CN113808071A - Non-invasive load monitoring method and system based on deep learning - Google Patents

Non-invasive load monitoring method and system based on deep learning Download PDF

Info

Publication number
CN113808071A
CN113808071A CN202110860151.4A CN202110860151A CN113808071A CN 113808071 A CN113808071 A CN 113808071A CN 202110860151 A CN202110860151 A CN 202110860151A CN 113808071 A CN113808071 A CN 113808071A
Authority
CN
China
Prior art keywords
load monitoring
data
monitored
invasive load
model
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.)
Pending
Application number
CN202110860151.4A
Other languages
Chinese (zh)
Inventor
周开乐
殷辉
丁涛
李兰兰
周昆树
胡定定
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.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
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 Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202110860151.4A priority Critical patent/CN113808071A/en
Publication of CN113808071A publication Critical patent/CN113808071A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformation in the plane of the image
    • G06T3/40—Scaling the whole image or part thereof
    • G06T3/4046—Scaling the whole image or part thereof using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image

Abstract

The invention provides a non-invasive load monitoring method and system based on deep learning, and relates to the technical field of power load monitoring. According to the method, the one-dimensional power data are converted into the two-dimensional image data to be monitored, the time domain information is incorporated into the original data, the data contain more effective information, meanwhile, a non-invasive load monitoring model based on deep learning is used, the model can consider the time dependence in the data, the accuracy of a prediction model is improved, and meanwhile, the appearance of gradient disappearance and gradient explosion in a CNN network can be avoided.

Description

Non-invasive load monitoring method and system based on deep learning
Technical Field
The invention relates to the technical field of power load monitoring, in particular to a non-invasive load monitoring method and system based on deep learning.
Background
With the rapid growth of population and the rapid development of economy, the number and types of electric equipment of users are rapidly increased, and more attention is paid to analysis algorithms which can be used by different users. The non-invasive load monitoring is to analyze and mine the running states of each electric appliance counted in the general table, including the start-stop time, the service cycle and the like, only through data in a single general table without additionally installing a sub-table in each electric appliance. The non-invasive load monitoring can provide detailed energy bills and personalized energy-saving suggestions for users, monitor fault equipment, assist in customer subdivision and enhance micro-grid demand side management.
Traditional monitoring methods, such as markov models, SVMs, LGs, decision trees, and the like, cannot simulate the nonlinear relationship between the electrical load and the related variables, and when there are many electrical devices, these linear classifiers are often poor in effect. Although the ANN can simulate the nonlinear relation, the ANN cannot simulate the space-time law of the electrical characteristics of the electrical appliance, so that a good monitoring result cannot be obtained.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a non-invasive load monitoring method and a non-invasive load monitoring system based on deep learning, and solves the technical problem that the existing non-invasive load monitoring can not simulate the time-space law of the electricity utilization characteristics of an electric appliance.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
in a first aspect, the present invention provides a deep learning-based non-intrusive load monitoring method, including the following steps:
s1, performing time window decomposition on the one-dimensional power data to be monitored collected in the summary table to obtain k pieces of one-dimensional power data to be monitored;
s2, performing image transformation on k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, scaling the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and performing zero-value normalization processing;
and S3, inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model, and obtaining the working state of each power load in each time period.
Preferably, the construction process of the pre-trained CNN-based non-invasive load monitoring model includes:
a1, performing time window decomposition and image conversion on power data acquired by a table in a preset historical time period to obtain two-dimensional image data in a modeling process;
a2, preprocessing the two-dimensional image data in the modeling process to obtain standard two-dimensional image data;
a3, performing label processing and segmentation processing on the standard two-dimensional image data to obtain a training set and a test set;
a4, obtaining an initial non-invasive load monitoring model based on the training set and the CNN model, and performing gradient test;
a5, testing the initial non-invasive load monitoring model based on the test set to obtain a test result, and obtaining an evaluation result based on the test result;
a6, optimizing the initial non-invasive load monitoring model according to the evaluation result to obtain a non-invasive load monitoring model.
Preferably, the image transformation comprises a gram angle field.
Preferably, the a3 specifically comprises:
corresponding to different electric appliances and collected d1Data of days, wherein
Figure BDA0003185409600000031
As a training set train, the training set train,
Figure BDA0003185409600000032
as test set test; aiming at training settrain, labeling according to the type of the electric appliance to obtain a label set label; the label type rule is consistent with the number of the electric appliances, namely labels of 1, 2 … i electric appliances are 1, 2 … i;
Figure BDA0003185409600000033
Figure BDA0003185409600000034
label＝[1,2,...i]T
wherein: each row in the training set train and the test set test represents one input variable, for a total of i variables.
Preferably, the a4 specifically comprises:
setting the number of layers of convolution layers in the CNN model and the size of convolution kernels of the layers, the number of neurons of all layers of all connection layers, regularization parameters, dropout parameters, batch size and learning rate size, and selecting Adam for gradient inspection; inputting a training set train and training to obtain an initial non-invasive load monitoring model;
the gradient test was performed using the following formula:
Figure BDA0003185409600000041
Figure BDA0003185409600000042
wherein: j (theta)i) Is a loss function with regularization parameter, ε 10-7And d theta is a derivative value calculated when the initial non-invasive load monitoring model propagates reversely.
Preferably, the a6 specifically comprises:
inputting the test set into an initial non-invasive load monitoring model to obtain a test result; evaluating the test result by using the root mean square error, the average absolute error and the average absolute error percentage according to the test result and the type of the electric appliances actually corresponding to the test set to obtain an evaluation result; the calculation formulas of the root mean square error, the average absolute error and the average absolute error percentage are as follows:
Figure BDA0003185409600000043
Figure BDA0003185409600000044
Figure BDA0003185409600000045
wherein: y isp,iObtaining a test result for each group of input data of the initial non-invasive load monitoring model, namely the type of the electric appliance predicted by the model; and y isr,iThe type of the electric appliance actually corresponding to the input data; n is the total number of appliance types in each set of input data.
In a second aspect, the present invention provides a deep learning based non-intrusive load monitoring system, the system comprising:
the data decomposition and conversion module is used for carrying out time window decomposition on the one-dimensional power data to be monitored, which are collected in the summary table, so as to obtain k pieces of one-dimensional power data to be monitored;
the data processing module is used for carrying out image conversion on the k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, scaling the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and carrying out zero value normalization processing;
and the non-invasive load monitoring and solving module is used for inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model to obtain the working state of each power load in each time interval.
Preferably, the data processing module is further configured to perform time window decomposition and image conversion on the power data of the preset historical time period acquired by the table, so as to obtain two-dimensional image data in the modeling process;
the data processing module is also used for carrying out unified scaling and normalization processing on the two-dimensional image data, dividing the processed image data into a training set and a testing set in the module, and carrying out labeling according to the type of the electric appliances.
Preferably, the pre-trained non-invasive load monitoring model based on the CNN is constructed by a model generation and optimization module;
the model generation and optimization module is used for inputting the generated data set into the CNN model, training and testing the CNN model to obtain an initial non-invasive load monitoring model, and performing gradient inspection; testing the initial non-invasive load monitoring model based on a test set to obtain a test result, and obtaining an evaluation result based on the test result; and optimizing the initial non-invasive load monitoring model according to the evaluation result to obtain the non-invasive load monitoring model.
In a third aspect, the present invention provides a computer readable storage medium storing a computer program for deep learning based non-intrusive load monitoring, wherein the computer program causes a computer to perform the deep learning based non-intrusive load monitoring method as described above.
(III) advantageous effects
The invention provides a non-intrusive load monitoring method and system based on deep learning. Compared with the prior art, the method has the following beneficial effects:
the method comprises the steps of carrying out time window decomposition on one-dimensional power data to be monitored collected in a summary table to obtain k pieces of one-dimensional power data to be monitored; performing image conversion on k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, scaling the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and performing zero value normalization processing; and inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model to obtain the working state of each power load in each time period. According to the method, the one-dimensional power data are converted into the two-dimensional image data to be monitored, the time domain information is incorporated into the original data, the data contain more effective information, meanwhile, a non-invasive load monitoring model based on deep learning is used, the model can consider the time dependence in the data, the accuracy of a prediction model is improved, and meanwhile, the appearance of gradient disappearance and gradient explosion in a CNN network can be avoided.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a block diagram of a deep learning-based non-intrusive load monitoring method according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a non-intrusive load monitoring method based on deep learning according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described, and it is obvious that the described embodiments are a part of the embodiments of the present invention, but 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.
The embodiment of the application provides a non-invasive load monitoring method and system based on deep learning, solves the technical problem that the existing non-invasive load monitoring cannot simulate the time-space law of the electricity utilization characteristics of an electric appliance, realizes the time dependence in data considered during monitoring, and improves the monitoring precision.
In order to solve the technical problems, the general idea of the embodiment of the application is as follows:
traditional monitoring methods, such as markov models, SVMs, LGs, decision trees, and the like, cannot simulate the nonlinear relationship between the electrical load and the related variables, and when there are many electrical devices, these linear classifiers are often poor in effect. Although the ANN can simulate the nonlinear relation, the ANN cannot simulate the space-time law of the electrical characteristics of the electrical appliance, so that a good monitoring result cannot be obtained. General RNNs, although capable of modeling the time dependence in time series, face the problems of gradient explosion and gradient disappearance.
Also, most of these methods require high frequency data, such as voltage, current, harmonics, phase angle, etc., that can reflect the operating characteristics of the appliance. The data cannot be directly utilized in subsequent demand side service, accurate energy utilization bills cannot be effectively provided for various users, or demand response regulation and control are carried out, and the value of non-invasive load monitoring cannot be exerted.
The embodiment of the invention aims to provide a non-intrusive load monitoring method and system based on deep learning. Firstly, time window decomposition is carried out on time sequence data of a single electric appliance, then the one-dimensional data is converted into a two-dimensional image through an imaging conversion method, and time domain characteristics in the data are stored. These image data are then preprocessed, and the processed data is input as a CNN model. The network model is then trained and tested and evaluated. And (3) performing time window decomposition on the data in the general table, then performing load monitoring based on the trained CNN model to obtain the start-stop time and the operation cycle data of each electric appliance, and finally obtaining the operation time and power consumption data of the electric appliances in each time period.
In order to better understand the technical solution, the technical solution will be described in detail with reference to the drawings and the specific embodiments.
The embodiment of the invention provides a non-intrusive load monitoring method based on deep learning, and as shown in fig. 1, the method comprises steps S1-S3;
s1, performing time window decomposition on the one-dimensional power data to be monitored collected in the summary table to obtain k pieces of one-dimensional power data to be monitored;
s2, performing image transformation on k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, scaling the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and performing zero-value normalization processing;
and S3, inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model, and obtaining the working state of each power load in each time period.
According to the embodiment of the invention, one-dimensional power data is converted into two-dimensional image data to be monitored, time domain information is incorporated into original data, so that the data contains more effective information, and meanwhile, a CNN-based non-invasive load monitoring model is used, so that the model can consider the time dependence in the data, improve the accuracy of a prediction model, and simultaneously avoid the appearance of gradient disappearance and gradient explosion in a CNN network.
The following describes the individual steps in detail, as shown in FIG. 2:
in step S1, time window decomposition is performed on the one-dimensional power data to be monitored collected in the summary table, so as to obtain k pieces of one-dimensional power data to be monitored. The specific implementation process is as follows:
and dividing the one-dimensional power data to be monitored collected in the summary table according to the same time window lambda and the moving step length s to obtain k pieces of one-dimensional power data to be monitored.
In step S2, image transformation is performed on the k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, the k two-dimensional images to be monitored are scaled into two-dimensional image data to be monitored with the same size, and zero value normalization processing is performed. The specific implementation process is as follows:
and converting the k pieces of segmented one-dimensional power data to be monitored by using a GAF method, then scaling the data into 255 x 255 two-dimensional image data to be monitored, and performing zero-value normalization processing.
Figure BDA0003185409600000091
Wherein: rkRepresenting the scaled image, Rk_meanRepresents RkMean value of Rk_varRepresents PkVariance of R'kThe two-dimensional image data to be monitored after standardization.
In step S3, the normalized two-dimensional image data to be monitored is input into a pre-trained CNN-based non-invasive load monitoring model, and the operating state of each power load in each time interval is obtained. The specific implementation process is as follows:
and inputting the standardized two-dimensional image data to be monitored into a pre-trained non-invasive load monitoring model based on the CNN, and analyzing the switching time and the power consumed in the operation period of each electrical appliance.
In the embodiment of the invention, the construction process of the pre-trained non-intrusive load monitoring model based on the CNN is as follows:
a1, performing time window decomposition and image conversion on the power data of the preset historical time period acquired by the table to obtain two-dimensional image data in the modeling process. The method specifically comprises the following steps:
assuming that the intelligent electric meter collects power data every 5 seconds, 17280 pieces of power data are obtained by each electric appliance every day, the data are divided according to a fixed time window, the size of the time window is set to be lambda, the lambda is an integral multiple of 5 seconds, the moving step length is s, and the s is set to be the minimum collection period (5 seconds) of the intelligent electric meter. The collected data of the single electric appliance in one day can be divided into j time sequence data. After subjecting these one-dimensional data to Graham Angular Field (GAF), j pieces of image data were obtained, where j has the following relationship with λ and s.
Figure BDA0003185409600000101
a2, preprocessing the two-dimensional image data in the modeling process to obtain standard two-dimensional image data. The method specifically comprises the following steps:
after the one-dimensional data is converted into two-dimensional image data, preprocessing of the image is required, and the preprocessing includes uniformly scaling the image into a square image of 255 × 255.
In addition, the images need to be subjected to zero-mean normalization processing.
Figure BDA0003185409600000111
Wherein: pi,jRepresenting the scaled image, Pi,j_meanRepresents Pi,jMean value of Pi,j_varRepresents Pi,jVariance of P'i,jFor the normalized two-dimensional image data, i represents the type of the electric appliance.
and a3, performing label processing and segmentation processing on the standard two-dimensional image data to obtain a training set and a test set. The method specifically comprises the following steps:
corresponding to different electric appliances and collected d1Data of days, wherein
Figure BDA0003185409600000112
As a training set train, the training set train,
Figure BDA0003185409600000113
as test set test. In addition, the training set train needs to be labeled manually, and labeling is performed according to the types of the electric appliances to obtain a label set label, which indicates that the label corresponding to each data is a specific electric appliance type. The label type rule is consistent with the number of the electric appliances, namely the labels of 1, 2 … i electric appliances are 1, 2 … i.
Figure BDA0003185409600000114
Figure BDA0003185409600000115
label＝[1,2,...i]T
Wherein: each row in the training set train and the test set test represents one input variable, for a total of i variables.
a4, obtaining an initial non-invasive load monitoring model based on the training set and the CNN model, and carrying out gradient test. The method specifically comprises the following steps:
firstly, the number of layers of convolution layers in a CNN model and the size of convolution kernels of the layers, the number of neurons of each layer of a full connection layer, a regularization parameter, a dropout parameter, a batch size and a learning rate size are set, and an Adam gradient descent algorithm is selected to optimize a network. And inputting a training set train and training to obtain an initial non-invasive load monitoring model.
The gradient test was performed using the following formula:
Figure BDA0003185409600000121
Figure BDA0003185409600000122
wherein: j (theta)i) Is a loss function with regularization parameters, ε is a very small constant, here designated 10-7And d theta is a derivative value calculated when the non-invasive load monitoring model (the trained CNN model) is propagated reversely. When c < epsilon indicates that the gradient descent method is correctly executed, the parameters of the non-intrusive load monitoring model are correctly learned.
a5, testing the initial non-invasive load monitoring model based on the test set to obtain a test result, and obtaining an evaluation result based on the test result. The method specifically comprises the following steps:
and inputting the test set into an initial non-invasive load monitoring model to obtain a test result. And simultaneously, according to the test result and the type of the electric appliance actually corresponding to the input data, evaluating the test result by using the root mean square error, the average absolute error and the average absolute error percentage to obtain an evaluation result, wherein the smaller the value of the evaluation result is, the higher the prediction precision is, and the calculation modes of the three evaluation indexes are as follows:
Figure BDA0003185409600000123
Figure BDA0003185409600000124
Figure BDA0003185409600000125
wherein: y isp,iObtaining a test result for each group of input data of the initial non-invasive load monitoring model, namely the type of the electric appliance predicted by the model; and y isr,iThe type of the electric appliance actually corresponding to the input data; n is the total number of appliance types in each set of input data.
a6, optimizing the initial non-invasive load monitoring model according to the evaluation result to obtain the non-invasive load monitoring model. The method specifically comprises the following steps:
and (d) adjusting parameters in the initial non-invasive load monitoring model obtained in the step a5 based on the evaluation result of the model, and dynamically optimizing the non-invasive load monitoring model until the non-invasive load monitoring model with the highest precision is obtained.
The embodiment of the invention also provides a non-invasive load monitoring system based on deep learning, which comprises:
the data decomposition and conversion module is used for carrying out time window decomposition on the one-dimensional power data to be monitored, which are collected in the summary table, so as to obtain k pieces of one-dimensional power data to be monitored;
the data processing module is used for carrying out image conversion on k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, zooming the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and carrying out zero value normalization processing;
and the non-invasive load monitoring and solving module is used for inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model to obtain the working state of each power load in each time interval.
The data processing module is further used for carrying out time window decomposition and image conversion on the power data of a preset historical time period acquired by the meter to obtain two-dimensional image data in the modeling process;
the data processing module is also used for carrying out unified scaling and normalization processing on the two-dimensional image data, dividing the processed image data into a training set and a testing set in the module, and carrying out labeling according to the type of the electric appliances.
The pre-trained CNN-based non-invasive load monitoring model is constructed through a model generation and optimization module.
The model generation and optimization module is used for inputting the generated data set into the CNN model, training and testing the CNN model to obtain an initial non-invasive load monitoring model, and performing gradient inspection; testing the initial non-invasive load monitoring model based on a test set to obtain a test result, and obtaining an evaluation result based on the test result; and optimizing the initial non-invasive load monitoring model according to the evaluation result to obtain the non-invasive load monitoring model.
It can be understood that, the non-intrusive load monitoring system based on deep learning provided by the embodiment of the present invention corresponds to the non-intrusive load monitoring method based on deep learning, and explanations, examples, and beneficial effects of relevant contents thereof may refer to corresponding contents in the non-intrusive load monitoring method based on deep learning, and are not described herein again.
Embodiments of the present invention also include a computer-readable storage medium storing a computer program for deep learning based non-intrusive load monitoring, wherein the computer program causes a computer to perform the deep learning based non-intrusive load monitoring method as described above.
In summary, compared with the prior art, the method has the following beneficial effects:
1. according to the embodiment of the invention, one-dimensional power data is converted into two-dimensional image data to be monitored, time domain information is incorporated into original data, so that the data contains more effective information, and a non-invasive load monitoring model based on deep learning is used.
2. The embodiment of the invention uses the power data as the identification parameter, and the output result of the model is directly associated with the power data, so that the load monitoring result has practical significance in the demand side management.
3. The embodiment of the invention can provide a basis for load optimization scheduling in the residential micro-grid by carrying out non-invasive monitoring on the general table data, thereby being beneficial to balancing the power supply and demand in the residential micro-grid.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The above examples are only intended to illustrate the technical solution of the present invention, but 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 (10)

1. A non-intrusive load monitoring method based on deep learning is characterized by comprising the following steps:
s1, performing time window decomposition on the one-dimensional power data to be monitored collected in the summary table to obtain k pieces of one-dimensional power data to be monitored;
s2, performing image transformation on k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, scaling the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and performing zero-value normalization processing;
and S3, inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model, and obtaining the working state of each power load in each time period.
2. The deep learning-based non-invasive load monitoring method according to claim 1, wherein the pre-trained CNN-based non-invasive load monitoring model is constructed by a process comprising:
a1, performing time window decomposition and image conversion on power data acquired by a table in a preset historical time period to obtain two-dimensional image data in a modeling process;
a2, preprocessing the two-dimensional image data in the modeling process to obtain standard two-dimensional image data;
a3, performing label processing and segmentation processing on the standard two-dimensional image data to obtain a training set and a test set;
a4, obtaining an initial non-invasive load monitoring model based on the training set and the CNN model, and performing gradient test;
a5, testing the initial non-invasive load monitoring model based on the test set to obtain a test result, and obtaining an evaluation result based on the test result;
a6, optimizing the initial non-invasive load monitoring model according to the evaluation result to obtain a non-invasive load monitoring model.
3. The deep learning-based non-invasive load monitoring method according to claim 2, wherein the image transformation comprises a gram angle field.
4. The deep learning-based non-invasive load monitoring method according to claim 2, wherein the a3 specifically comprises:
corresponding to different electric appliances and collected d1Data of days, wherein
Figure FDA0003185409590000021
As a training set train, the training set train,
Figure FDA0003185409590000022
as test set test; labeling the training set train according to the type of the electric appliance to obtain a label set label; the label type rule is consistent with the number of the electric appliances, namely labels of 1, 2 … i electric appliances are 1, 2 … i;
Figure FDA0003185409590000023
Figure FDA0003185409590000024
label＝[1,2,...i]T
wherein: each row in the training set train and the test set test represents one input variable, for a total of i variables.
5. The deep learning-based non-invasive load monitoring method according to claim 2, wherein the a4 specifically comprises:
setting the number of layers of convolution layers in the CNN model and the size of convolution kernels of the layers, the number of neurons of all layers of all connection layers, regularization parameters, dropout parameters, batch size and learning rate size, and selecting Adam for gradient inspection; inputting a training set train and training to obtain an initial non-invasive load monitoring model;
the gradient test was performed using the following formula:
Figure FDA0003185409590000031
Figure FDA0003185409590000032
wherein: j (theta)i) Is a loss function with regularization parameter, ε 10-7And d theta is a derivative value calculated when the initial non-invasive load monitoring model propagates reversely.
6. The deep learning-based non-invasive load monitoring method according to claim 2, wherein the a6 specifically comprises:
inputting the test set into an initial non-invasive load monitoring model to obtain a test result; evaluating the test result by using the root mean square error, the average absolute error and the average absolute error percentage according to the test result and the type of the electric appliances actually corresponding to the test set to obtain an evaluation result; the calculation formulas of the root mean square error, the average absolute error and the average absolute error percentage are as follows:
Figure FDA0003185409590000033
Figure FDA0003185409590000034
Figure FDA0003185409590000035
wherein: y isp,iObtaining a test result for each group of input data of the initial non-invasive load monitoring model, namely the type of the electric appliance predicted by the model; and y isr,iThe type of the electric appliance actually corresponding to the input data; n is the total number of appliance types in each set of input data.
7. A deep learning based non-intrusive load monitoring system, the system comprising:
the data decomposition and conversion module is used for carrying out time window decomposition on the one-dimensional power data to be monitored, which are collected in the summary table, so as to obtain k pieces of one-dimensional power data to be monitored;
the data processing module is used for carrying out image conversion on the k pieces of one-dimensional power data to be monitored to obtain k two-dimensional images to be monitored, scaling the k two-dimensional images to be monitored into two-dimensional image data to be monitored with the same size, and carrying out zero value normalization processing;
and the non-invasive load monitoring and solving module is used for inputting the normalized two-dimensional image data to be monitored into a pre-trained CNN-based non-invasive load monitoring model to obtain the working state of each power load in each time interval.
8. The deep learning based non-invasive load monitoring system according to claim 7,
the data processing module is also used for carrying out time window decomposition and image conversion on the power data of the preset historical time period acquired by the meter to obtain two-dimensional image data in the modeling process;
the data processing module is also used for carrying out unified scaling and normalization processing on the two-dimensional image data, dividing the processed image data into a training set and a testing set in the module, and carrying out labeling according to the type of the electric appliances.
9. The deep learning-based non-invasive load monitoring system according to claim 8, wherein the pre-trained CNN-based non-invasive load monitoring model is constructed by a model generation and optimization module;
the model generation and optimization module is used for inputting the generated data set into the CNN model, training and testing the CNN model to obtain an initial non-invasive load monitoring model, and performing gradient inspection; testing the initial non-invasive load monitoring model based on a test set to obtain a test result, and obtaining an evaluation result based on the test result; and optimizing the initial non-invasive load monitoring model according to the evaluation result to obtain the non-invasive load monitoring model.
10. A computer-readable storage medium storing a computer program for deep learning based non-intrusive load monitoring, wherein the computer program causes a computer to perform the deep learning based non-intrusive load monitoring method as recited in any of claims 1 to 6.
CN202110860151.4A 2021-07-28 2021-07-28 Non-invasive load monitoring method and system based on deep learning Pending CN113808071A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110860151.4A CN113808071A (en) 2021-07-28 2021-07-28 Non-invasive load monitoring method and system based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110860151.4A CN113808071A (en) 2021-07-28 2021-07-28 Non-invasive load monitoring method and system based on deep learning

Publications (1)

Publication Number Publication Date
CN113808071A true CN113808071A (en) 2021-12-17

Family

ID=78942640

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110860151.4A Pending CN113808071A (en) 2021-07-28 2021-07-28 Non-invasive load monitoring method and system based on deep learning

Country Status (1)

Country Link
CN (1) CN113808071A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI802245B (en) * 2022-01-24 2023-05-11 台灣電力股份有限公司 Power consumption analysis system and power consumption analysis method based on non-intrusive appliance load monitoring

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI802245B (en) * 2022-01-24 2023-05-11 台灣電力股份有限公司 Power consumption analysis system and power consumption analysis method based on non-intrusive appliance load monitoring

Similar Documents

Publication Publication Date Title
CN109598451B (en) Non-invasive load identification method based on PCA (principal component analysis) and LSTM (least Square TM) neural network
CN113156917B (en) Power grid equipment fault diagnosis method and system based on artificial intelligence
KR20200128232A (en) Apparatus for predicting electricity demand and method thereof
CN111160626A (en) Power load time sequence control method based on decomposition and fusion
CN113902062A (en) Transformer area line loss abnormal reason analysis method and device based on big data
CN114742283A (en) Medium-voltage line loop closing current estimation and risk assessment method based on deep learning network
CN112614004A (en) Method and device for processing power utilization information
CN115169719A (en) Platform area state prediction model based on fuzzy clustering and BP neural network
Dai et al. Fault diagnosis of data-driven photovoltaic power generation system based on deep reinforcement learning
CN113808071A (en) Non-invasive load monitoring method and system based on deep learning
CN112836876A (en) Power distribution network line load prediction method based on deep learning
CN115952928A (en) Short-term power load prediction method, device, equipment and storage medium
CN109767037A (en) A kind of prediction technique of summer Daily treatment cost
CN114819292A (en) Park microgrid intelligent scheduling method and system based on comprehensive safety evaluation
Ma et al. A non-intrusive load decomposition algorithm for residents
CN113591391A (en) Power load control device, control method, terminal, medium and application
Kong et al. Running state prediction and evaluation of power transformers
Zhang et al. Load prediction based on depthwise separable convolution model
CN115410142A (en) Non-invasive load monitoring method and system based on low-frequency load characteristic coding
Chen et al. A dual attention-based CNN-GRU model for short-term electric load forecasting
CN114638555B (en) Power consumption behavior detection method and system based on multilayer regularization extreme learning machine
Wei et al. Predicting Electricity Usage Based on Deep Neural Network
CN115618752B (en) Power load control method and system based on Autoformer model
Xu et al. A Wavelet Transform and ResNet Based Method for Sensing the Operational Status of Secondary Equipment in the Distribution Room
Yang et al. Lasso-GBDT-based Investment Forecasting for Distribution Transformer Replacement Projects

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