CN113920375A - Fusion characteristic typical load recognition method and system based on combination of Faster R-CNN and SVM - Google Patents
Fusion characteristic typical load recognition method and system based on combination of Faster R-CNN and SVM Download PDFInfo
- Publication number
- CN113920375A CN113920375A CN202111284397.8A CN202111284397A CN113920375A CN 113920375 A CN113920375 A CN 113920375A CN 202111284397 A CN202111284397 A CN 202111284397A CN 113920375 A CN113920375 A CN 113920375A
- Authority
- CN
- China
- Prior art keywords
- characteristic
- load
- wavelet
- network
- signal
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 230000004927 fusion Effects 0.000 title claims abstract description 18
- 238000012706 support-vector machine Methods 0.000 claims abstract description 37
- 238000000605 extraction Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 10
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 238000010586 diagram Methods 0.000 claims description 35
- 230000007246 mechanism Effects 0.000 claims description 16
- 238000001514 detection method Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 8
- 230000009021 linear effect Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000011176 pooling Methods 0.000 claims 1
- 238000004134 energy conservation Methods 0.000 abstract description 6
- 238000013527 convolutional neural network Methods 0.000 description 29
- 230000006870 function Effects 0.000 description 5
- 230000005611 electricity Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000009434 installation Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Complex Calculations (AREA)
Abstract
The invention relates to a fusion characteristic typical load recognition method and a system based on combination of fast R-CNN and SVM, wherein the method comprises the following steps: s1: collecting the load of electrical equipment, constructing a signal data sample, selecting a corresponding wavelet base, and determining the number of decomposition layers; decomposing the noisy signals to obtain a group of wavelet coefficients; s2: performing soft threshold function processing on the wavelet coefficient to obtain an estimated wavelet coefficient; s3: reconstructing the wavelet by using the estimated wavelet coefficient to obtain a de-noised reconstruction signal; s4: carrying out normalization processing on the reconstruction signal to obtain a normalized reconstruction signal; s5: and constructing a load characteristic curve image based on the normalized reconstructed signal, performing characteristic extraction on the load characteristic curve image by using an Faster R-CNN network, and classifying by using an SVM (support vector machine) to obtain a final load identification result. The method provided by the invention improves the accuracy of typical load characteristic identification, guides energy conservation, reduces the power consumption cost of users and improves the utilization rate of electric energy.
Description
Technical Field
The invention relates to the field of intelligent electric meters, in particular to a fusion characteristic typical load identification method and system based on combination of Faster R-CNN and SVM.
Background
The charge control intelligent electric meter is directly installed on a user total current inlet bus to collect power information parameters of a plurality of electric equipment in a user house as an intelligent metering instrument. The system can carry out multiple accurate measurement records on the electric energy and support real-time multi-rate of electricity price, and is widely applied. Load identification is an important technology of the fee-controlled intelligent electric meter, and the total power can be refined into the load. The power utilization list of the main electric appliances is provided for the user, and guidance is given to energy conservation, so that the power utilization cost of the user can be reduced, and the utilization rate of electric energy can be improved.
Load identification is broadly divided into two categories, invasive and non-invasive. The non-invasive load identification is to identify the electrical characteristics of the load such as current, power and the like when the current is in the house. Non-invasive load identification refers to the installation of identification devices at each type of electrical consumer interface. Since the installation, maintenance and other operations of the intrusive load identification device are difficult to realize, the non-intrusive load identification is widely accepted. The non-invasive load identification technology of the intelligent electric meter is an important electric energy classification metering technology in an intelligent power grid and is also one of important measures for realizing energy conservation and emission reduction. At present, the defects of precision loss and confusion of similar load characteristic electric appliances still exist when the load is identified by utilizing the identification capability of a convolutional neural network to an image, so that how to improve the identification accuracy needs to be solved urgently.
Disclosure of Invention
In order to solve the technical problems, the invention provides a fusion characteristic typical load identification method and system based on the combination of Faster R-CNN and SVM.
The technical solution of the invention is as follows: a fusion characteristic typical load recognition method based on combination of fast R-CNN and SVM comprises the following steps:
step S1: selecting n days of a day to be measured, collecting M point loads of the electrical equipment each day, constructing a signal data sample to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics, and determining the number of decomposition layers; decomposing the noisy signals to obtain a group of wavelet coefficients;
step S2: determining a threshold value of each wavelet coefficient of each layer in a grading manner, and performing soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients;
step S3: reconstructing the wavelet by using the estimated wavelet coefficient to obtain a de-noised reconstructed signal;
step S4: carrying out normalization processing on the reconstruction signal to obtain a normalized reconstruction signal;
step S5: constructing a load characteristic curve image based on the normalized reconstruction signal, performing characteristic extraction on the load characteristic curve image by using an Faster R-CNN network to obtain a characteristic diagram, calculating to obtain an interested region, generating an interested region characteristic diagram, generating a characteristic sequence based on the interested region characteristic diagram, obtaining a characteristic sequence with weight by using an attention mechanism, and classifying the characteristic sequence with weight by using an SVM to obtain a final load identification result; wherein the Faster R-CNN network comprises: the method comprises the steps of a feature extraction network, a regional suggestion network, a regional feature map generation network and a classification and regression detection network.
Compared with the prior art, the invention has the following advantages:
the invention provides a fusion characteristic typical load recognition method based on Fast R-CNN, a Support Vector Machine (SVM) and an attention mechanism, wherein the Fast R-CNN has super-strong learning ability and can realize characteristic extraction by using a small amount of sample data, and the network is more suitable for capturing local characteristics; the introduced attention mechanism can focus on the relationship between elements in the global sequence more clearly, and the two methods are combined to carry out more efficient learning and identification on local features and global relations in the feature sequence. In addition, the invention uses a support vector machine for classification, and combines the generalization capability of the SVM to the small sample data set with the feature extraction capability of the Faster R-CNN, thereby improving the applicability of the whole model to the small sample data set, improving the accuracy of typical load feature identification, guiding energy conservation, reducing the electricity consumption cost of users and improving the utilization rate of electric energy.
Drawings
FIG. 1 is a flowchart of a typical load recognition method based on fusion characteristics of combination of Faster R-CNN and SVM in the embodiment of the present invention;
FIG. 2 is a schematic diagram of an attention mechanism in a Faster R-CNN network according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a fast R-CNN network according to an embodiment of the present invention;
FIG. 4 is a block diagram of a typical load recognition system based on fusion characteristics of the combination of Faster R-CNN and SVM in the embodiment of the present invention.
Detailed Description
The invention provides a fusion characteristic typical load identification method based on combination of fast R-CNN and SVM, which improves the accuracy of typical load characteristic identification, guides energy conservation, reduces the electricity consumption cost of users and improves the utilization rate of electric energy.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings.
Example one
As shown in fig. 1, a typical load recognition method based on fusion characteristics of combination of fast R-CNN and SVM according to an embodiment of the present invention includes the following steps:
step S1: selecting n days of a day to be measured, collecting M point loads of the electrical equipment each day, constructing a signal data sample to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics, and determining the number of decomposition layers; decomposing the noisy signals to obtain a group of wavelet coefficients;
step S2: determining a threshold value of each wavelet coefficient of each layer in a grading manner, and performing soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients;
step S3: reconstructing the wavelet by using the estimated wavelet coefficient to obtain a de-noised reconstruction signal;
step S4: carrying out normalization processing on the reconstruction signal to obtain a normalized reconstruction signal;
step S5: based on the normalized reconstruction signal, constructing a load characteristic curve image, performing characteristic extraction on the load characteristic curve image by using an Faster R-CNN network to obtain a characteristic diagram, calculating to obtain an interested region, generating a characteristic diagram of the interested region, generating a characteristic sequence based on the characteristic diagram of the interested region, obtaining a characteristic sequence with weight by using an attention mechanism, and classifying the characteristic sequence with weight by using an SVM to obtain a final load identification result; wherein, the Faster R-CNN network comprises: the method comprises the steps of a feature extraction network, a regional suggestion network, a regional feature map generation network and a classification and regression detection network.
In one embodiment, the step S1: selecting n days of a day to be measured, constructing a signal data sample by using M load points each day to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics, and determining the number j of decomposition layers; decomposing the noisy signal to obtain a group of wavelet coefficients, specifically comprising:
step S11: selecting n days of a day to be measured, collecting M point loads of the electrical equipment each day, constructing a signal data sample, and obtaining a load characteristic set as X ═ X1,x2,...,xi,...,xNN is the number of load characteristics of the electrical equipment, xiIs the ith characteristic in the load characteristic set X;
step S12: a noisy signal is calculated according to the following equation (1):
g(k)=f(k)+s(k)k=1,2,3,…,M-1 (1)
wherein k is time, f (k) is real signal, s (k) is noise signal, g (k) is noise-containing signal;
step S13: selecting a corresponding wavelet basis based on the noise-containing signals g (k), and selecting a smooth continuous wavelet basis when the noise-containing signals are smooth; when the linear property of the noisy signal is stronger, selecting a linear wavelet base;
step S14: decomposing the noise-containing signals g (k) according to a preset decomposition layer number j to obtain a group of wavelet coefficients omegaj,k。
In one embodiment, the step S2: determining a threshold value of each wavelet coefficient of each layer in stages, and performing soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients, wherein the method specifically comprises the following steps:
performing soft threshold function processing on the wavelet coefficient according to formula (2) to obtain an estimated wavelet coefficient
Wherein T is a threshold value; when wavelet coefficient omegaj,kWhen the absolute value of (1) is greater than T, the wavelet coefficient is estimatedHas an absolute value of | ωj,kl-T, the sign remains unchanged; when wavelet coefficient omegaj,kIs less than or equal to T, the wavelet coefficient is estimatedIs zero.
In one embodiment, the step S3: using estimated wavelet coefficientsReconstructing the wavelet to obtain a de-noised reconstructed signal
In one embodiment, the step S4: for the reconstructed signalAnd carrying out normalization processing to normalize the value to a (0, 1) interval to obtain a normalized reconstruction signal.
The preprocessing operation is carried out on the data samples through the steps S1-S4, and the noise processing and normalization processing are carried out on the images through wavelet denoising, so that the influence of bad data on the subsequent neural network training is weakened.
In one embodiment, the step S5: based on the normalized reconstruction signal, constructing a load characteristic curve image, performing characteristic extraction on the load characteristic curve image by using an Faster R-CNN network to obtain a characteristic diagram, calculating to obtain an interested region, generating a characteristic diagram of the interested region, generating a characteristic sequence based on the characteristic diagram of the interested region, obtaining a characteristic sequence with weight by using an attention mechanism, and classifying the characteristic sequence with weight by using an SVM to obtain a final load identification result; wherein, the Faster R-CNN network comprises: the method comprises the following steps of extracting a network, suggesting a network in a region, generating a network by a regional characteristic graph and classifying and regressing a detection network, and specifically comprises the following steps:
step S51: constructing a load characteristic curve image based on the normalized reconstruction signal: inputting the load characteristic curve image into a characteristic extraction network to obtain a characteristic diagram;
the embodiment of the invention adopts a VGG network to realize the feature extraction of the load feature curve image;
step S52: inputting the feature map into a region suggestion network (RPN), and calculating to obtain a target region to be identified on the feature map to obtain a region of interest;
step S53: inputting the characteristic diagram of the region of interest into a network, generating the characteristic diagram of the region of interest by using ROI poolingCalculating the over-convolution layer to obtain the characteristic sequenceComputing feature sequences using an attention mechanismObtaining a weighted characteristic sequence y ═ y1,y2,...,yt,...,yN]The method comprises the following specific steps:
according to the following formula (3), a query vector q ═ q is calculated1,q2,...,qi,...,qN]The key vector m ═ m1,m2,...,mi,...,mN]Sum vector v ═ v1,v2,...,vi,...,vN]:
Wherein, Wq、WmAnd WvIs a preset weight matrix coefficient;
according to equation (4), each is calculatedAttention score of ct,iI.e. at presentCorrelation between the power value at time t and the power values of other vectors in the signature sequence;
ct,i=qt·mi (4)
the attention score is normalized according to equation (5):
the normalized attention scores are weighted and summed according to equation (6):
wherein, ytWeighted sequence y ═ y for output after attention mechanism calculation1,y2,...,yt,...,yN]The value at time t.
FIG. 2 is a schematic diagram of the attention mechanism in the Faster R-CNN network.
In the training process of the Faster R-CNN network, the embodiment of the invention allocates a weight coefficient to each element in the characteristic sequence by introducing an attention mechanism, so that the attention of the network can be put on the details such as the peak-valley length with more judgment information, and the like, and the influence of other fluctuations on the classification result is weakened.
Step S54: inputting the sequences with weights into a classification and regression detection network, and classifying the sequences by using a one-to-one method Support Vector Machine (SVM), wherein the classification specifically comprises the following steps:
according to the weighted sequence y ═ y1,y2,...,yt,...,yN]And classifying the classes by using an SVM to obtain z classes, judging again by using z (z-1)/2 secondary classifiers, voting for the corresponding classes, and taking the class with the most votes as a final load identification result.
According to the embodiment of the invention, the SVM is selected as the feature classifier, so that the generalization capability of small samples can be improved.
FIG. 3 is a schematic diagram of the fast R-CNN network structure.
The invention provides a fusion characteristic typical load recognition method based on Fast R-CNN, a Support Vector Machine (SVM) and an attention mechanism, wherein the Fast R-CNN has super-strong learning ability and can realize characteristic extraction by using a small amount of sample data, and the network is more suitable for capturing local characteristics; the introduced attention mechanism can focus on the relationship between elements in the global sequence more clearly, and the two methods are combined to carry out more efficient learning and identification on local features and global relations in the feature sequence. In addition, the invention uses a support vector machine for classification, and combines the generalization capability of the SVM to the small sample data set with the feature extraction capability of the Faster R-CNN, thereby improving the applicability of the whole model to the small sample data set, improving the accuracy of typical load feature identification, guiding energy conservation, reducing the electricity consumption cost of users and improving the utilization rate of electric energy.
Example two
As shown in fig. 4, an embodiment of the present invention provides a fusion characteristic typical load recognition system based on the combination of fast R-CNN and SVM, which includes the following modules:
the wavelet coefficient acquisition module 61 is used for selecting n days to be measured, acquiring M point loads of the electrical equipment each day, constructing a signal data sample to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics, and determining the decomposition layer number; decomposing the noisy signals to obtain a group of wavelet coefficients;
an estimated wavelet coefficient obtaining module 62, configured to determine a threshold of each layer of wavelet coefficients in a staged manner, and perform soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients;
a reconstructed signal module 63, configured to reconstruct the wavelet by using the estimated wavelet coefficient to obtain a denoised reconstructed signal;
a normalization module 64, configured to perform normalization processing on the reconstructed signal to obtain a normalized reconstructed signal;
the load identification module 65 is configured to construct a load characteristic curve image based on the normalized reconstructed signal, perform characteristic extraction on the load characteristic curve image by using an fast R-CNN network to obtain a characteristic diagram, calculate an interested region, generate an interested region characteristic diagram, generate a characteristic sequence based on the interested region characteristic diagram, obtain a characteristic sequence with weights by using an attention mechanism, and classify the characteristic sequence to obtain a final load identification result; wherein, the Faster R-CNN network comprises: the method comprises the steps of a feature extraction network, a regional suggestion network, a regional feature map generation network and a classification and regression detection network.
The above examples are provided only for the purpose of describing the present invention, and are not intended to limit the scope of the present invention. The scope of the invention is defined by the appended claims. Various equivalent substitutions and modifications can be made without departing from the spirit and principles of the invention, and are intended to be within the scope of the invention.
Claims (5)
1. A fusion characteristic typical load recognition method based on combination of fast R-CNN and SVM is characterized by comprising the following steps:
step S1: selecting n days of a day to be measured, collecting M point loads of the electrical equipment each day, constructing a signal data sample to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics, and determining the number of decomposition layers; decomposing the noisy signals to obtain a group of wavelet coefficients;
step S2: determining a threshold value of each wavelet coefficient of each layer in a grading manner, and performing soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients;
step S3: reconstructing the wavelet by using the estimated wavelet coefficient to obtain a de-noised reconstructed signal;
step S4: carrying out normalization processing on the reconstruction signal to obtain a normalized reconstruction signal;
step S5: constructing a load characteristic curve image based on the normalized reconstruction signal, performing characteristic extraction on the load characteristic curve image by using an Faster R-CNN network to obtain a characteristic diagram, calculating to obtain an interested region, generating an interested region characteristic diagram, generating a characteristic sequence based on the interested region characteristic diagram, obtaining a characteristic sequence with weight by using an attention mechanism, and classifying the characteristic sequence with weight by using an SVM to obtain a final load identification result; wherein the Faster R-CNN network comprises: the method comprises the steps of a feature extraction network, a regional suggestion network, a regional feature map generation network and a classification and regression detection network.
2. The fusion feature canonical load recognition method based on the combination of fast R-CNN and SVM according to claim 1, wherein the step S1: selecting n days of a day to be measured, constructing a signal data sample by using M load points each day to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics, and determining the number j of decomposition layers; decomposing the noisy signal to obtain a group of wavelet coefficients, specifically comprising:
step S11: selecting n days of a day to be measured, collecting M point loads of the electrical equipment each day, constructing a signal data sample, and obtaining a load characteristic set as X ═ X1,x2,...,xi,...,xNN is the number of load characteristics of the electrical equipment, xiThe ith characteristic in the load characteristic set X is taken as the characteristic;
step S12: a noisy signal is calculated according to the following equation (1):
g(k)=f(k)+s(k)k=1,2,3,…,M-1 (1)
wherein k is time, f (k) is real signal, s (k) is noise signal, g (k) is noise-containing signal;
step S13: selecting a corresponding wavelet basis based on the noise-containing signals g (k), and selecting smooth continuous wavelet basis when the noise-containing signals are smooth; when the linear property of the noise-containing signal is stronger, selecting a linear wavelet base;
step S14: decomposing the noise-containing signals g (k) according to a preset decomposition layer number j to obtain a group of wavelet coefficients omegaj,k。
3. The fusion feature canonical load recognition method based on the combination of fast R-CNN and SVM according to claim 1, wherein the step S2: determining a threshold value of each wavelet coefficient of each layer in stages, and performing soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients, wherein the method specifically comprises the following steps:
performing soft threshold function processing on the wavelet coefficient according to a formula (2) to obtain an estimated wavelet coefficient
4. The fusion feature canonical load recognition method based on the combination of fast R-CNN and SVM according to claim 1, wherein the step S5: constructing a load characteristic curve image based on the normalized reconstruction signal, performing characteristic extraction on the load characteristic curve image by using an Faster R-CNN network to obtain a characteristic diagram, calculating to obtain an interested region, generating an interested region characteristic diagram, generating a characteristic sequence based on the interested region characteristic diagram, obtaining a characteristic sequence with weight by using an attention mechanism, and classifying the characteristic sequence to obtain a final load identification result; wherein the Faster R-CNN network comprises: the method comprises the following steps of extracting a network, suggesting a network in a region, generating a network by a regional characteristic graph and classifying and regressing a detection network, and specifically comprises the following steps:
step S51: and constructing a load characteristic curve image based on the normalized reconstruction signal: inputting the load characteristic curve image into the characteristic extraction network to obtain a characteristic diagram;
step S52: inputting the characteristic diagram into a region suggestion network (RPN), and calculating a target region to be identified on the characteristic diagram to obtain a region of interest;
step S53: inputting the characteristic diagram of the region of interest into a region characteristic diagram generation network, generating the characteristic diagram of the region of interest by using ROI pooling, and obtaining a characteristic sequence through convolutional layer calculationComputing the sequence of features using an attention mechanismObtaining a weighted characteristic sequence y ═ y1,y2,...,yt,...,yN]The method comprises the following specific steps:
according to the following formula (3), a query vector q ═ q is calculated1,q2,...,qi,...,qN]The key vector m ═ m1,m2,...,mi,...,mN]Sum vector v ═ v1,v2,...,vi,...,vN]:
Wherein, Wq、WmAnd WvIs a preset weight matrix coefficient;
according to equation (4), each is calculatedAttention score of ct,iI.e. at presentCorrelation between the power value at time t and the power values of other vectors in the signature sequence;
ct,i=qt·mi (4)
normalizing the attention score according to equation (5):
the normalized attention scores are weighted and summed according to equation (6):
wherein, ytWeighted sequence y ═ y for output after attention mechanism calculation1,y2,...,yt,...,yN]The value at time t;
step S54: inputting the weighted sequence into a classification and regression detection network, and classifying the weighted sequence by using a one-to-one method Support Vector Machine (SVM), wherein the classification specifically comprises the following steps:
according to the weighted sequence y ═ y1,y2,...,yt,...,yN]The load recognition method comprises the steps of classifying the load by using an SVM (support vector machine) to obtain z classes, judging the z classes by using z (z-1)/2 classifiers, voting for the corresponding classes, and taking the class with the largest vote as a final load recognition result.
5. A fusion characteristic typical load recognition system based on combination of fast R-CNN and SVM is characterized by comprising the following modules:
the wavelet coefficient acquisition module is used for selecting n days to be measured, acquiring M point loads of the electrical equipment each day, constructing a signal data sample to obtain a load characteristic set, selecting a corresponding wavelet base according to the load characteristics and determining the decomposition layer number; decomposing the noisy signals to obtain a group of wavelet coefficients;
the wavelet coefficient estimation module is used for determining a threshold value of each wavelet coefficient in stages and performing soft threshold function processing on the wavelet coefficients to obtain estimated wavelet coefficients;
the reconstructed signal module is used for reconstructing the wavelet by using the estimated wavelet coefficient to obtain a de-noised reconstructed signal;
the normalization module is used for performing normalization processing on the reconstruction signal to obtain a normalized reconstruction signal;
the load identification module is used for constructing a load characteristic curve image based on the normalized reconstruction signal, extracting the characteristics of the load characteristic curve image by using an Faster R-CNN network to obtain a characteristic diagram, calculating to obtain an interested area, generating an interested area characteristic diagram, generating a characteristic sequence based on the interested area characteristic diagram, obtaining the characteristic sequence with the weight by using an attention mechanism, and classifying the characteristic sequence to obtain a final load identification result; wherein the Faster R-CNN network comprises: the method comprises the steps of a feature extraction network, a regional suggestion network, a regional feature map generation network and a classification and regression detection network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111284397.8A CN113920375B (en) | 2021-11-01 | 2021-11-01 | Fusion characteristic typical load identification method and system based on combination of Faster R-CNN and SVM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111284397.8A CN113920375B (en) | 2021-11-01 | 2021-11-01 | Fusion characteristic typical load identification method and system based on combination of Faster R-CNN and SVM |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113920375A true CN113920375A (en) | 2022-01-11 |
CN113920375B CN113920375B (en) | 2024-06-25 |
Family
ID=79244809
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111284397.8A Active CN113920375B (en) | 2021-11-01 | 2021-11-01 | Fusion characteristic typical load identification method and system based on combination of Faster R-CNN and SVM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113920375B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114332443A (en) * | 2022-03-16 | 2022-04-12 | 北京智芯微电子科技有限公司 | Non-intrusive load recognition device, method, storage medium, and electronic apparatus |
CN116722557A (en) * | 2023-05-30 | 2023-09-08 | 国网北京市电力公司 | Demand response rebound time length analysis method and system based on wavelet decomposition |
CN117496256A (en) * | 2023-11-14 | 2024-02-02 | 中国人民解放军军事科学院***工程研究院 | Small sample electromagnetic signal identification method and device |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106908671A (en) * | 2017-01-13 | 2017-06-30 | 中国海洋大学 | A kind of non-intrusion type household loads intelligent detecting method and system |
CN110070229A (en) * | 2019-04-26 | 2019-07-30 | 中国计量大学 | The short term prediction method of home electrical load |
CN110188826A (en) * | 2019-05-31 | 2019-08-30 | 湖南科技大学 | Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data |
CN110514889A (en) * | 2019-07-19 | 2019-11-29 | 浙江万胜智能科技股份有限公司 | A kind of method and system of non-intrusion type household electricity remained capacity |
CN111242423A (en) * | 2019-12-31 | 2020-06-05 | 深圳能源售电有限公司 | Active demand response potential analysis method, device, equipment and storage medium |
US20200219620A1 (en) * | 2019-01-08 | 2020-07-09 | International Business Machines Corporation | Personalized Smart Home Recommendations Through Cognitive Load Analysis |
WO2021208502A1 (en) * | 2020-04-16 | 2021-10-21 | 中国科学院深圳先进技术研究院 | Remote-sensing image target detection method based on smooth bounding box regression function |
-
2021
- 2021-11-01 CN CN202111284397.8A patent/CN113920375B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106908671A (en) * | 2017-01-13 | 2017-06-30 | 中国海洋大学 | A kind of non-intrusion type household loads intelligent detecting method and system |
US20200219620A1 (en) * | 2019-01-08 | 2020-07-09 | International Business Machines Corporation | Personalized Smart Home Recommendations Through Cognitive Load Analysis |
CN110070229A (en) * | 2019-04-26 | 2019-07-30 | 中国计量大学 | The short term prediction method of home electrical load |
CN110188826A (en) * | 2019-05-31 | 2019-08-30 | 湖南科技大学 | Household electrical appliance operating status non-invasive inspection methods based on intelligent electric meter data |
CN110514889A (en) * | 2019-07-19 | 2019-11-29 | 浙江万胜智能科技股份有限公司 | A kind of method and system of non-intrusion type household electricity remained capacity |
CN111242423A (en) * | 2019-12-31 | 2020-06-05 | 深圳能源售电有限公司 | Active demand response potential analysis method, device, equipment and storage medium |
WO2021208502A1 (en) * | 2020-04-16 | 2021-10-21 | 中国科学院深圳先进技术研究院 | Remote-sensing image target detection method based on smooth bounding box regression function |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114332443A (en) * | 2022-03-16 | 2022-04-12 | 北京智芯微电子科技有限公司 | Non-intrusive load recognition device, method, storage medium, and electronic apparatus |
CN114332443B (en) * | 2022-03-16 | 2022-06-24 | 北京智芯微电子科技有限公司 | Non-intrusive load recognition device, method, storage medium, and electronic apparatus |
CN116722557A (en) * | 2023-05-30 | 2023-09-08 | 国网北京市电力公司 | Demand response rebound time length analysis method and system based on wavelet decomposition |
CN117496256A (en) * | 2023-11-14 | 2024-02-02 | 中国人民解放军军事科学院***工程研究院 | Small sample electromagnetic signal identification method and device |
Also Published As
Publication number | Publication date |
---|---|
CN113920375B (en) | 2024-06-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113920375A (en) | Fusion characteristic typical load recognition method and system based on combination of Faster R-CNN and SVM | |
CN113191215B (en) | Rolling bearing fault diagnosis method integrating attention mechanism and twin network structure | |
CN112435142B (en) | Power load identification method and load power utilization facility knowledge base construction method thereof | |
CN113157771A (en) | Data anomaly detection method and power grid data anomaly detection method | |
CN112598303A (en) | Non-invasive load decomposition method based on combination of 1D convolutional neural network and LSTM | |
CN114841293B (en) | Multimode data fusion analysis method and system for power Internet of things | |
Liu et al. | A novel dual-attention optimization model for points classification of power quality disturbances | |
CN115410258A (en) | Human face expression recognition method based on attention image | |
CN111428772A (en) | Photovoltaic system depth anomaly detection method based on k-nearest neighbor adaptive voting | |
CN114626625A (en) | Bearing life prediction method based on multi-module U-BiLSTM network | |
CN113987910A (en) | Method and device for identifying load of residents by coupling neural network and dynamic time planning | |
CN117351307A (en) | Model training method, device, equipment and storage medium | |
CN117611830A (en) | Random class target positioning and counting method based on few sample labeling | |
CN115456957B (en) | Method for detecting change of remote sensing image by full-scale feature aggregation | |
Zheng et al. | A general fault diagnosis framework for rotating machinery and its flexible application example | |
CN116304735A (en) | Method, system and equipment for diagnosing surface damage of steel rail based on dependency measurement | |
Zhao et al. | Week-ahead hourly solar irradiation forecasting method based on ICEEMDAN and TimesNet networks | |
CN112241922B (en) | Power grid asset comprehensive value assessment method based on improved naive Bayesian classification | |
CN116011654A (en) | District popularity index prediction method based on power load data | |
Lian | Wind speed prediction based on CEEMD-SE and multiple echo state network with Gauss–Markov fusion | |
CN114839586A (en) | Low-voltage transformer area metering device misalignment calculation method based on EM algorithm | |
Liu et al. | Support vector regression for bus travel time prediction using wavelet transform | |
CN110852597B (en) | Electricity consumption peak period resident load ratio calculation method based on generation of countermeasure network | |
CN114706054A (en) | Method for identifying human body motion micro Doppler signal | |
Wang et al. | Real-time power quality waveform recognition with a programmable digital signal processor |
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 |