CN114359674A - Non-invasive load identification method based on metric learning - Google Patents

Non-invasive load identification method based on metric learning Download PDF

Info

Publication number
CN114359674A
CN114359674A CN202210032592.XA CN202210032592A CN114359674A CN 114359674 A CN114359674 A CN 114359674A CN 202210032592 A CN202210032592 A CN 202210032592A CN 114359674 A CN114359674 A CN 114359674A
Authority
CN
China
Prior art keywords
load
feature
identified
sample
switching event
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
CN202210032592.XA
Other languages
Chinese (zh)
Inventor
于淼
王丙楠
陆玲霞
赵强
包哲静
程卫东
魏萍
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.)
Zhejiang University ZJU
Holley Technology Co Ltd
Original Assignee
Zhejiang University ZJU
Holley Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU, Holley Technology Co Ltd filed Critical Zhejiang University ZJU
Priority to CN202210032592.XA priority Critical patent/CN114359674A/en
Publication of CN114359674A publication Critical patent/CN114359674A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a non-invasive load identification method based on metric learning. The method can realize effective identification of unknown load and has strong generalization capability; on the other hand, metric learning, which is one of the methods for small sample learning, can reduce the dependence on training samples, and has high practicability.

Description

Non-invasive load identification method based on metric learning
Technical Field
The invention relates to the field of non-intrusive load monitoring (NILM), in particular to a non-intrusive load identification method based on metric learning.
Background
With the rapid development of society, the demand for energy is increasing, and electric energy is one of the main energy utilization methods as the main secondary energy. How to improve the electric energy use efficiency, realizing that intelligent power consumption arouses extensive concern, it is very important to master user side detailed power consumption information. Traditional intrusive load monitoring needs to install a collecting and communicating device at each electric load to detect the load state, needs to modify the existing electric appliances or lines, and is difficult to implement and high in cost. The non-intrusive load monitoring technology analyzes the state of each load in a line by monitoring the bus, has the advantages of strong universality, low cost and the like, and becomes a research hotspot in recent years.
Most of the existing non-invasive load identification researches are based on optimization and pattern recognition methods, but all the methods have certain defects. Firstly, most of the models rely on a large number of label samples for model training, but in practical application, enough label data cannot be obtained or the obtaining cost is high; secondly, the models usually assume that all loads in a scene are known, and after a new load is added in the scene, the original model cannot identify the new load, and even influences the identification effect of the original load; finally, these models have poor generalization performance, and the models often need to be retrained after scene change, and training and establishing the models individually for each scene will result in greatly increased maintenance complexity and cost.
Disclosure of Invention
The invention aims to provide a non-invasive load identification method based on metric learning, which aims at solving the problems in the prior art, takes load current as a characteristic, maps the current characteristic to a new metric space through a convolutional neural network, realizes clustering by taking triple loss as network loss during network training, and only needs to measure the distance of a load characteristic vector in the metric space during identification. The method can realize effective identification of unknown load and has strong generalization capability; on the other hand, metric learning, which is one of the methods for small sample learning, can reduce the dependence on training samples, and has high practicability.
The technical scheme of the invention is as follows:
a non-intrusive load identification method based on metric learning comprises the following steps:
step 1, monitoring power bus power data in real time by using an event detection algorithm, positioning the occurrence time of a load switching event, and differentiating steady-state data before and after the occurrence time of the load switching event to separate out voltage and current and power data of switching of a load to be identified;
step 2, starting information acquisition of a complete period of the current data at the voltage positive zero crossing point, and performing resampling and normalization processing on the acquired current information;
step 3, using the trained feature extraction network to extract the features of the current to obtain the feature vector information of the load switching event to be identified;
step 4, checking whether the feature library exists or not, if not, establishing the feature library, storing feature vector information and power data of the load switching event to be identified and setting a corresponding feature number;
if the load switching event is the same as the sample in the feature library, the feature number of the sample in the feature library is used as the feature number of the load switching event to be identified; and if all the characteristic vector similarity conditions or the power similarity conditions are not met, the load switching event to be identified is a new event, the characteristic vector information and the power data of the load switching event to be identified are stored in a characteristic library, and corresponding characteristic numbers are set.
And 5, mapping the obtained feature numbers into actual load switching events, wherein the load switching events comprise load types and load state conversion information.
The feature number of the sample setting newly added into the feature library is mapped with the load switching event by the following method:
and when a sample newly added into the feature library exists, informing a user, and judging the actual load type of the newly added number and the state conversion information of the load by the user according to the historical switching record and the current-day actual use condition.
The historical switching record can be used for visual display of various types of load historical work information and the like.
Further, the feature extraction network is composed of a plurality of one-dimensional convolution layers, an activation function layer, a plurality of residual modules, a global average pooling layer, a linear layer and an L2 regularization layer which are connected in sequence.
Further, the feature extraction network uses the triplet loss as the network loss during training.
Further, when the feature extraction network is trained, the used training samples regard different working states of the same electrical appliance as separate devices.
Further, in the step 4, a cosine similarity calculation method is adopted to calculate the similarity between the feature vector of each sample in the feature library and the feature vector of the load to be identified, and a difference calculation method is adopted to calculate the similarity between the power of the sample and the power of the load to be identified.
Further, in the mapping of the feature number mapping and the actual load switching event, the single-state load and the multi-state load are processed differently, specifically as follows:
for single-state load, only two states of on/off are provided, the state conversion is simple, and only the characteristics of the state conversion process need to be labeled; for multi-state load, because the load may be switched among various states, in addition to marking the load type information, the change information of the load state also needs to be marked during marking, and the multi-state load switching information and the current state information can be accurately identified only by mastering the characteristic information during switching among various states.
The invention has the beneficial effects that:
firstly, metric learning is used as one of small sample learning methods, the requirement of a non-invasive load recognition model on the number of training samples can be reduced, and better recognition performance can be obtained through a small amount of sample training; secondly, load identification is realized in a mode of matching with the feature library, unknown loads are automatically added into the feature library for identification and classification, and later-stage user labeling equipment information is waited; finally, the metric learning model has strong generalization performance, retraining or model building is not needed when the metric learning model is migrated to a new scene, migration cost is greatly reduced, and the metric learning model has higher practical value. Meanwhile, in the process of establishing mapping between the sample feature number newly added into the feature library and the load switching event, the method also marks the current state of the electric appliance/load, and can obtain more accurate detailed power utilization information of the user side.
Drawings
Fig. 1 shows the current resampling and normalization process.
Fig. 2 is a schematic diagram of a current feature extraction network structure.
Fig. 3 is a schematic diagram of triple loss calculation.
FIG. 4 is a flowchart illustrating the overall recognition process of the present invention.
Fig. 5 is a diagram illustrating switching of a multi-state load state.
Fig. 6 shows the recognition result (confusion matrix) of the appliance test in the cool part when the model is trained in house6 in the plid.
Detailed Description
In order to verify the characteristics and effects of the present invention, the present invention will be further explained below with reference to the white data set, the PLAID data set, and the cool data set.
The invention provides a non-invasive load identification method based on metric learning, which is characterized by obtaining characteristic vector information through a trained characteristic extraction network, comparing the similarity of characteristic vector information and power information of a load switching event to be identified and samples in a characteristic library, judging whether the load switching event to be identified is a new event or not by combining the similarity, and determining the event type of the load switching event to be identified, which is not the new event. In the embodiment, the feature extraction network is constructed and trained, and then the method of the present invention is described in detail in combination with the trained feature extraction network.
Constructing and training a feature extraction network:
(1) and (3) training set construction:
the training set of the invention can adopt the samples of the public data sets in the existing non-invasive load identification field, taking the samples of house6 in the PLAID data set as an example, the construction method of the training set is as follows:
the house6 comprises 6 electric appliances of fluorescent lamp, air conditioner, refrigerator, fan, blower and notebook computer, wherein, the air conditioner and refrigerator respectively have 3 and 2 working states, and during model training, the different working states are treated as independent load types.
When training samples are collected from each load type sample, the number of test cases of each load type is different, and in order to ensure sample balance, when the number of samples is insufficient, the samples are expanded by using a oversampling technology (SMOTE), and the main process is as follows:
1) randomly selecting two samples x from the same sample0And x1；
2) The new sample was obtained as follows:
xnew＝x0+rand(0,1)(x1-x0)
rand (0,1) represents a random function, producing a random number of 0 or 1.
3) The steps are repeated until a specified number of samples are obtained.
And preprocessing the current information of each acquired sample, resampling the current information to a specified frequency, normalizing the frequency and using the frequency as network input. The detailed steps are as follows:
a, starting current collection when a voltage positive zero crossing point in order to obtain current phase information;
b, recording current information of a complete period;
c, resampling the current information, and obtaining the current information with the appointed sampling frequency by using a linear interpolation method, wherein the specific calculation process is as follows:
Figure BDA0003467101990000041
Figure BDA0003467101990000042
Figure BDA0003467101990000043
i'n＝(ceil(loc)-loc)·ifloor(loc)+(loc-floor(loc))·iceil(loc)
in the formula, N0And N1The number of points per period, Ts, of the original data and the sampled data respectively0And Ts1Respectively the sampling time before and after sampling, n is the nth data point after sampling, loc is the position of the nth data point corresponding to the original sequence, ceil and floor are respectively the upward and downward rounding functions, iceil(loc)And ifloor(loc)Current values, i'nThe current value of the nth sampled data point is obtained by linear interpolation calculation.
d, normalizing the sampled current and scaling the current amplitude to be between-1 and 1, namely performing the following operations on the current:
Figure BDA0003467101990000051
wherein I 'is a normalized current, and I ═ I'1,i’2,…,i’N1,]Is a sequence of sampled currents. max (abs (i)) represents the maximum absolute value of the sampled current sequence.
(2) Constructing a feature extraction network model: the feature extraction network in the present invention may be any feature extraction network, and in this embodiment, the reference ResNet model is constructed based on a one-dimensional convolution residual block, and is a measurement network model composed of a plurality of one-dimensional convolution layers, an activation function layer, a plurality of residual modules, a global average pooling layer, a linear layer, and an L2 regularization layer (not shown in the figure) connected in sequence. Fig. 2 is a schematic diagram of an exemplary structure, in which 64 one-dimensional convolution kernels with a size of 7 are first used to perform convolution operation on current, a step length is selected to be 2 to perform downsampling, then downsampling and feature extraction are sequentially performed through three residual modules, finally dimensionality expansion is performed after a global average pooling layer is performed, a feature vector extracted based on load current is obtained after a linear layer is performed, in order to eliminate the influence of a feature vector modulus value during distance calculation, an L2 regularization layer is used to unitize a feature vector output by a network to make the feature vector fall on a unit hypersphere, and at this time, an euclidean distance and a cosine distance can be regarded as equivalent. In order to accelerate convergence speed, BN (Batchnormalization) layer is added in the convolution layer and the residual module after each convolution, and the characteristic value is normalized. In the aspect of activation function selection, the network inputs current sampling values of [ -1,1], so that the Tanh function is selected as the activation function.
The triple loss function is used in the network training, and if each pair of samples is input into the network for training, the spatial complexity is O (N)3) And as the training process is carried out, the number of the triples with the loss of 0 is increased, and the network loss mean value is pulled down, so that the network training is slowed down until the triples fall into the local optimum. In order to solve the problem, triples need to be screened when a structure sample is trained each time, and a certain number of triples with loss not 0 are selected as a training sample group as shown in fig. 3, so that the network training process is accelerated, and the situation that the triples fall into local optimum is avoided.
Specifically, the triplet loss is taken as a loss function of network training, and is as follows:
L＝max(d(a,p)-d(a,n)+margin,0)
in the formula, the triplet is < a, p, n >, a is called anchor (anchor), p is positive instance (positive), and belongs to the same category as a, n is negative instance (negative), and a is not in the same category. d is a distance measurement function, the Euclidean distance (because normalization is carried out, the farther the distance on a unit hypersphere is, the larger the included angle between two vectors is, and certain equivalence is realized on the distance between the two vectors and the cosine distance) is used in training, and margin is a constant, so that the loss is 0 when the positive sample and the negative sample are both very close to the anchor point sample.
After the training is completed, a trained feature extraction network is obtained, and the trained feature extraction network is utilized to realize the non-invasive load identification method based on metric learning, as shown in fig. 4, the method comprises the following steps:
step 1, monitoring power bus power data by using an event detection algorithm in real time, positioning the occurrence time of a load switching event, and differentiating the steady-state data before and after the occurrence time of the load switching event to separate voltage and current and power data of the load switching event to be identified;
step 2, starting current information acquisition of a complete period of voltage and current data at a positive voltage zero-crossing point, and performing resampling and normalization processing on the acquired current information;
step 3, using the trained feature extraction network to extract the features of the current to obtain the feature vector information of the load switching event to be identified;
step 4, checking whether the feature library exists or not, if not, establishing the feature library, storing feature vector information and power data of the load switching event to be identified and setting a corresponding feature number;
if the load switching event is the same as the sample in the feature library, the feature number of the sample in the feature library is used as the feature number of the load switching event to be identified; and if all the characteristic vector similarity conditions or the power similarity conditions are not met, the load switching event to be identified is a new event, the characteristic vector information and the power data of the load switching event to be identified are stored in a characteristic library, and corresponding characteristic numbers are set.
The similarity calculation result is used to determine the similarity, and may adopt a cosine similarity, a difference calculation, an euclidean distance, or other calculation methods, and preferably, in this embodiment, the similarity between the feature vector of each sample in the feature library and the feature vector of the load to be identified is calculated by adopting a cosine similarity calculation method, which is specifically as follows:
Figure BDA0003467101990000061
in the formula, x and y represent two feature vectors, and n represents a feature vector length. The closer the value is to 1, the more similar the two features are, and in practice, the similarity condition can be set reasonably according to the situation.
Calculating the similarity between the power of the sample and the power of the load to be identified by adopting a difference value calculation method, which specifically comprises the following steps:
relative difference of (P)max-Pmin)/Pmin
Wherein P ismaxAnd PminRespectively a large value and a small value in the power of the load switching event to be identified and the power of the sample in the characteristic library.
In practice, the similarity condition can be set reasonably according to the situation, and different thresholds can be set for different power sections, so as to improve the identification precision.
The feature number of the sample setting newly added into the feature library is mapped with the load switching event by the following method:
and when a sample newly added into the feature library exists, informing a user, and judging the actual load type of the newly added feature number and the state conversion information of the load by the user according to the historical switching record and the current-day actual use condition. For the single-state load, only two states of on/off are provided, the state conversion is simple, and the characteristics of the state conversion process only need to be labeled; for multi-state load, as shown in fig. 5, because the load may be switched among various states, different states have different characteristic information, when labeling, in addition to labeling the load type information, the change information of the load state can be labeled, the characteristic information when switching among various states is grasped, the multi-state load switching information and the current state information can be accurately identified, and thus more and more accurate detailed power utilization information of the user side can be obtained.
To illustrate the superiority of the process of the present invention, two examples are constructed for illustration.
In the two embodiments, in the aspect of model parameter selection, the extracted current sampling frequency is 128 points per cycle, the feature dimension extracted by the measurement network is 16, when cosine similarity is judged, a similarity threshold value is set to be 0.8, when the cosine similarity of the two features is greater than 0.8, the two features are considered to be similar, otherwise, the two features are not similar, and in the same way, when power matching is performed, a power relative difference threshold value is set to be 0.2.
In the aspect of measuring the indexes of the test results, F is selected1The score and accuracy (Acc) is measured by a common index, which is calculated as follows:
Figure BDA0003467101990000071
Figure BDA0003467101990000072
Figure BDA0003467101990000073
Figure BDA0003467101990000074
in the formula, TP, FP, TN, FN respectively refer to the number of true positive, false positive, true negative, and false negative, precision is accuracy, and recall is recall.
Example 1:
to illustrate the universal recognition capabilities of the present invention, 80% white data set samples were used as training set training feature extraction networks, and the tests were performed on the remaining samples, following the common V-I method: document [ 1]](De Baets L,Dhaene T,Deschrijver D,et al.VI-Based Appliance Classification Using Aggregated Power Consumption Data[C]//2018 IEEE International Conference on Smart Computing (SMARTCOMP). Taormina: IEEE,2018: 179-186.), and document [2 ]](Wangyang, Yangwei, Xiaoxiaozongshu, Zhangshu. non-invasive resident load monitoring method based on U-I trajectory curve refined identification [ J]Electric network technology, 2021,45(10):4104-](De Baets L,Ruyssinck J,Develder C,et al.Appliance Classification Using VI Trajectories and Convolutional Neural Networks[J]Energy and Buildings,2018,158: 32-36.) using F1The scores were used as performance indicators, and the results are shown in Table 1.
TABLE 1 general identification Capacity comparison
Figure BDA0003467101990000081
It can be seen that the method has better identification capability than the existing V-I method (including V-I track and power two-stage identification) by using current data and power data and combining a one-dimensional convolutional neural network, and has a simpler network structure and lower computation complexity.
Example 2:
in order to verify the universality and the small sample learning capability of the method, only a sample of house6 in the PLAID data set is used as a training set, and appliances in partial COOLL data sets are selected for testing.
The electrical appliances selected in the COOLL data set comprise 9 types of air conditioners, modems, chargers, travel chargers, drilling machines, fans 1, fans 2, soldering irons and dust collectors, 20 groups of information of each electrical appliance are collected to be used as test samples, and a confusion matrix of identification results is obtained and is shown in figure 6.
Due to signal noise or current fluctuation of the electric appliance in steady-state operation and the like, part of the electric appliances are identified as a plurality of numbers, as shown by marks in a confusion matrix, 2 numbers are respectively identified by the electric appliance 1 and the electric appliance 5, namely, the difference of the similar inner wave forms exceeds a threshold value set by a model, and some samples are identified as new equipment. However, the identification effect is ideal on the whole, the identification accuracy of each electric appliance is shown in table 2, the electric appliances identified as a plurality of numbers can be solved by mapping the numbers to the same type of electric appliances in the mapping and labeling process of the characteristic numbers set by the samples added into the characteristic library to the real load switching events, and the method has good robustness.
TABLE 2 Cross-dataset generalization Capacity test results
Figure BDA0003467101990000082
Figure BDA0003467101990000091
In summary, it can be seen from the embodiments that, compared with other methods based on V-I trajectories, the method of the present invention is better in general identification performance, and more importantly, the method of the present invention is still excellent in generalization capability across datasets and small sample learning test, only limited sample training is required (only load in house6 in the PLAID dataset is used as a training set), and multiple identification numbers can be mapped to the same electrical appliance in the later period to further improve the identification effect, and the method of the present invention has a higher practical value.
It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. This need not be, nor should all embodiments be exhaustive. And obvious variations or modifications of the invention may be made without departing from the scope of the invention.

Claims (5)

1. A non-intrusive load identification method based on metric learning is characterized by comprising the following steps:
step 1, monitoring power bus power data in real time by using an event detection algorithm, positioning the occurrence time of a load switching event, and differentiating steady-state data before and after the occurrence time of the load switching event to separate out voltage and current and power data of switching of a load to be identified;
step 2, starting information acquisition of a complete period of the current data at the voltage positive zero crossing point, and performing resampling and normalization processing on the acquired current information;
step 3, using the trained feature extraction network to extract the features of the current to obtain the feature vector information of the load switching event to be identified;
step 4, checking whether the feature library exists or not, if not, establishing the feature library, storing feature vector information and power data of the load switching event to be identified and setting a corresponding feature number;
if the load switching event is the same as the sample in the feature library, the feature number of the sample in the feature library is used as the feature number of the load switching event to be identified; and if all the characteristic vector similarity conditions or the power similarity conditions are not met, the load switching event to be identified is a new event, the characteristic vector information and the power data of the load switching event to be identified are stored in a characteristic library, and corresponding characteristic numbers are set.
And 5, mapping the obtained feature numbers into actual load switching events, wherein the load switching events comprise load types and load state conversion information.
The feature number of the sample setting newly added into the feature library is mapped with the load switching event by the following method:
and when a sample newly added into the feature library exists, informing a user, and judging the actual load type of the newly added number and the state conversion information of the load by the user according to the historical switching record and the current-day actual use condition.
2. The non-invasive load identification method based on metric learning as claimed in claim 1, wherein the feature extraction network is composed of a plurality of one-dimensional convolution layers, an activation function layer, a plurality of residual modules, a global average pooling layer and a linear layer, and an L2 regularization layer, which are connected in sequence.
3. The non-intrusive load identification method based on metric learning as claimed in claim 1, wherein the feature extraction network training uses triplet loss as network loss.
4. The non-intrusive load identification method based on metric learning as claimed in claim 1, wherein in the feature extraction network training, training samples used in the training process regard different working states of the same electrical appliance as separate devices.
5. The non-invasive load identification method based on metric learning as claimed in claim 1, wherein in the step 4, the cosine similarity calculation method is used to calculate the similarity between the feature vector of each sample in the feature library and the feature vector of the load to be identified, and the difference calculation method is used to calculate the similarity between the power of the sample and the power of the load to be identified.
CN202210032592.XA 2022-01-12 2022-01-12 Non-invasive load identification method based on metric learning Pending CN114359674A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210032592.XA CN114359674A (en) 2022-01-12 2022-01-12 Non-invasive load identification method based on metric learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210032592.XA CN114359674A (en) 2022-01-12 2022-01-12 Non-invasive load identification method based on metric learning

Publications (1)

Publication Number Publication Date
CN114359674A true CN114359674A (en) 2022-04-15

Family

ID=81108491

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210032592.XA Pending CN114359674A (en) 2022-01-12 2022-01-12 Non-invasive load identification method based on metric learning

Country Status (1)

Country Link
CN (1) CN114359674A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662624A (en) * 2022-05-25 2022-06-24 浙江大学 Non-intrusive load identification method and system based on retraining twin network
CN116108350A (en) * 2023-01-06 2023-05-12 中南大学 Non-invasive electrical appliance identification method and system based on multitasking learning
CN117496243A (en) * 2023-11-06 2024-02-02 南宁师范大学 Small sample classification method and system based on contrast learning
CN117496243B (en) * 2023-11-06 2024-05-31 南宁师范大学 Small sample classification method and system based on contrast learning

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114662624A (en) * 2022-05-25 2022-06-24 浙江大学 Non-intrusive load identification method and system based on retraining twin network
CN114662624B (en) * 2022-05-25 2022-09-13 浙江大学 Non-invasive load identification method and system based on retraining twin network
CN116108350A (en) * 2023-01-06 2023-05-12 中南大学 Non-invasive electrical appliance identification method and system based on multitasking learning
CN116108350B (en) * 2023-01-06 2023-10-20 中南大学 Non-invasive electrical appliance identification method and system based on multitasking learning
CN117496243A (en) * 2023-11-06 2024-02-02 南宁师范大学 Small sample classification method and system based on contrast learning
CN117496243B (en) * 2023-11-06 2024-05-31 南宁师范大学 Small sample classification method and system based on contrast learning

Similar Documents

Publication Publication Date Title
WO2022110557A1 (en) Method and device for diagnosing user-transformer relationship anomaly in transformer area
CN110443302B (en) Load identification method based on feature fusion and deep learning and application thereof
CN107491792B (en) Power grid fault classification method based on feature mapping transfer learning
CN114359674A (en) Non-invasive load identification method based on metric learning
CN109829497B (en) Supervised learning-based station area user identification and discrimination method
CN109492667A (en) A kind of feature selecting discrimination method for non-intrusive electrical load monitoring
CN111830347B (en) Two-stage non-invasive load monitoring method based on event
CN109165604A (en) The recognition methods of non-intrusion type load and its test macro based on coorinated training
CN109828184B (en) Voltage sag source identification method based on mutual approximate entropy
CN113255236B (en) Non-invasive load self-adaptive identification method based on twin network
CN109858522A (en) A kind of management line loss abnormality recognition method based on data mining
CN111628494B (en) Low-voltage distribution network topology identification method and system based on logistic regression method
CN110569876A (en) Non-invasive load identification method and device and computing equipment
CN113191253A (en) Non-invasive load identification method based on feature fusion under edge machine learning
CN111275069B (en) Non-invasive load monitoring method
CN113659564B (en) Low-voltage distribution network topology identification method and system based on voltage fluctuation feature clustering
Kumar et al. Accurate fault diagnosis and classification scheme based on non-parametric, statistical-frequency features and neural networks
Xia et al. A Switch State Recognition Method based on Improved VGG19 network
Paul et al. An intelligent system for domestic appliance identification using deep dense 1-D convolutional neural network
CN109000924B (en) Method for monitoring state of ball screw pair based on K mean value
CN116486232A (en) Load identification method and system based on deep learning theory
CN115081933B (en) Low-voltage user topology construction method and system based on improved spectral clustering
CN114841266A (en) Voltage sag identification method based on triple prototype network under small sample
CN114004285A (en) Non-invasive load identification method based on improved kNN algorithm
He et al. Application and Comparative Analysis of Traditional Machine Learning and Deep Learning in Transmission Line Fault Classification

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