CN111639586B - Non-invasive load identification model construction method, load identification method and system - Google Patents
Non-invasive load identification model construction method, load identification method and system Download PDFInfo
- Publication number
- CN111639586B CN111639586B CN202010460278.2A CN202010460278A CN111639586B CN 111639586 B CN111639586 B CN 111639586B CN 202010460278 A CN202010460278 A CN 202010460278A CN 111639586 B CN111639586 B CN 111639586B
- Authority
- CN
- China
- Prior art keywords
- similarity
- load identification
- module
- loss
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a non-invasive load identification model construction method, a load identification method and a system, belonging to the field of non-invasive load identification and comprising the following steps: establishing a model to be trained, which comprises a feature extraction module, a classifier module and a similarity measurement module; forming a training set by electric appliance samples of a plurality of families; selecting h families from the training set, selecting k samples from each family, and inputting the samples into a feature extraction module to obtain the features of the electric appliance; inputting the electric appliance characteristics into a classifier module to convert the electric appliance characteristics into class labels, and calculating class loss; inputting the electrical appliance characteristics into a similarity measurement module, correspondingly splicing the electrical appliance characteristics of every two families, and calculating the similarity loss; updating the model parameters to minimize the overall loss by taking the sum of the two losses as the overall loss; after repeated training for multiple rounds, the load identification model is formed by the feature extraction module and the classifier module. The invention can construct a load identification model with domain generalization capability and improve the accuracy of non-invasive load identification.
Description
Technical Field
The invention belongs to the field of non-invasive load identification, and particularly relates to a non-invasive load identification model construction method, a load identification method and a system.
Background
The intelligent power grid and related applications are being developed and deployed in various countries in the world, and the advantages of the intelligent power grid can be exerted to the greatest extent by using data collected by the intelligent electric meters. The data collected by the intelligent electric meter can be used for identifying the type of the electric appliance in the family of the user, namely, the load identification is realized.
The load recognition method can be classified into two types, invasive and non-invasive, from the number of sensors. Intrusive load identification requires the installation of a corresponding sensor for each appliance, and additional equipment and higher cost make the intrusive method difficult to popularize. The non-intrusive load identification only needs to collect data from a single bus intelligent electric meter installed in a family, and decomposes the total household power consumption into the energy consumption of a single electric appliance through data analysis of the intelligent electric meter, so that the feedback of the power consumption condition is facilitated, a user is helped to save energy, and meanwhile, the accurate charging of a supply side is facilitated. Compared with invasive load identification, the non-invasive load identification has low cost and is easy to popularize, so that the method is widely researched.
The non-intrusive load identification method usually needs to identify the type of an electric appliance by means of a model, in practical application occasions, the electric appliances of different families have large differences due to user behaviors, electric appliance models and the like, and a labeled data set used by a training model is difficult to cover all possible electric appliance samples from different families, so different application scenes are constructed in different family environments. In addition, common machine learning algorithms and deep learning models are easy to be over-fitted to training set data, so how to acquire a load identification model which is cross-application scenes and is generalized and universal is a non-intrusive load identification difficulty.
Disclosure of Invention
Aiming at the defects and the improvement requirements of the prior art, the invention provides a non-invasive load identification model construction method, a load identification method and a system, and aims to construct a load identification model which is cross-application scene and generalized and universal so as to improve the identification precision of non-invasive load identification.
To achieve the above object, according to one aspect of the present invention, there is provided a non-intrusive load identification model construction method with domain generalization capability, including:
establishing a model to be trained based on the deep neural network, wherein the model to be trained comprises: the device comprises a feature extraction module, a classifier module and a similarity measurement module;
intercepting current signal data containing an electric appliance switching event from historical current signal data collected at a bus, converting the current signal data into a three-dimensional complex spectrogram, and forming an electric appliance sample by the three-dimensional complex spectrogram and a corresponding electric appliance category; forming a training set by electric appliance samples of a plurality of families;
randomly selecting h families from the training set, and randomly selecting k families from the electric appliance samples of each family as training samples of the current round; inputting the three-dimensional complex frequency spectrograms in all the training samples into a feature extraction module for feature extraction to obtain corresponding electrical appliance features; inputting the obtained electric appliance characteristics into a classifier module, converting the electric appliance characteristics into corresponding class labels, and calculating class loss by combining real electric appliance classes; inputting the obtained electric appliance characteristics into a similarity measurement module, splicing the electric appliance characteristics of every two families, forming a splicing characteristic by the electric appliance characteristics of the two different families, calculating a similarity score between the two electric appliance characteristics in each splicing characteristic, and calculating a similarity loss by combining real electric appliance categories; taking the sum of the category loss and the similarity loss as the overall loss of the current round, and updating parameters of each layer of the model to be trained by using a back propagation algorithm to minimize the overall loss, thereby completing the training of the current round;
performing multi-round training on the model to be trained, and forming a load identification model by using the trained feature extraction module and the classifier module after the training is finished;
wherein, the similarity score is used for measuring the similarity between the two electric appliance characteristics; h is more than or equal to 2, and k is more than or equal to 1.
In the model training process, the established model to be trained is provided with a similarity measurement module which can be used for extracting similarity scores among the characteristics of electrical appliances belonging to different families besides a characteristic extraction module and a classifier module, and similarity loss can be further calculated based on the similarity scores; in the training process, the sum of the similarity loss and the category loss is used as the overall loss, and the distances between the characteristics of the similar electrical appliances in different user domains can be shortened by minimizing the overall loss on the basis of ensuring that the identification result of the load category is as close to the real category as possible, so that the similar electrical appliances belonging to different families are similar or even identical in a characteristic space, and therefore when the established load identification model is used for carrying out load identification on any new sample, the load type corresponding to the new sample can be accurately identified on the basis of the similar old sample appearing in the training process, the user domain difference caused by different families can be effectively eliminated, and the user domain generalization is realized. Therefore, the load identification model constructed by the invention has domain generalization capability, and when the load identification model constructed by the invention is used for carrying out non-invasive load identification, the influence caused by user domain difference can be eliminated, and the accuracy of the non-invasive load identification is effectively improved.
Further, the method for constructing the non-invasive load identification model with the domain generalization capability provided by the invention further comprises the following steps:
(S1) after the load identification model is established, selecting 1 or more household electrical appliance samples to form a test set, wherein the test set and the training set are not overlapped with each other;
(S2) testing the load identification model by using the electric appliance samples in the test set to evaluate whether the identification precision of the load identification model meets the application requirement, if not, turning to the step (S3); otherwise, go to step (S4);
(S3) training the model to be trained again after adjusting the training parameters, and after the training is finished, reconstructing a load recognition model by the feature extraction module and the classifier module, and turning to the step (S2);
(S4) the test is ended.
After the load identification model is constructed, the invention further utilizes the electric appliance samples in different user domains to form a test set, evaluates the training effect of the load identification model, and carries out model training again when the identification precision of the model does not meet the application requirement, thereby further ensuring that the constructed load identification model has better generalization capability and further ensuring the accuracy of load identification.
Further, h is 2.
In the model training process, the electrical appliance samples of two families are randomly selected from the training set each time to serve as training samples, so that when the similarity measurement module carries out feature splicing, the samples of the two families only need to be correspondingly spliced once without considering other combination conditions, and the problems that the calculated amount is too large and even the calculated amount exceeds the hardware calculating capacity due to too many training samples in one round can be avoided while the model has strong domain generalization capacity.
Further, when similarity loss is calculated, taking the similarity score between two electrical appliance features in the splicing features and the binary cross entropy of the similarity label of the splicing features as the similarity loss of a single splicing feature, and taking the average value of the similarity losses of all the splicing features in the current round as the similarity loss of the current round;
the lower the similarity score is, the higher the similarity between the two electrical appliance characteristics in the splicing characteristics is; the similarity label of the stitching feature is used to indicate whether the genres of the two appliances involved in the stitching feature are the same.
The method takes the binary cross entropy of the similarity score and the similarity label as the similarity loss of a single splicing feature, and takes the average value of the similarity loss of the splicing feature as the similarity loss of the current turn, so that the overall training loss is continuously reduced in the training process, the similarity loss of each turn is also continuously reduced, the similarity loss of the single splicing feature is also continuously reduced, and finally the uniform feature expression of the same type of electric appliance samples from different families is realized, the user domain difference caused by different families can be effectively eliminated, and the user domain generalization is realized.
Further, the similarity loss of a single splice feature is:
wherein l1Representing a loss of similarity of individual stitching features; y is a similarity label of the splicing characteristics, when the real categories of the two electric appliances related to the splicing characteristics are the same, y is 0, when the real categories of the two electric appliances related to the splicing characteristics are different, y is 1;
a similarity score between two electrical features in the stitched feature.
Further, the class penalty is a cross-entropy penalty.
Further, current signal data containing electric appliance switching events are converted into a three-dimensional complex frequency spectrogram through short-time Fourier transform.
According to another aspect of the present invention, there is provided a non-intrusive load identification method with domain generalization capability, comprising: monitoring a target load to be identified in real time, and intercepting current signal data containing a switching event from current signal data collected from a bus when the switching event of the target load is detected;
and converting the intercepted current signal data into a three-dimensional complex spectrogram, taking the three-dimensional complex spectrogram as input, and obtaining the type of the target load identified by the load identification model by using the non-invasive load identification model construction method with the domain generalization capability provided by the invention.
Because the load identification model constructed by the invention has domain generalization capability, the invention utilizes the model to carry out non-invasive load identification, and can effectively improve the identification precision.
According to yet another aspect of the present invention, there is provided a non-intrusive load identification system with domain generalization capability, comprising: the system comprises a monitoring module, a conversion module and a load identification module;
the monitoring module is used for monitoring a target load to be identified in real time, intercepting current signal data containing a switching event from current signal data collected from a bus when the switching event of the target load is detected, and triggering the conversion module;
the conversion module is used for converting the intercepted current signal data into a three-dimensional complex frequency spectrogram and triggering the load identification module;
and the load identification module is used for obtaining the type of the target load identified by the load identification model by taking the three-dimensional complex spectrogram output by the conversion module as input and utilizing the non-invasive load identification model construction method with the domain generalization capability provided by the invention.
Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained:
(1) according to the invention, the similarity measurement module for calculating the similarity scores between the electric appliance characteristics belonging to different families is introduced into the model to be trained, and the sum of the similarity loss and the category loss is taken as the overall loss in the training process, so that the uniform characteristic expression of the same type of electric appliance samples from different families can be realized, the user domain difference caused by different families is effectively eliminated, and the user domain generalization is realized. Therefore, the load identification model constructed by the invention has domain generalization capability, and when the load identification model constructed by the invention is used for carrying out non-invasive load identification, the influence caused by user domain difference can be eliminated, and the accuracy of the non-invasive load identification is effectively improved.
(2) The invention can monitor the load switching event to be identified in real time, and only needs to intercept a section of current signal data containing the switching event of the electrical appliance during identification, and the feedback is timely and the accuracy is high.
(3) The invention utilizes the idea of domain generalization, the training set and the testing set adopt the electric appliance samples of different families, and the similar electric appliance samples of each family in the training set are enabled to obtain uniform characteristic expression, the user domain characteristics of the electric appliance samples are weakened, the generalization capability of the model is enhanced, and when the method is applied in a new user scene, the method still can keep better precision without retraining.
Drawings
FIG. 1 is a flowchart of a method for constructing a non-intrusive load identification model with domain generalization capability according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a model to be trained according to an embodiment of the present invention;
fig. 3 is a flowchart of a non-intrusive load identification method with domain generalization capability according to an embodiment of the present invention.
Detailed Description
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 and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In the present application, the terms "first," "second," and the like (if any) in the description and the drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
In order to construct a non-invasive load identification model with domain generalization capability, in an embodiment of the present invention, a method for constructing a non-invasive load identification model with domain generalization capability is provided, as shown in fig. 1, including:
establishing a model to be trained based on the deep neural network, as shown in fig. 2, the model to be trained includes: the device comprises a feature extraction module, a classifier module and a similarity measurement module; the characteristic extraction module is used for extracting the characteristics of the three-dimensional complex frequency spectrogram to obtain corresponding electrical appliance characteristics; the classifier module is used for converting the electric appliance features extracted by the feature extraction module into category labels of the electric appliances; the similarity measurement module is used for splicing the two electrical appliance characteristics belonging to different families to obtain splicing characteristics, and calculating a similarity score between the two electrical appliance characteristics in the splicing characteristics, wherein the similarity score is used for measuring the similarity between the two electrical appliance characteristics;
intercepting current signal data containing an electric appliance switching event from historical current signal data collected at a bus, converting the current signal data into a three-dimensional complex spectrogram, and forming an electric appliance sample by the three-dimensional complex spectrogram and a corresponding electric appliance category; forming a training set by electric appliance samples of a plurality of families;
randomly selecting h-2 families from the training set, randomly selecting k-32 electrical appliance samples from the electrical appliance samples of each family, and taking the k-64 electrical appliance samples as the training samples of the current round; inputting the three-dimensional complex frequency spectrograms in all the training samples into a feature extraction module for feature extraction to obtain corresponding electrical appliance features, and obtaining 64 electrical appliance features in total; inputting the obtained electric appliance characteristics into a classifier module, converting the electric appliance characteristics into corresponding class labels, and calculating class loss by combining real electric appliance classes; inputting the obtained electrical appliance characteristics into a similarity measurement module, splicing the electrical appliance characteristics of the two selected families, forming a splicing characteristic by the two electrical appliance characteristics belonging to different families, specifically, correspondingly splicing the 32 electrical appliance characteristics from the first family and the 32 electrical appliance characteristics from the second family to obtain 32 splicing characteristics, calculating similarity scores between the two electrical appliance characteristics in each splicing characteristic to obtain 32 similarity scores, and calculating similarity loss by combining real electrical appliance categories; taking the sum of the category loss and the similarity loss as the overall loss of the current round, and updating parameters of each layer of the model to be trained by using a back propagation algorithm to minimize the overall loss, thereby completing the training of the current round;
and (3) performing 30 rounds of training on the model to be trained, and after the training is finished, forming a load identification model by using the trained feature extraction module and the classifier module.
As shown in fig. 2, in this embodiment, the feature extraction module is a convolutional neural network DenseNet-121, and the similarity measurement module includes: 3 cascaded full connection layers FC (1) -FC (3), wherein the classifier module comprises 1 full connection layer FC (4); it should be noted that this is only one alternative embodiment of the present invention and should not be construed as the only limitation of the present invention; in other embodiments of the present invention, the feature extraction module may also be replaced by other models that can perform feature extraction, the number of cascaded fully-connected layers in the similarity measurement module may also be set as other parameters, and even other embodiments may be used to implement the similarity measurement module, and similarly, the classifier module may also be implemented by multiple cascaded fully-connected layers, or other embodiments;
in this embodiment, the activation function of the last fully-connected layer in the similarity measurement module is a sigmoid function, the activation function of the last fully-connected layer in the classifier module is a softmax function, and the activation functions of the remaining fully-connected layers are all ReLU functions; likewise, the present invention is only an alternative embodiment and should not be construed as the only limitation of the present invention.
As shown in fig. 1, the method for constructing a non-intrusive load identification model with domain generalization capability according to this embodiment further includes:
(S1) after the load identification model is established, selecting 1 or more household electrical appliance samples to form a test set, wherein the test set and the training set are not overlapped with each other;
(S2) testing the load identification model by using the electric appliance samples in the test set to evaluate whether the identification precision of the load identification model meets the application requirement, if not, turning to the step (S3); otherwise, go to step (S4);
(S3) training the model to be trained again after adjusting the training parameters, and after the training is finished, reconstructing a load recognition model by the feature extraction module and the classifier module, and turning to the step (S2);
(S4) the test is ended.
In the embodiment, after the load identification model is constructed, the electric appliance samples in different user domains are further utilized to form a test set, the training effect of the load identification model is evaluated, and the model training is carried out again when the identification precision of the model does not meet the application requirement, so that the established load identification model can be further ensured to have better generalization capability, and the accuracy of load identification is ensured.
In this embodiment, h is 2, that is, the electrical appliance samples of two families are randomly selected from the training set each time as the training samples of the current round, so that when the similarity measurement module performs feature splicing, the samples of two groups of families only need to be correspondingly spliced once, compared with other values, for example, when h is 3, two families are combined in pairs and need to be correspondingly spliced 3 times in total, and when h is 4, two families are combined in pairs and need to be correspondingly spliced 6 times in total, this embodiment can effectively reduce the number of the training samples of each round, thereby reducing the amount of calculation, and experimental data indicates that only the electrical appliance samples of two families are selected as the training samples in each round, and the finally obtained load identification model has a strong domain generalization capability. Therefore, the embodiment can avoid the problem that the calculation amount is too large and even exceeds the hardware calculation capacity due to too many training samples in one round while ensuring that the model has stronger domain generalization capacity;
it should be noted that h-2 is only one preferred embodiment of the present invention and should not be construed as the only limitation of the present invention; in other application scenarios, if there is a higher requirement for the domain generalization capability of the model, h can also be set to a larger value accordingly.
In this embodiment, when the similarity loss is calculated, the similarity score between two electrical appliance features in the splicing features and the binary cross entropy of the similarity label of the splicing features are used as the similarity loss of a single splicing feature, and the average value of the similarity losses of all the splicing features in the current round is used as the similarity loss of the current round;
the lower the similarity score is, the higher the similarity between the two electrical appliance features in the splicing feature is, and the similarity score can also be understood as the distance between the two electrical appliance features after normalization, and the value of the similarity score is between 0 and 1; the similarity label of the splicing characteristics is used for indicating whether the real categories of the two electric appliances related to the splicing characteristics are the same or not;
specifically, the formula for calculating the similarity loss of a single stitching feature is:
wherein l1Representing a loss of similarity of individual stitching features; y is a similarity label of the splicing characteristics, when the real categories of the two electric appliances related to the splicing characteristics are the same, y is 0, when the real categories of the two electric appliances related to the splicing characteristics are different, y is 1;
scoring a similarity score between two appliance characteristics in the stitching characteristics;
in other embodiments of the present invention, the single splicing feature and the overall similarity loss may also be calculated according to other manners such as a mean square error;
in this embodiment, the category loss is cross entropy loss, and the corresponding calculation formula is:
wherein l2Indicates class loss, C is the number of classes, yiThe ith bit of the C-length class label vector after the coding of the class label one-hot code, only the class position of the C-bit vector in the coding result of the one-hot code is 1, other bits are 0, z isiFor the ith bit of the vector with the length of C output by the classifier module, each bit of the vector output by the classifier module represents a probability value belonging to a corresponding category;
in this embodiment, the intercepted current signal data including the switching event of the electrical appliance have the same length (for example, all the current signal data are 7s), and in the three-dimensional complex spectrogram obtained through conversion, three dimensions are respectively a frequency dimension, a time dimension, a real part dimension and an imaginary part dimension; optionally, in this embodiment, the current signal data containing the appliance switching event is converted into a three-dimensional complex spectrogram by short-time fourier transform.
In summary, in the embodiment, a similarity measurement module for calculating similarity scores between electrical appliance features belonging to different families is introduced into a model to be trained, and in the training process, the sum of similarity loss and category loss is taken as the overall loss, so that the uniform feature expression of similar electrical appliance samples from different families can be obtained, the user domain difference caused by different families is effectively eliminated, and the user domain generalization is realized.
In another embodiment of the present invention, a non-intrusive load identification method with domain generalization capability is provided, as shown in fig. 3, including:
obtaining power data samples: monitoring a target load to be identified in real time, and intercepting current signal data containing a switching event from current signal data collected from a bus when the switching event of the target load is detected;
converting the sample into a three-dimensional complex spectrogram: converting the intercepted current signal data into a three-dimensional complex spectrogram;
detecting the target load category to be identified: the three-dimensional complex spectrogram is used as input, and the type of the target load identified by the load identification model is obtained by using the non-invasive load identification model construction method with the domain generalization capability provided by the embodiment.
Because the load identification model constructed in the above embodiment has domain generalization capability, the present embodiment uses the model to perform non-invasive load identification, and can effectively improve identification accuracy.
In yet another embodiment of the present invention, a non-intrusive load identification system with domain generalization capability is provided, comprising: the system comprises a monitoring module, a conversion module and a load identification module;
the monitoring module is used for monitoring a target load to be identified in real time, intercepting current signal data containing a switching event from current signal data collected from a bus when the switching event of the target load is detected, and triggering the conversion module;
the conversion module is used for converting the intercepted current signal data into a three-dimensional complex frequency spectrogram and triggering the load identification module;
and the load identification module is used for obtaining the type of the target load identified by the load identification model by taking the three-dimensional complex spectrogram output by the conversion module as input and utilizing the non-invasive load identification model construction method with the domain generalization capability provided by the embodiment.
The technical solution and the beneficial effects obtained by the present invention are further described below with reference to a specific application example:
a public household electrical appliance data set UK-DALE (UK domestic application-level electric utility dataset) is used, and power load data of 5 households (sequentially referred to as household 1 to household 5) is included in the data set, and 1/6Hz bus and branch table data are recorded. The households 1, 2 and 5 also contain 16kHz high-frequency bus data, and the current data of the 3 households are selected in order to draw a time spectrogram and apply the load identification model construction method and the load identification method provided in the embodiment. Meanwhile, in order to avoid the imbalance of the aspect ratio of the spectrogram caused by the overlong frequency axis, the data is down-sampled to 2 kHz.
In order to embody the effect of the invention on the network generalization capability, the data of families 1 and 5 are used as a training set, the data of family 2 is used as a test set, 7-second current data containing the state change of an electric appliance is intercepted as a sample, meanwhile, the current data is subjected to short-time Fourier transform to be processed into a plurality of time spectrograms, the size of the obtained time spectrogram is 224 multiplied by 100 multiplied by 2, three dimensions respectively represent frequency, time and real/imaginary parts, and the time spectrograms are used as model input.
Selecting 7 electric appliances which are common to 3 families in a data set UK-DALE, wherein the time is as follows: kettles, refrigerators, dishwashers, microwave ovens, washing machines, computers, and treadmills.
The load identification model shown in fig. 2 is established, the similarity measurement module comprises full connection layers FC (1), FC (2) and FC (3), and the output dimensions are 1024, 64 and 1 respectively. The classifier module contains 1 full connection layer FC (4), and the output dimension is 7. The activation functions of FC (1) and FC (2) are ReLU, FC (3) and FC (4) are sigmoid and softmax respectively. In the training process, an Adam optimizer is used, the learning rate is 0.001, the classification loss is multi-class cross entropy, the similarity loss is binary cross entropy, and the total iteration number is 1500.
And (3) performing model training by using the samples of the family 1 and the family 5, and testing the sample of the family 2 after the training is finished. The results were measured using an F1 score, F1 score (F1-score) being the harmonic mean of precision and recall, calculated as follows:
where TP represents the number of events for which the sample was correctly identified, FN represents the number of events for which the sample was of the category but not identified as the category, and FP represents the number of events for which the sample was identified as the category but not the category.
TABLE 1 recognition Effect of load recognition models on UK-DALE datasets
The load identification results are shown in table 1, and it can be known from the results that the non-intrusive load identification method with domain generalization capability can make the model have better generalization capability and perform well in the scene across users, especially for the appliances with large domain difference in the two multimode modes of washing machine and dishwasher.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (9)
1. A non-invasive load identification model construction method with domain generalization capability is characterized by comprising the following steps:
establishing a model to be trained based on a deep neural network, wherein the model to be trained comprises: the device comprises a feature extraction module, a classifier module and a similarity measurement module;
intercepting current signal data containing an electric appliance switching event from historical current signal data collected at a bus, converting the current signal data into a three-dimensional complex spectrogram, and forming an electric appliance sample by the three-dimensional complex spectrogram and a corresponding electric appliance category; forming a training set by electric appliance samples of a plurality of families; in the three-dimensional complex frequency spectrogram, three dimensions are respectively a frequency dimension, a time dimension, a real part dimension and an imaginary part dimension;
randomly selecting h families from the training set, and randomly selecting k families from the electric appliance samples of each family as training samples of the current round; inputting the three-dimensional complex frequency spectrograms in all the training samples into the feature extraction module for feature extraction to obtain corresponding electrical appliance features; inputting the obtained electric appliance characteristics into the classifier module, converting the electric appliance characteristics into corresponding category labels, and calculating category loss by combining real electric appliance categories; inputting the obtained electrical appliance characteristics into the similarity measurement module, splicing the electrical appliance characteristics of every two families, forming a splicing characteristic by the two electrical appliance characteristics belonging to different families, calculating a similarity score between the two electrical appliance characteristics in each splicing characteristic, and calculating a similarity loss by combining real electrical appliance categories; taking the sum of the category loss and the similarity loss as the overall loss of the current round, and updating parameters of each layer of the model to be trained by using a back propagation algorithm to minimize the overall loss so as to finish the training of the current round;
performing multi-round training on the model to be trained, and after the training is finished, forming a load identification model by using the trained feature extraction module and the classifier module;
wherein, the similarity score is used for measuring the similarity between the two electric appliance characteristics; h is more than or equal to 2, and k is more than or equal to 1.
2. The method for constructing a non-invasive load recognition model with domain generalization capability according to claim 1, further comprising:
(S1) after the load recognition model is established, selecting 1 or more household electrical appliance samples to form a test set, wherein the test set and the training set are not overlapped with each other;
(S2) testing the load identification model by using the electric appliance samples in the test set to evaluate whether the identification precision of the load identification model meets the application requirement, and if not, turning to the step (S3); otherwise, go to step (S4);
(S3) regulating training parameters, then retraining the model to be trained, after training, reconstructing a load recognition model by the feature extraction module and the classifier module, and turning to the step (S2);
(S4) the test is ended.
3. The method for constructing a non-invasive load identification model with domain generalization capability according to claim 1 or 2, wherein h is 2.
4. The method for constructing the non-invasive load identification model with the domain generalization capability according to claim 1 or 2, wherein when the similarity loss is calculated, the similarity loss of a single splicing feature is determined by using the similarity score between two electrical appliance features in the splicing feature and the binary cross entropy of the similarity label of the splicing feature, and the average value of the similarity losses of all the splicing features in the current round is determined as the similarity loss in the current round;
the lower the similarity score is, the higher the similarity between the two electrical appliance characteristics in the splicing characteristics is; the similarity label of the stitching feature is used to indicate whether the genres of the two appliances involved in the stitching feature are the same.
5. The method for constructing a non-invasive load recognition model with domain generalization capability according to claim 4, wherein the similarity loss of a single splicing feature is:
wherein l1Representing a loss of similarity of individual stitching features; y is a similarity label of the splicing characteristics, when the real categories of the two electric appliances related to the splicing characteristics are the same, y is 0, when the real categories of the two electric appliances related to the splicing characteristics are different, y is 1;
a similarity score between two electrical features in the stitched feature.
6. The method for constructing a non-invasive load recognition model with domain generalization capability according to claim 1 or 2, wherein the class loss is a cross-entropy loss.
7. The method for constructing a non-invasive load identification model with domain generalization capability according to claim 1 or 2, wherein the current signal data containing the appliance switching event is converted into a three-dimensional complex spectrogram by short-time Fourier transform.
8. A method for non-intrusive load identification with domain generalization capability, comprising: monitoring a target load to be identified in real time, and intercepting current signal data containing a switching event from current signal data collected from a bus when the switching event of the target load is detected;
converting the intercepted current signal data into a three-dimensional complex spectrogram, taking the three-dimensional complex spectrogram as an input, and identifying the type of the target load by using the load identification model obtained by the non-invasive load identification model construction method with the domain generalization capability of any one of claims 1 to 7.
9. A non-intrusive load identification system with domain generalization capability, comprising: the system comprises a monitoring module, a conversion module and a load identification module;
the monitoring module is used for monitoring a target load to be identified in real time, intercepting current signal data containing a switching event from current signal data collected from a bus when the switching event of the target load is detected, and triggering the conversion module;
the conversion module is used for converting the intercepted current signal data into a three-dimensional complex frequency spectrogram and triggering the load identification module;
the load identification module is used for obtaining a load identification model by taking the three-dimensional complex spectrogram output by the conversion module as input and utilizing the non-invasive load identification model construction method with the domain generalization capability of any one of claims 1 to 7 to identify the type of the target load.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010460278.2A CN111639586B (en) | 2020-05-27 | 2020-05-27 | Non-invasive load identification model construction method, load identification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010460278.2A CN111639586B (en) | 2020-05-27 | 2020-05-27 | Non-invasive load identification model construction method, load identification method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111639586A CN111639586A (en) | 2020-09-08 |
| CN111639586B true CN111639586B (en) | 2022-03-18 |
Family
ID=72329056
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010460278.2A Active CN111639586B (en) | 2020-05-27 | 2020-05-27 | Non-invasive load identification model construction method, load identification method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111639586B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112180193B (en) * | 2020-09-28 | 2021-11-02 | 华中科技大学 | Non-invasive load identification system and method based on track image identification |
| CN113033821A (en) * | 2021-03-29 | 2021-06-25 | 上海思创电器设备有限公司 | Load identification machine learning system applied to algorithm core unit |
| CN113393121A (en) * | 2021-06-15 | 2021-09-14 | 贵州电网有限责任公司 | Non-invasive load identification method based on load power fingerprint characteristics |
| CN116610922A (en) * | 2023-07-13 | 2023-08-18 | 浙江大学滨江研究院 | Non-invasive load identification method and system based on multi-strategy learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8560134B1 (en) * | 2010-09-10 | 2013-10-15 | Kwangduk Douglas Lee | System and method for electric load recognition from centrally monitored power signal and its application to home energy management |
| CN106096726A (en) * | 2016-05-31 | 2016-11-09 | 华北电力大学 | A kind of non-intrusion type load monitoring method and device |
| CN108132379A (en) * | 2017-12-11 | 2018-06-08 | 武汉大学 | Non-intrusion type load monitor system and recognition methods based on cloud platform |
| CN109508908A (en) * | 2018-12-25 | 2019-03-22 | 深圳市城市公共安全技术研究院有限公司 | Non-invasive load recognition model training method, load monitoring method and device |
| CN110119545A (en) * | 2019-04-24 | 2019-08-13 | 西安理工大学 | A kind of non-intrusive electrical load recognition methods based on stack self-encoding encoder |
| KR102081912B1 (en) * | 2018-11-15 | 2020-02-26 | 광주과학기술원 | Operating method of building load identification system using mask estimation method based deep learning |
| CN110991263A (en) * | 2019-11-12 | 2020-04-10 | 华中科技大学 | Non-invasive load identification method and system for resisting background load interference |
-
2020
- 2020-05-27 CN CN202010460278.2A patent/CN111639586B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8560134B1 (en) * | 2010-09-10 | 2013-10-15 | Kwangduk Douglas Lee | System and method for electric load recognition from centrally monitored power signal and its application to home energy management |
| CN106096726A (en) * | 2016-05-31 | 2016-11-09 | 华北电力大学 | A kind of non-intrusion type load monitoring method and device |
| CN108132379A (en) * | 2017-12-11 | 2018-06-08 | 武汉大学 | Non-intrusion type load monitor system and recognition methods based on cloud platform |
| KR102081912B1 (en) * | 2018-11-15 | 2020-02-26 | 광주과학기술원 | Operating method of building load identification system using mask estimation method based deep learning |
| CN109508908A (en) * | 2018-12-25 | 2019-03-22 | 深圳市城市公共安全技术研究院有限公司 | Non-invasive load recognition model training method, load monitoring method and device |
| CN110119545A (en) * | 2019-04-24 | 2019-08-13 | 西安理工大学 | A kind of non-intrusive electrical load recognition methods based on stack self-encoding encoder |
| CN110991263A (en) * | 2019-11-12 | 2020-04-10 | 华中科技大学 | Non-invasive load identification method and system for resisting background load interference |
Non-Patent Citations (2)
| Title |
|---|
| "Household Load Forecasting Based on a Pre-Processing Non-Intrusive Load Monitoring Techniques";Ahmed F. Ebrahim 等;《2018 IEEE Green Technologies Conference (GreenTech)》;20180607;107-114 * |
| "场景自适应的非侵入式负荷识别算法研究";阿蓉;《中国优秀硕士学位论文全文数据库-工程科技Ⅱ辑》;20190615;第2019年卷(第6期);C042-383 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111639586A (en) | 2020-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111639586B (en) | Non-invasive load identification model construction method, load identification method and system | |
| CN109387712B (en) | Non-invasive load detection and decomposition method based on state matrix decision tree | |
| CN108616120B (en) | Non-invasive power load decomposition method based on RBF neural network | |
| Chen et al. | Temporal and spectral feature learning with two-stream convolutional neural networks for appliance recognition in NILM | |
| CN111160798B (en) | Non-invasive household appliance load identification method based on bee colony algorithm | |
| CN111368904B (en) | Electrical equipment identification method based on electric power fingerprint | |
| CN110991263B (en) | Non-invasive load identification method and system for resisting background load interference | |
| CN111027408A (en) | Load identification method based on support vector machine and V-I curve characteristics | |
| Ghosh et al. | An improved load feature extraction technique for smart Homes using fuzzy-based NILM | |
| CN111382789B (en) | Power load identification method and system based on machine learning | |
| CN109840691B (en) | Non-invasive polynomial electric quantity estimation method based on deep neural network | |
| CN109470957B (en) | Non-invasive electrical appliance identification method based on Frey's distance | |
| CN109359665A (en) | A kind of family's electric load recognition methods and device based on support vector machines | |
| CN112215405A (en) | Non-invasive type residential electricity load decomposition method based on DANN domain adaptive learning | |
| CN112598303A (en) | Non-invasive load decomposition method based on combination of 1D convolutional neural network and LSTM | |
| Saha et al. | Comprehensive NILM framework: device type classification and device activity status monitoring using capsule network | |
| CN116796271A (en) | Resident energy abnormality identification method | |
| Luo et al. | A generalizable method for practical non-intrusive load monitoring via metric-based meta-learning | |
| Dinesh et al. | Non-intrusive load monitoring based on low frequency active power measurements | |
| CN113010985A (en) | Non-invasive load identification method based on parallel AANN | |
| Cannas et al. | NILM techniques applied to a real-time monitoring system of the electricity consumption | |
| Herath et al. | Comprehensive analysis of convolutional neural network models for non-instructive load monitoring | |
| CN115859190A (en) | Non-invasive household electrical classification method based on causal relationship | |
| CN114239762A (en) | Non-invasive load identification method and system based on structured load characteristic spectrum | |
| CN113420436B (en) | GRU and ARIMA-T based non-invasive event detection method |
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 |