total load data of real-time operation of the user, and repeating S1-S5 to output the states of all the devices.
Compared with the prior art, the invention has the following beneficial effects: the invention provides a load identification method integrating event detection and a neural network, which comprises the steps of firstly, obtaining historical operation data of a total load of a user, and marking time on the historical operation data of the total load of the user according to the state of equipment; performing event detection on active power in the marked historical operating data of the total load of the user, determining transient and steady-state operating sections of the electric equipment, and acquiring a steady-state current waveform of each working state of each electric equipment through a vector difference between steady-state currents of adjacent steady-state operating sections; randomly combining the obtained marked equipment steady-state current waveform and the current waveform of the no-electric-appliance operation section to obtain a multi-equipment current random combination; extracting combined current characteristics from the random combination of the multiple devices of current, namely performing fast Fourier transform on the combined current of the analog scene, and obtaining the amplitude and the phase angle of the harmonic current for multiple times; and taking the data obtained by the previous processing as a training sample, and outputting the state of each device through the trained neural network model. The event method and the neural network pattern recognition method are fused, so that the power consumption data of each piece of user power equipment is not required to be measured in advance, a large amount of manual marking cost is saved, a large amount of samples can be provided for training of a neural network model only by marking short-time power consumption information, and further practical possibility of popularization and implementation is provided for application of the neural network pattern recognition method in the non-invasive load recognition field. Meanwhile, the total load electricity consumption information is used as an identification object, so that the adaptability of the method to complex scenes is improved.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only one embodiment of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
FIG. 1 is a flow chart of a method of load identification incorporating event detection and neural networks of the present invention;
FIG. 2 is a flow chart of S2 of the present invention;
FIG. 3 is a diagram illustrating the results of a user event detection over a period of time in accordance with an embodiment of the present invention;
FIG. 4 is a graph of a voltage digital signal and its spline interpolation function of the present invention;
FIG. 5 is a graph of a current digital signal and its spline interpolation function of the present invention;
FIG. 6 is a current waveform of the combined current of the two devices of the present invention;
FIG. 7 is a schematic diagram of an RBF neural network model of the present invention.
Detailed Description
The technical solutions in the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, the method for load identification by fusing event detection and neural network provided by the present invention includes the following steps:
and S1, acquiring historical operation data of the total load of the user within a period of time, and marking time on the historical operation data of the total load of the user according to the state of the equipment, wherein the historical operation data of the total load of the user at the marked time also comprises data of sections in which the electric appliances are operated, so that the operation moments of different electric appliances can be positioned in the historical operation data of the total load of the user. The data types of the historical operation data of the total load of the user comprise original voltage waveform data, current waveform data and active power data. Specifically, historical operating data of the total load of the user in one day can be obtained.
When the historical operating data of the total load of the user is acquired, the user is required to turn on and off (guarantee that the data are different) all the electric equipment in the family or perform all possible state conversion operations on the equipment within the period of acquiring the data, and mark the starting time and the time when the state conversion occurs of each equipment.
And S2, performing event detection on active power in the historical operation data of the total load of the user marked by the S1, determining transient state and steady state operation sections of the electric equipment, and acquiring a steady state current waveform of each working state of each electric equipment through a vector difference between steady state currents of adjacent steady state operation sections so as to obtain a steady state current waveform of a single equipment.
As shown in fig. 2, S2 includes the steps of:
and S21, carrying out event detection on active power in the historical operation data of the total load of the user marked by the S1, and dividing the historical data of the user marked by the S1 into a transient section and a steady-state section which are alternately distributed. The event detection method adopts a variable point detection method:
the active power of a user with marked historical data is represented by a formula (1), P is an active power sequence, and the P is detected point by point;
P=[P1,P2,P3,...Pi,...Pm] (1)
in the formula (1), PiRepresenting the total load at the ith time pointAnd the active power value m represents the total time length of the historical operating data of the total load of the user.
ΔPi=|Pi+1-Pi| (2)
When the absolute value of the difference between the active power value at a certain time point and the active power value at the later time point shown in formula (2) is greater than a certain threshold value, Δ PthThe time point is regarded as a change point, and when the continuous occurrence frequency of the non-change point exceeds a certain threshold value tthThen, the section containing the non-change point is regarded as the steady-state section, and the remaining section is the event transient section.
Exemplified by a user's active power data for a segment of a day, Δ PthIs 30W, tthThe result of the test at S21 is shown in fig. 3, where the curve represents the active power curve, the time between two adjacent points is 1 second, and the circle point mark represents the detected event.
And S22, selecting voltage and current with a certain time length in a steady-state section before and after each device is started to reconstruct the voltage and current analog signals, and obtaining reconstructed voltage and current analog signals.
In the formula (3), Vi,IiIs the instantaneous value of the voltage and current at the ith sampling point in the selected time period, and n represents the total number of sampling points contained in the selected time period in the steady-state section.
Considering that actually acquired voltage and current waveform signals are digital discrete, as shown in formula (3), when current waveform data is subjected to addition and subtraction combination, data points of different current waveforms have to be ensured to correspond one to one under the same voltage phase reference, so that the requirement on the sampling frequency of a digital signal is very high, and the sampling frequency is often very low in practical application, so that analog signal reconstruction needs to be considered.
Therefore, the present embodiment utilizes a cubic spline interpolation function v (t), i (t) to fit the discrete points of the voltage and current digital signals; and V (t), I (t) is a function defined over the interval [1, n ] satisfying the following condition:
a. a form of a cubic polynomial is satisfied over each interval [ i, i +1] (i ═ 1, 2.. n-1).
b.V (t), I (t) is continuously derivable of the second order over the interval [1, n ], i.e. satisfying formula (4):
in formula (4), k represents the order of derivation.
A segment of sampling frequency f obtained in the steady-state sectionsFig. 4 and 5 show digital signals of voltage and current with 3200Hz and 256 sampling points n, and a spline function curve in the graph can be obtained according to S22, that is, the reconstruction of analog signals of voltage and current is completed as shown in fig. 4 and 5.
S23, in order to ensure that the current waveforms of adjacent steady-state sections are performed under the same voltage reference during subtraction, the starting point of the current waveform for waveform subtraction must be reasonably determined, and the zero crossing point of the voltage signal is selected as the selected starting point of the current waveform, namely the limiting condition of the formula (5) is met;
the zero-crossing point t of the first satisfied equation (5) in the voltage analog signal reconstructed from S2210At the beginning, the fourth zero-crossing point t satisfying equation (5)40And stopping, and taking the current analog signal waveform under the corresponding time scale as a steady-state current waveform.
Meanwhile, the actual voltage frequency may fluctuate around 50Hz, and therefore, the time length shown in formula (6) needs to be expanded and contracted for the current analog signal waveform, so as to complete the steady-state current waveform acquisition.
In the formula (6), Itr(t) watchShowing the current analog signal expression, f, after time length scaling0Representing a standard frequency of voltage 50Hz, fsThe sampling frequency of voltage and current is represented, and 3200Hz is selected in the embodiment.
S24, for each start event (corresponding to the transient process of the start of the electrical appliance), after each event is started, there may be multiple steady-state segments corresponding to different working states of the corresponding electrical appliance, or there may be only one steady-state segment (depending on the number of states included in the electrical appliance), so according to the steady-state current waveform obtained in step S23, the steady-state current waveform in the steady-state segment after the event is started is subtracted from the steady-state current waveform before the event is started, so as to obtain the steady-state current waveform difference of the single device.
Here, by the processing of S21 to S23, the acquired steady-state current waveform has been processed into an analog signal of uniform length, and thus can be directly subtracted.
And S25, because each starting event corresponds to the starting transient process of the user electric equipment, repeating S21-S24 according to the marked historical operating data of the user to acquire the steady-state current waveforms of all the electric equipment in each working state.
S3, randomly combining the marked equipment steady-state current waveform and the current waveform of the no-electric-appliance operation section obtained in the step S2 (waveform discrete values are directly added), and obtaining a multi-equipment current random combination; and then, simulating the scenes of simultaneous operation of various electrical appliances of a user according to the random combination of the currents of the multiple devices, wherein the number of the scenes is determined according to the number of the devices and the number of the steady-state current waveforms of each device, and the scenes are increased in an exponential level manner, and the combined current of the scenes represents the total load current generated by the simulated scenes.
Wherein, the current waveform of the section without electric appliance operation is a section without electric appliance operation taken from the user historical data marked by S1, and the voltage and current signals of the section without electric appliance operation are processed by S22-S23 to obtain the background current waveform; in order to simulate the situation of the total load current of a user in actual operation, the current of each device and the waveform of the background current are randomly combined. Taking 2 device combinations as an example (shown in fig. 6), the number of combinations is shown in formula (7):
in the formula (7), num _ com is the number of combinations, num _ dev is the number of all the electric devices of the user, l is the number of steady-state current waveforms obtained in each working state of each electric device, nsiThe status number of the ith electric equipment is shown.
S4, extracting combined current characteristics from the random combination of the multiple devices of current, namely, carrying out fast Fourier transform on the combined current of the simulation scene, and obtaining the amplitude and the phase angle of the first 5 harmonic current.
Specifically, discrete point acquisition is carried out on the obtained combined current waveform of each simulated scene, and the distance between two points is 1/fsAnd then, performing fast Fourier transform on the harmonic current, and taking the amplitude and phase angle of the first 5 harmonic currents to obtain a data format as shown in a formula 8:
in formula (8), IciThe magnitude and phase angle of the first 5 harmonic currents of the combined current in the ith simulation scenario are respectively represented, i being 1,2, 3.
And S5, taking the data obtained by the processing of S1-S4 as training samples, and outputting the state of each device through the trained neural network model. I.e. the amplitude and phase angle of the first 5 harmonic currents of the combined current as input, and the actual labeled state of the respective device as output. The invention selects RBF neural network.
As shown in fig. 7, training samples of the neural network model need to be determined, the simulation scenarios obtained through S1-S3 are used as the training samples, one simulation scenario is a training sample, and the simulation scenario obtained through S3 is normalized through S4 to obtain a final training sample. That is, the data obtained in S4 is used as a training sample, and if the input of the training sample has 10 variables in total, as shown in equation (8);
before model training, fixed value normalization processing is required, and for the amplitude of harmonic current, the maximum value of the harmonic current in a sample floats upwards by 1A to serve as a normalization fixed value; for the harmonic phase angle, selecting pi as a normalized fixed value, namely as shown in an equation (9);
in the formula (9), XiRepresents the input of the i-th training sample, which contains elements of magnitude and phase angle of the first 5 harmonic currents of the normalized combined current.
The sample output is the working state quantity s of all the electric equipment in the useriE {0,1}, i 1,2,3And (4) respectively. When s isiTaking 1 indicates that the ith operating state among all operating states of all devices is in an operating state. Then the input and output of the RBF neural network model can be represented by equation (10):
in the formula (10), sijState quantity, w, representing the jth appliance state in the output of the ith training samplekjRepresents the weight from the kth hidden layer neuron to the jth output neuron, q is the number of hidden layer neurons contained, ρ (X)i,ck) For the radial basis function, the present invention selects a commonly used gaussian radial basis function, as shown in formula (11). c. CkThe center corresponding to the kth hidden layer neuron.
Wherein σkFor the width of the kth hidden layer neuron, it determines the width of the radial basis function around the center point。
For the determination of the specific parameters of the RBF neural network model, the invention utilizes the negative gradient training principle to automatically adjust the data center and the width of each neuron and the weight between neurons in each layer.
And S6, collecting the total load data of the real-time operation of the user, and repeating S1-S5 to output the states of all the devices.
The operation method of the load identification method combining event detection and neural network of the present invention will be described in detail to make those skilled in the art more aware of the present invention:
taking a user for example, the user contains 8 kinds of electric equipment in total, and relates to 11 kinds of equipment operating states in total, and the number of hidden layer neurons of the RBF neural network model is set to q-10. When single device steady-state current acquisition is performed, the number of steady-state current waveforms l is 5 per device per state acquisition. And combining the number of the training samples num _ com which is 31850496, and selecting a proper number of samples from the training samples to train parameters of the RBF neural network model so as to save training time. The trained neural network is used for identifying the states of the electric equipment of the user under different scenes, and the following results are obtained in the following table 1:
TABLE 1 load recognition Effect under multiple scenarios for the user
Note: TP is true positive, FP is false positive, FN is false negative
In summary, the load identification method integrating the event detection and the neural network, disclosed by the invention, integrates the event method and the neural network pattern identification method, so that the power consumption data of each piece of electric equipment of a user is not required to be measured in advance, a large amount of manual marking cost is saved, a large amount of samples can be provided for the training of a neural network model only by marking short-time power consumption information, and further, the practical possibility of popularization and implementation is provided for the application of the neural network pattern identification method in the non-invasive load identification field. Meanwhile, the total load electricity consumption information is used as an identification object, so that the adaptability of the method to complex scenes is improved. Through verification, the load identification effect of the method has higher accuracy, and a technical basis can be provided for a series of advanced applications of the non-invasive load identification technology.
The above disclosure is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of changes or modifications within the technical scope of the present invention, and shall be covered by the scope of the present invention.

Claims (6)

1. A load identification method integrating event detection and a neural network is characterized in that: the method comprises the following steps:
s1, acquiring historical operation data of the total load of the user, and marking time on the historical operation data of the total load of the user according to the state of the equipment;
s2, performing event detection on active power in the historical operation data of the total load of the user marked by the S1, determining transient state and steady state operation sections of the electric equipment, and acquiring a steady state current waveform of each working state of each electric equipment through a vector difference between steady state currents of adjacent steady state operation sections;
the S2 includes the steps of:
s21, carrying out event detection on active power in the historical operation data of the total load of the user marked by the S1, and dividing the historical data of the user marked by the user into a transient section and a steady-state section which are alternately distributed; the event detection method adopts a variable point detection method:
the active power of a user with marked historical data is represented by a formula (1), P is an active power sequence, and the P is detected point by point;
P=[P1,P2,P3,...Pi,...Pm] (1)
in the formula (1), PiIs shown asThe total load active power values of the i time points, and m represents the total time length of the historical operating data of the total load of the user;
ΔPi=|Pi+1-Pi| (2)
when the absolute value of the difference between the active power value at a certain time point and the active power value at the later time point shown in formula (2) is greater than a certain threshold value, Δ PthIf the time point is regarded as a change point, the continuous occurrence frequency of the non-change point exceeds a certain threshold value tthThen, the section containing the non-variable point is regarded as a steady-state section, and the rest sections are event transient sections;
s22, selecting voltage and current with a certain time length in a steady-state section before and after each device is started to reconstruct voltage and current analog signals to obtain reconstructed voltage and current analog signals;
the voltage and current for a certain time are shown in formula (3):
in the formula (3), Vi,IiIs the instantaneous value of the voltage and the current at the ith sampling point in the selected time period, and n represents the total number of the sampling points contained in the selected time period in the steady-state section;
fitting the formula (3) by a cubic spline interpolation function V (t), I (t) and V (t), I (t) are functions defined on the interval [1, n ] and meet the limiting conditions of the cubic spline interpolation function;
s23, selecting a zero crossing point of the voltage signal as a starting point of the selected current waveform, namely, meeting the limiting condition of the formula (5); the zero-crossing point t of the first satisfied equation (5) in the voltage analog signal reconstructed from S2210At the beginning, the fourth zero-crossing point t satisfying equation (5)40Stopping, and taking the current analog signal waveform under the corresponding time scale as a steady-state current waveform;
s24, subtracting the steady-state current waveform of the steady-state section after the event is started from the steady-state current waveform before the event is started according to the steady-state current waveform obtained in the step S23 to obtain the steady-state current waveform difference of the single device;
before S24, the current analog signal waveform needs to be subjected to time length expansion as shown in formula (6), so that the steady-state current waveform acquisition is completed;
in the formula (6), Itr(t) an expression of a current analog signal after time-length expansion and contraction, f0Representing a standard frequency of voltage 50Hz, fsRepresenting the sampling frequency of voltage and current;
s25, repeating S21-S24 to obtain the steady-state current waveforms of all the electric equipment in each working state;
s3, randomly combining the marked equipment steady-state current waveform and the current waveform of the no-electric-appliance operation section obtained in the step S2 to obtain a multi-equipment current random combination;
the current waveform of the no-electric appliance operation section is a section without electric appliance operation taken from user historical data marked by S1, and a background current waveform is obtained through S22-S23 for voltage and current signals of the section; then randomly combining the current of each device with the background current waveform;
s4, extracting combined current characteristics from the multi-device current random combination, namely performing fast Fourier transform on the combined current of the simulation scene, and obtaining the amplitude and phase angle of the harmonic current for multiple times;
s5, taking the data obtained by the processing of S1-S4 as training samples, and outputting the state of each device through a trained neural network model;
and S6, collecting the total load data of the real-time operation of the user, and repeating S1-S5 to output the states of all the devices.
2. The method of fusing event detection and load recognition of a neural network as claimed in claim 1, wherein: the data types of the historical operation data of the total load of the user comprise original voltage waveform data, current waveform data and active power data.
3. The method of fusing event detection and load recognition of a neural network as claimed in claim 1, wherein: a constraint of the cubic spline interpolation function;
a. a form of a cubic polynomial is satisfied over each interval [ i, i +1] (i ═ 1, 2.. n-1);
b.V (t), I (t) is continuously derivable of the second order over the interval [1, n ], i.e. satisfying formula (4):
in formula (4), k represents the order of derivation.
4. The method of fusing event detection and load recognition of a neural network as claimed in claim 1, wherein: f. ofsIs 3200 Hz.
5. The method of fusing event detection and load recognition of a neural network as claimed in claim 1, wherein: in S4, the amplitude and phase angle of the first 5 th harmonic current are obtained.
6. The method of fusing event detection and load recognition of a neural network as claimed in claim 1, wherein: the neural network model is an RBF neural network.
CN201911115166.7A 2019-11-14 2019-11-14 Load identification method integrating event detection and neural network Active CN110991818B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911115166.7A CN110991818B (en) 2019-11-14 2019-11-14 Load identification method integrating event detection and neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911115166.7A CN110991818B (en) 2019-11-14 2019-11-14 Load identification method integrating event detection and neural network

Publications (2)

Publication Number Publication Date
CN110991818A CN110991818A (en) 2020-04-10
CN110991818B true CN110991818B (en) 2020-12-18

Family

ID=70084602

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911115166.7A Active CN110991818B (en) 2019-11-14 2019-11-14 Load identification method integrating event detection and neural network

Country Status (1)

Country Link
CN (1) CN110991818B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120290230A1 (en) * 2009-07-01 2012-11-15 Carnegie Mellon University Methods and Apparatuses for Monitoring Energy Consumption and Related Operations
CN106096726A (en) * 2016-05-31 2016-11-09 华北电力大学 A kind of non-intrusion type load monitoring method and device
CN108616120A (en) * 2018-04-28 2018-10-02 西安理工大学 A kind of non-intrusive electrical load decomposition method based on RBF neural

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120290230A1 (en) * 2009-07-01 2012-11-15 Carnegie Mellon University Methods and Apparatuses for Monitoring Energy Consumption and Related Operations
CN106096726A (en) * 2016-05-31 2016-11-09 华北电力大学 A kind of non-intrusion type load monitoring method and device
CN108616120A (en) * 2018-04-28 2018-10-02 西安理工大学 A kind of non-intrusive electrical load decomposition method based on RBF neural

Also Published As

Publication number Publication date
CN110991818A (en) 2020-04-10

Similar Documents

Publication Publication Date Title
Xu et al. Real-time transient stability assessment model using extreme learning machine
CN103268519B (en) Based on the power-system short-term load forecasting method and the device that improve Lyapunov index
Mazumdar et al. Neural network based method for predicting nonlinear load harmonics
Valtierra-Rodriguez et al. A new methodology for tracking and instantaneous characterization of voltage variations
Sirojan et al. Deep neural network based energy disaggregation
CN105590144A (en) Wind speed prediction method and apparatus based on NARX neural network
KR20180058113A (en) System and Method for household appliance classification using extracting eigen value of household appliance
CN110991818B (en) Load identification method integrating event detection and neural network
Vanfretti et al. A phasor measurement unit based fast real-time oscillation detection application for monitoring wind-farm-to-grid sub–synchronous dynamics
Dash et al. Power quality monitoring using an integrated Fourier linear combiner and fuzzy expert system
Stošović et al. Classification of nonlinear loads based on artificial neural networks
Papadopoulos et al. Online parameter identification and generic modeling derivation of a dynamic load model in distribution grids
Saraf et al. Online oscillation monitoring of synchronous generators using Parallel-Prony analysis
CN108152629B (en) Method for drawing voltage tolerance curve of thermal power plant
Arshad et al. Wind power prediction using genetic programming based ensemble of artificial neural networks (GPeANN)
CN104330742B (en) A kind of series mode interference suppressing method and device
JP6417152B2 (en) Device power estimation method and system
Park et al. A novel method for assessing the contribution of harmonic sources to voltage distortion in power systems
CN110501568B (en) Non-invasive equipment load monitoring method based on graph signal processing
CN106226700A (en) Lithium battery detection platform system and method for testing
El-Sharkh et al. Forecasting electricity demand using dynamic artificial neural network model
Kwater et al. Application of artificial neural networks in non-invasive identification of electric energy receivers
CN113363979A (en) Non-contact power load decomposition sensing method and system
Wang et al. Rolling bearing fault diagnosis based on wavelet packet-neural network characteristic entropy
Esfahani et al. A New Dynamic Intelligent Time Domain Arc Furnace Modeling based on Combination Adaptive Neuro-fuzzy Inference System and Chain Code

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