CN112989133B - Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment - Google Patents
Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment Download PDFInfo
- Publication number
- CN112989133B CN112989133B CN202110335453.XA CN202110335453A CN112989133B CN 112989133 B CN112989133 B CN 112989133B CN 202110335453 A CN202110335453 A CN 202110335453A CN 112989133 B CN112989133 B CN 112989133B
- Authority
- CN
- China
- Prior art keywords
- electrical
- power
- electrical equipment
- data
- fingerprint
- 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
Abstract
The invention provides a graph data modeling power fingerprint identification method, a storage medium and a system of electrical equipment, which comprise the following steps: detecting an input event of the electrical equipment in real time at the power utilization bus, and extracting a plurality of electrical operation characteristics of the input electrical equipment; establishing an incidence relation between every two of the plurality of electrical operation characteristics according to a mutual information entropy theory; taking the extracted electrical operation characteristics as nodes, taking the incidence relation between every two electrical operation characteristics as edges, and generating power fingerprint diagram data containing the nodes and the edges according to a diagram theory; graph representation learning is carried out on the power fingerprint graph data by using a graph convolution network, and a mapping relation between the power fingerprint graph data and the electrical equipment is constructed to realize load identification. The mapping relation is determined without a large amount of calculation in the load identification process, and the electrical operation characteristics of the electrical equipment can be directly obtained based on the power fingerprint image data, so that the non-invasive load identification algorithm is simple, the identification speed is high, and the operation efficiency is high.
Description
Technical Field
The invention relates to the technical field of informatization communication, in particular to a graph data modeling electric power fingerprint identification method, a storage medium and a system of electrical equipment.
Background
The ubiquitous power internet of things is an important development direction of a current smart grid, and the ubiquitous power internet of things is mainly used for promoting interactive intelligent power utilization of users and mining incremental value of power utilization big data, and how to acquire fine particle information of the power utilization of the users is a key point and a difficult point in the ubiquitous power internet of things. The advanced measurement system can realize the analysis and monitoring of the power utilization information of the user, particularly the load identification, and is beneficial to realizing the bidirectional information interaction between the user and the power grid.
Non-intrusive load monitoring (NILM) was first proposed by professor Hart in the 80 th 20 th century by installing a sensor at the customer bus to perform load decomposition based on the total electricity consumption of the customer to obtain the operating status of each load. Compared with the Intrusive Load Monitoring (ILM), the method has the advantages of low cost, easiness in installation, no relation to user privacy and the like, and therefore the method has a wider development prospect.
Currently, for NILM, researchers have studied and obtained certain results at home and abroad. The literature (Hart, G.W.1992, "non-systematic application Load monitoring." Proceedings of the IEEE 80 (12): 1870-1891.) proposes to detect the amount of change in active and reactive power for Load identification, which, although simple, is difficult to identify accurately for characteristic overlapping and multi-state loads; the method comprises the following steps of (Kolter J Z, jaakkola T. Adaptive induced behavior HMMs with application to energy differentiation [ C ]. La Palma, spain: microme Publishing, 2012.) constructing an additive factor hidden Markov model of load decomposition, and carrying out optimization solution by using a method approximate to maximum probability to obtain better load decomposition performance at that time; the literature (Lam, H.Y., G.S.K.Fung, and W.K.Lee.2007."A Novel Method to Construct Taxonomy electric applications Based on Load signatures." IEEE Transactions on Consumer Electronics 53 (2): 653-660.) proposes to use the V-I trajectory Method to differentiate loads using an index describing the characteristics of the trajectory, which, although achieving some success, is difficult to differentiate for small loads and difficult to convert the algorithm into programming language; the method comprises the following steps that (according to a document, a Venulon, a Ching, a non-invasive load decomposition method [ J ] based on a depth sequence translation model, a power grid technology 2020,44 (1): 27-34.) is combined with a sequence translation model theory, and a sequence translation model is used for constructing a mapping relation between a signal to be decomposed and a state code of an electric appliance; the application of RPROP neural network in non-invasive load decomposition [ J ] power system protection and control, 2016,44 (07): 55-61 ] proposes to use RPROP artificial neural network for load recognition, although good recognition effect is achieved in training set, its generalization capability is still considered. Although the method has a good effect, a mapping relation is not established between the electrical equipment and the electrical operation characteristic data of the electrical equipment, and the mapping relation can be determined only through a large amount of calculation in the load identification process, so that the algorithm structure is complex, and in the data processing process of the method, the electrical operation characteristics of the electrical equipment are required to be input into a non-invasive load identification algorithm in sequence to realize load identification, so that the non-invasive load identification algorithm is slow in identification speed and low in operation efficiency.
Disclosure of Invention
The invention aims to provide the electric power fingerprint identification method which is simple in algorithm, high in identification speed and high in operation efficiency.
In order to solve the technical problem, the invention provides a graph data modeling power fingerprint identification method for electrical equipment, which comprises the following steps:
step S1: detecting an input event of electrical equipment in real time at an electricity utilization bus, and if the input event of the electrical equipment is detected, extracting a plurality of electrical operation characteristics of the electrical equipment;
step S2: establishing an incidence relation between every two of the plurality of electrical operation characteristics according to a mutual information entropy theory;
and step S3: taking the extracted electrical operation characteristics as nodes, taking the incidence relation between every two electrical operation characteristics as edges, and generating electric power fingerprint graph data containing the nodes and the edges according to a graph theory;
and step S4: and constructing a load identification model based on a graph convolution network to carry out graph representation learning on the electric power fingerprint graph data, and constructing a mapping relation between the electric power fingerprint graph data and the electric equipment so as to realize load identification.
Preferably, in step S1, whether a commissioning event of an electrical appliance occurs is determined by detecting an accumulated sum of active power changes at the power bus.
Preferably, the specific detection method of the cumulative sum of active power changes at the power utilization bus is as follows:
defining a steady state detection window and a transient state detection window for active power at an electric bus, wherein the lengths of the steady state detection window and the transient state detection window are a and b respectively, and the steady state detection window and the transient state detection window jointly form an event detection window; respectively calculating the active power average value P of the steady-state part a And the active power average value P of the transient part b The method comprises the following steps:
then define the forward cumulative sum
And the negative cumulative sum
Wherein
For recording the accumulated course of the electrical equipment investment,
the method is used for recording the cumulative process of cutting off the electrical equipment and comprises the following specific steps:
wherein E r Representing the normal fluctuation level of active power when the electrical equipment operates;
then defining the cumulative sum of switching events of the electrical equipmentThreshold H, i.e. cumulative sum of positive changes in active power
When the threshold H is reached, it is considered that the event of switching the electric device has occurred at this time.
Preferably, in step S1, the electrical operation characteristics of the electrical equipment include active power, reactive power, current effective value, current 3-order harmonic amplitude, current 5-order harmonic amplitude, current 7-order harmonic amplitude, current harmonic total distortion rate, and instantaneous power peak value; the extraction of the plurality of electrical operation characteristics of the electrical equipment means that the data collected at the power utilization bus after the input event occurs and the same type of data collected at the power utilization bus before the input event occurs are subtracted, so that the electrical operation characteristics of the input electrical equipment are obtained.
In the step S2, the incidence relation between each two electrical operation features refers to correlation characteristics and irrelevance characteristics between each two extracted electrical operation features; the mutual entropy refers to information shared between two associated electrical operating characteristics, wherein if the information of one of the two associated electrical operating characteristics is obtained, the unpredictability of the other electrical operating characteristic is reduced.
The specific calculation formula of the mutual information entropy is as follows:
where p (x) and p (y) are probability density functions of the variables x and y, and p (x, y) is a joint probability density function of the two variables.
In the step S3, all the electrical operating characteristics extracted in the step S1 are defined as nodes in the electrical fingerprint map data, and the mutual entropy between every two electrical operating characteristics calculated in the step S2 is defined as edges in the electrical fingerprint map data, so as to construct the electrical fingerprint map data G (a, X) of the electrical equipment with unique identifier, where a is an adjacent matrix of the map, and X is a characteristic matrix of the nodes in the map.
For node features, the data is linearly transformed using maximum-minimum normalization, mapping the data to [0,1] intervals, which are defined as follows:
where max is the maximum value in the data; min is the minimum in the data;
for the side weight, a softmax function is used for carrying out normalization processing on the mutual information entropy, and the calculation formula of each side weight value is as follows:
wherein k (x) i ,x j ) Is the mutual information entropy of the edge between the node i and the node j.
Preferably, in the step S4, for each piece of power fingerprint data G (a, X) containing original feature information, a graph convolution layer is used to aggregate the information, and new power fingerprint data G (a, X') with more abstract and more identification is output, and the process is expressed by the following formula:
in the formula: theta is a weight matrix; deg (i) and deg (j) represent degrees of node i and node j; σ denotes the activation function.
Preferably, after the electric power fingerprint map data of the electric equipment is built, the electric power fingerprint map data of each electric equipment is integrated to build an electric power fingerprint database of the electric equipment, and the electric power fingerprint map data in the electric power fingerprint database is divided into a training set, a verification set and a test set according to the proportion of 3.
Preferably, in the step S4, a load identification model based on a graph convolution network is constructed, and the power fingerprint library of the electrical equipment established in the step S3 is used for training and learning the load identification model, the input of the load identification model is power fingerprint data constructed by electrical operation characteristics and mutual information entropy between every two electrical operation characteristics, and the output of the load identification model is the type of the electrical equipment corresponding to the power fingerprint data.
Preferably, in step S4, the load identification model is constructed specifically: firstly, a load identification model carries out feature transformation on input power fingerprint image data by using a feature transformation layer, secondly, at least three image convolution layers are stacked to extract structural features of the power fingerprint image data, secondly, a graph pooling layer is used to aggregate full image information, and finally, at least four full connection layers are stacked to realize nonlinear mapping between the power fingerprint image data and the type of electrical equipment, so that a mapping relation between the power fingerprint image data and the electrical equipment is constructed.
The invention also provides a computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method as described above.
The invention also provides a graph data modeling power fingerprint identification system of electrical equipment, which comprises a computer readable storage medium and a processor which are connected with each other, wherein the computer readable storage medium is as described above.
The invention has the following beneficial effects: after the occurrence of a switching event of the electrical equipment is detected, extracting a plurality of electrical operation characteristics capable of reflecting the starting state of the electrical equipment, establishing an incidence relation between every two electrical operation characteristics according to a mutual information entropy theory, taking the extracted electrical operation characteristics as nodes, taking the incidence relation between every two electrical operation characteristics as edges, generating electrical fingerprint graph data containing the nodes and the edges according to a graph theory, and establishing a mapping relation between the electrical fingerprint graph data and the electrical equipment after the electrical fingerprint graph data containing the electrical operation characteristics of the electrical equipment are formed, so that the mapping relation is determined without a large amount of calculation in a load identification process, and a non-invasive load identification algorithm is simple; because the electrical operation characteristics of the electrical equipment are contained in the electrical fingerprint image data, the electrical operation characteristics of the electrical equipment can be directly obtained based on the electrical fingerprint image data in the load identification process, so that the electrical fingerprint image data is beneficial to fully extracting the operation characteristics of the electrical equipment to realize load identification, the electrical operation characteristics of the electrical equipment do not need to be input into a non-invasive load identification algorithm in sequence to realize load identification, and the non-invasive load identification algorithm is high in identification speed and high in operation efficiency.
Drawings
Fig. 1 is a general flow chart of a power fingerprint load identification method.
Fig. 2 is a sliding window active power bilateral accumulation and event detection schematic.
Fig. 3 is a schematic diagram of electric power fingerprint map data construction of an electric appliance.
FIG. 4 is a schematic diagram of a GCN map illustrating the learning process.
FIG. 5 is a schematic diagram of a GCN-based load recognition model.
Detailed Description
The invention is described in further detail below with reference to specific embodiments.
In the ubiquitous power internet of things, a non-invasive load identification algorithm can be used for load decomposition so as to obtain the running state of a load, but the non-invasive load identification algorithm is complex, low in identification speed and low in running efficiency, and has the problem that the non-invasive load identification algorithm is difficult to apply to actual engineering. To this end, the present embodiment provides a graph data modeling power fingerprint identification system for an electrical device, the system includes a computer-readable storage medium and a processor, which are connected to each other, where the computer-readable storage medium stores a computer program, and in order to enable the system to achieve non-intrusive load decomposition and achieve an algorithm effect of fast, accurate, and efficient identification, when the computer program is executed by the processor, the method implements a graph data modeling power fingerprint identification method for an electrical device, as shown in fig. 1, the method includes the following steps:
step S1: the method comprises the steps of detecting an input event of the electrical equipment in real time at an electricity utilization bus, and if the input event of the electrical equipment is detected, extracting a plurality of electrical operation characteristic difference values of the electrical equipment.
In this embodiment, the sliding window active power bilateral accumulation sum algorithm is used to realize real-time detection of the input event of the electrical equipment at the power utilization bus, and the main idea of the algorithm is as follows: if the active power has obvious variation in real time detected at the power utilization bus, the running state of the electrical equipment of the power utilization bus is judged to be changed, namely, the switching-in event of the electrical equipment is judged to occur, so that whether the switching-in event of the electrical equipment occurs or not can be judged by detecting the accumulated sum of the active power variation at the power utilization bus. Under the condition that no input event of the electrical equipment occurs, a relatively stable active power sequence can be formed at the power utilization bus, and when the input event of the electrical equipment occurs, obvious change of the active power can be caused at the power utilization bus, so that the relatively stable active power sequence is damaged, and therefore, the obvious change quantity of the active power at the power utilization bus can be detected according to the accumulation of the change of the active power. The specific detection method comprises the following steps:
the active power sequence at the utility bus P = { P (t) }, t =1,2, \8230, where t is the sampling instant and P (t) is the active power at the instant t. As shown in fig. 2, a steady-state detection window and a transient-state detection window are defined for active power at an electricity-using bus, that is, an active power sequence is divided into a steady-state portion and a transient-state portion, the length of the steady-state detection window is a, the length of the transient-state detection window is b, and the steady-state detection window and the transient-state detection window together form an event detection window; then respectively calculating the active power average value P of the steady-state part a And the active power average value P of the transient part b The method comprises the following steps:
then define the forward cumulative sum
And negative cumulative sum
Wherein
Used for recording the accumulated process of the investment of the electrical equipment,
the cumulative process for recording the cutting of the electrical equipment is as follows:
wherein E r Representing the normal fluctuation level of active power when the electrical equipment operates;
then, a threshold value H for the cumulative sum of switching events of the electrical equipment is defined, namely, the cumulative sum of positive changes of active power
When the threshold H is reached, it is considered that the event of switching the electric device has occurred at this time. Wherein the conventional reference value of the threshold H is 32.5.
The electrical operation characteristics of the electrical equipment mainly comprise active power, reactive power, a current effective value, a current 3-order harmonic amplitude, a current 5-order harmonic amplitude, a current 7-order harmonic amplitude, a current harmonic total distortion rate, an instantaneous power peak value and the like; the method comprises the steps of extracting a plurality of electrical operation characteristics of the electrical equipment after detecting that an input event of the electrical equipment occurs, specifically subtracting data collected at an electricity utilization bus after the input event occurs from the same type of data collected at the electricity utilization bus before the input event occurs, and obtaining the electrical operation characteristics of the input electrical equipment.
Step S2: and establishing an incidence relation between every two electrical operation characteristics based on a mutual information entropy theory.
The incidence relation between every two of the plurality of electrical operation characteristics refers to the correlation characteristics and the irrelevance characteristics between every two of the plurality of electrical operation characteristics extracted in the step S1; mutual entropy, which refers to information shared between two electrical operating characteristics, reduces the unpredictability of one of the two associated electrical operating characteristics if the information of the other electrical operating characteristic is obtained. The specific calculation formula of the mutual information entropy is as follows:
where p (x) and p (y) are probability density functions of the variables x and y, respectively, and p (x, y) is a joint probability density function of the two variables x and y.
And step S3: and taking the extracted electrical operation characteristics as nodes, taking the incidence relation between every two electrical operation characteristics as edges, and generating electric power fingerprint graph data containing the nodes and the edges based on a graph theory.
Firstly, all the electrical operation characteristics extracted in step S1 are defined as nodes in the electrical fingerprint map data, and then the mutual information entropy between every two electrical operation characteristics calculated in step S2 is defined as edges in the electrical fingerprint map data, which means that all the electrical operation characteristics have an association relationship, but different electrical devices have different electrical operation characteristics, and the mutual information entropy between every two electrical operation characteristics has different differences in different electrical devices, so that the electrical fingerprint map data G (a, X) with unique identification can be constructed for the electrical devices, wherein a is an adjacent matrix of the map, and X is a characteristic matrix of the nodes in the map, as shown in fig. 3 in particular.
For node feature x * The data is mapped to [0,1] using a linear transformation of the data using maximum and minimum normalization]The intervals are specifically as follows:
where max is the maximum value in the data; min is the minimum in the data.
For edge weight W ij And carrying out normalization processing on the mutual information entropy by using a softmax function, wherein the calculation formula of each edge weight value is as follows:
wherein k (x) i ,x j ) Is the mutual information entropy of the edge between the node i and the node j.
After the electric power fingerprint image data of the electric equipment is built, the electric power fingerprint image data of each electric equipment are integrated to build an electric power fingerprint database of the electric equipment, the electric power fingerprint image data in the electric power fingerprint database are divided into a training set, a verification set and a test set according to the proportion of 1, and the training set, the verification set and the test set are used for training a load identification model.
And step S4: and constructing a load identification model based on a graph convolution network to carry out graph representation learning on the electric power fingerprint graph data, and constructing a mapping relation between the electric power fingerprint graph data and the electric equipment so as to realize load identification.
The raw features in the power fingerprint map data may be further combined and form more abstract high-level features, which is referred to as graph representation learning. As shown in fig. 4, for each piece of power fingerprint data G (a, X) including the original feature information, the graph convolution layer may be used to aggregate the information, and new power fingerprint data G (a, X') with more abstract and more identification may be output. The essence of graph representation learning lies in that nodes are mapped into a dense and low-dimensional vector space, network information is kept as much as possible in the mapping process, and similarity and difference among the nodes are mined. The key role is a Graph Convolution Network (GCN), the feature extraction mode of the GCN is similar to the convolution operation of the traditional european space, and feature extraction is realized by transmitting, converting and aggregating neighbor node messages. This process can be represented by the following sub-formula:
in the formula: theta is a weight matrix; deg (i) represents the degree of node i, deg (j) represents the degree of node j; σ denotes the activation function.
In order to realize the identification of the electrical equipment, a load identification model based on a graph convolution network is constructed, the power fingerprint library of the electrical equipment established in the step S3 is used for training and learning the load identification model, and an Adam optimizer is used for correcting trainable parameters in the model by using a back propagation algorithm. The input of the load identification model is electric power fingerprint data constructed by electric operation characteristics and mutual information entropy between every two electric operation characteristics, and the output of the load identification model is the type of electric equipment corresponding to the electric power fingerprint data. As shown in fig. 5, firstly, the load identification model performs feature transformation on input power fingerprint data by using a feature transformation layer, secondly, three graph convolution layers (GCN layers) are stacked to extract structural features of the power fingerprint data, secondly, a graph pooling layer is used to aggregate full graph information, and finally, four full connection layers are stacked to realize nonlinear mapping between the power fingerprint data and the type of the electrical equipment, so that a mapping relation between the power fingerprint data and the electrical equipment is constructed, and a mapping relation between the power fingerprint data and the electrical equipment is constructed, so that the mapping relation is determined without a large amount of calculation in the load identification process, and a non-invasive load identification algorithm is simple; because the electrical operation characteristics of the electrical equipment are contained in the electrical fingerprint image data, the electrical operation characteristics of the electrical equipment can be directly obtained based on the electrical fingerprint image data in the load identification process, so that the electrical fingerprint image data is beneficial to fully extracting the operation characteristics of the electrical equipment to realize load identification, the electrical operation characteristics of the electrical equipment do not need to be input into a non-invasive load identification algorithm in sequence to realize load identification, and the non-invasive load identification algorithm is high in identification speed and high in operation efficiency.
The above description is only the embodiments of the present invention, and the scope of protection is not limited thereto. The insubstantial changes or substitutions will now be made by those skilled in the art based on the teachings of the present invention, which fall within the scope of the claims.
Claims (14)
1. A graph data modeling electric power fingerprint identification method of electric equipment is characterized by comprising the following steps:
step S1: detecting an input event of electrical equipment in real time at an electricity utilization bus, and if the input event of the electrical equipment is detected, extracting a plurality of electrical operation characteristics of the electrical equipment;
step S2: establishing an incidence relation between every two of a plurality of electrical operation characteristics according to a mutual information entropy theory;
and step S3: taking the extracted electrical operation characteristics as nodes, taking the incidence relation between every two electrical operation characteristics as edges, and generating electric power fingerprint graph data containing the nodes and the edges according to a graph theory;
and step S4: and constructing a load identification model based on a graph convolution network to carry out graph representation learning on the electric power fingerprint graph data, and constructing a mapping relation between the electric power fingerprint graph data and the electric equipment so as to realize load identification.
2. The method as claimed in claim 1, wherein in step S1, the sum of the cumulative active power changes at the power bus is detected to determine whether the event of switching the electrical equipment occurs.
3. The method as claimed in claim 2, wherein the specific detection method of the cumulative sum of active power changes at the power utilization bus is:
defining a steady-state detection window and a transient-state detection window for active power at an electric bus, wherein the lengths of the steady-state detection window and the transient-state detection window are a and b respectively, and the steady-state detection window and the transient-state detection window jointly form an event detection window; respectively calculating the active power of the steady partPower average value P a And the active power average value P of the transient part b The method comprises the following steps:
then define the forward cumulative sum
And negative cumulative sum
Wherein
For recording the accumulated course of the electrical equipment investment,
the cumulative process for recording the cutting of the electrical equipment is as follows:
wherein E r Representing the normal fluctuation level of active power when the electrical equipment operates;
then, a threshold value H for the cumulative sum of switching events of the electrical equipment is defined, namely, the cumulative sum of positive changes of active power
When the threshold H is reached, it is considered that the event of putting the electrical equipment into operation has occurred at this time.
4. The method of claim 1, further comprising: in the step S1, the electrical operation characteristics of the electrical equipment include active power, reactive power, a current effective value, a current 3-order harmonic amplitude, a current 5-order harmonic amplitude, a current 7-order harmonic amplitude, a current harmonic total distortion rate, and an instantaneous power peak value; the extraction of the plurality of electrical operation characteristics of the electrical equipment means that the data collected at the power utilization bus after the input event occurs and the same type of data collected at the power utilization bus before the input event occurs are subtracted, so that the electrical operation characteristics of the input electrical equipment are obtained.
5. The method of claim 1, wherein: in the step S2, the incidence relation between each two electrical operation features refers to correlation characteristics and irrelevance characteristics between each two extracted electrical operation features; the mutual entropy refers to information shared between two associated electrical operating characteristics, wherein if information of one of the electrical operating characteristics is obtained, unpredictability of the other electrical operating characteristic is reduced.
6. The method as claimed in claim 5, wherein the mutual information entropy is calculated by the following formula:
where p (x) and p (y) are probability density functions of the variables x and y, and p (x, y) is a joint probability density function of the two variables.
7. The method as claimed in claim 1, wherein in step S3, all the electrical operating characteristics extracted in step S1 are defined as nodes in the electrical fingerprint map data, and the mutual entropy between every two electrical operating characteristics calculated in step S2 is defined as edges in the electrical fingerprint map data, so as to construct the electrical fingerprint map data G (a, X) of the electrical equipment with unique identifier, wherein a is an adjacent matrix of the map, and X is a characteristic matrix of the nodes in the map.
8. The method of claim 7, wherein for node characteristics, the data is linearly transformed using maximum and minimum normalization, mapping the data to [0,1] intervals, defined as follows:
where max is the maximum value in the data; min is the minimum in the data;
for the side weight, a softmax function is used for carrying out normalization processing on the mutual information entropy, and the calculation formula of each side weight value is as follows:
wherein k (x) i ,x j ) The mutual information entropy of the edge between the node i and the node j is obtained.
9. The method according to claim 8, wherein in step S4, for each power fingerprint data G (a, X) containing original feature information, the graph convolution layer is used to perform information aggregation to output more abstract and more identifiable new power fingerprint data G (a, X'), which is expressed by the following formula:
in the formula: theta is a weight matrix; deg (i) and deg (j) represent degrees of node i and node j; σ denotes the activation function.
10. The method as claimed in claim 8, wherein after the construction of the power fingerprint data of the electrical equipment is completed, the power fingerprint data of each electrical equipment is integrated to construct a power fingerprint database of the electrical equipment, and the power fingerprint data in the power fingerprint database is divided into a training set, a verification set and a test set according to the proportion of 3.
11. The method as claimed in claim 10, wherein in the step S4, a load recognition model based on a graph convolution network is constructed, and the power fingerprint library of the electrical equipment established in the step S3 is used for training and learning the load recognition model, wherein the input of the load recognition model is power fingerprint map data constructed by electrical operation characteristics and mutual information entropy between every two electrical operation characteristics, and the output of the load recognition model is the type of the electrical equipment corresponding to the power fingerprint map data.
12. The method according to claim 11, wherein in step S4, a load recognition model is constructed, in particular: firstly, a load identification model carries out feature transformation on input power fingerprint image data by using a feature transformation layer, secondly, at least three image convolution layers are stacked to extract structural features of the power fingerprint image data, secondly, a graph pooling layer is used to aggregate full image information, and finally, at least four full connection layers are stacked to realize nonlinear mapping between the power fingerprint image data and the type of electrical equipment, so that a mapping relation between the power fingerprint image data and the electrical equipment is constructed.
13. Computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 12.
14. A graph data modeling power fingerprinting system for electrical appliances, comprising a computer readable storage medium and a processor connected to each other, characterized in that the computer readable storage medium is as claimed in claim 13.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110335453.XA CN112989133B (en) | 2021-03-29 | 2021-03-29 | Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110335453.XA CN112989133B (en) | 2021-03-29 | 2021-03-29 | Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112989133A CN112989133A (en) | 2021-06-18 |
| CN112989133B true CN112989133B (en) | 2022-10-04 |
Family
ID=76337966
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110335453.XA Active CN112989133B (en) | 2021-03-29 | 2021-03-29 | Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112989133B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114050613B (en) * | 2021-11-29 | 2023-10-27 | 国网湖南省电力有限公司 | Online identification and tracing method and system for power grid voltage transient event |
| CN115563511B (en) * | 2022-12-05 | 2023-03-10 | 广州水沐青华科技有限公司 | Edge side power fingerprint identification method and device based on machine learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111914484A (en) * | 2020-08-07 | 2020-11-10 | 中国南方电网有限责任公司 | Recursive graph convolution network system for power grid transient stability evaluation |
| CN111965476A (en) * | 2020-06-24 | 2020-11-20 | 国网江苏省电力有限公司淮安供电分公司 | Low-voltage diagnosis method based on graph convolution neural network |
| CN112257841A (en) * | 2020-09-03 | 2021-01-22 | 北京大学 | Data processing method, device and equipment in graph neural network and storage medium |
| CN112435142A (en) * | 2020-12-16 | 2021-03-02 | 北京航空航天大学 | Power load identification method and load power utilization facility knowledge base construction method thereof |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180284758A1 (en) * | 2016-05-09 | 2018-10-04 | StrongForce IoT Portfolio 2016, LLC | Methods and systems for industrial internet of things data collection for equipment analysis in an upstream oil and gas environment |
| US10921801B2 (en) * | 2017-08-02 | 2021-02-16 | Strong Force loT Portfolio 2016, LLC | Data collection systems and methods for updating sensed parameter groups based on pattern recognition |
| US20200341987A1 (en) * | 2019-04-25 | 2020-10-29 | Microsoft Technology Licensing, Llc | Ranking database query results |
-
2021
- 2021-03-29 CN CN202110335453.XA patent/CN112989133B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111965476A (en) * | 2020-06-24 | 2020-11-20 | 国网江苏省电力有限公司淮安供电分公司 | Low-voltage diagnosis method based on graph convolution neural network |
| CN111914484A (en) * | 2020-08-07 | 2020-11-10 | 中国南方电网有限责任公司 | Recursive graph convolution network system for power grid transient stability evaluation |
| CN112257841A (en) * | 2020-09-03 | 2021-01-22 | 北京大学 | Data processing method, device and equipment in graph neural network and storage medium |
| CN112435142A (en) * | 2020-12-16 | 2021-03-02 | 北京航空航天大学 | Power load identification method and load power utilization facility knowledge base construction method thereof |
Non-Patent Citations (2)
| Title |
|---|
| L2-GCN: Layer-Wise and Learned Efficient Training of Graph Convolutional Networks;Yuning You et al.;《2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)》;20200619;第2124-2132页 * |
| 配电网馈线故障类型识别方法研究;陈伟凡;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20190415;C042-1267页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112989133A (en) | 2021-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109546659B (en) | Power distribution network reactive power optimization method based on random matrix and intelligent scene matching | |
| Li et al. | Predicting short-term electricity demand by combining the advantages of arma and xgboost in fog computing environment | |
| CN112989133B (en) | Graph data modeling power fingerprint identification method, storage medium and system for electrical equipment | |
| Shi et al. | An approach of electrical load profile analysis based on time series data mining | |
| Yun et al. | Research on intelligent fault diagnosis of power acquisition based on knowledge graph | |
| CN112989131B (en) | Graph representation learning electric appliance equipment power fingerprint decomposition method, storage medium and system | |
| CN113887912A (en) | Non-invasive load identification method for deeply learning downward embedded equipment | |
| Fatouh et al. | New semi-supervised and active learning combination technique for non-intrusive load monitoring | |
| Han et al. | Non‐intrusive load monitoring by voltage–current trajectory enabled asymmetric deep supervised hashing | |
| CN115758188A (en) | Non-invasive load identification method, device, equipment and medium | |
| Guohua et al. | Research on non-intrusive load monitoring based on random forest algorithm | |
| Zhou et al. | Non-intrusive extraction and forecasting of residential electric vehicle charging load | |
| CN111815022A (en) | Power load prediction method based on time-delay coordinate embedding method | |
| Liu et al. | Home appliances classification based on multi-feature using ELM | |
| CN111199014A (en) | Time sequence based seq2point NILM method and device | |
| Liu et al. | Multi-scale residual network for energy disaggregation | |
| Chhaya et al. | Application of data mining in smart grid technology | |
| Wang et al. | A non-intrusive load decomposition method for residents | |
| CN113191656B (en) | Low-voltage distribution network equipment load and topology linkage method based on data correlation analysis | |
| Lu et al. | Anomaly Recognition Method for Massive Data of Power Internet of Things Based on Bayesian Belief Network | |
| Yu et al. | Nonintrusive load disaggregation method based on graph signal processing | |
| CN116823338B (en) | Method for deducing economic attribute missing value of power consumer | |
| CN116599036A (en) | Two-stage non-invasive load decomposition method based on TCN and Informar | |
| Zhai et al. | Short-term wind speed prediction based on quadratic decomposition improved particle swarm optimization extreme learning machine | |
| Qi et al. | ARIMA-LSTM Line Loss Anomaly Analysis Method Based on Wavelet Decomposition |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: Room 604, Building 2, No. 5 Yunkai Road, Huangpu District, Guangzhou City, Guangdong Province, 510535 Applicant after: Guangzhou Shuimu Qinghua Technology Co.,Ltd. Address before: 510700 room 509, building D, 39 Ruihe Road, Huangpu District, Guangzhou City, Guangdong Province Applicant before: Guangzhou Shuimu Qinghua Technology Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |