WO2021107422A1

WO2021107422A1 - Nonintrusive load monitoring method using energy consumption data

Info

Publication number: WO2021107422A1
Application number: PCT/KR2020/014994
Authority: WO
Inventors: 지영민
Original assignee: 한국전자기술연구원
Priority date: 2019-11-28
Filing date: 2020-10-30
Publication date: 2021-06-03
Also published as: KR20210066534A; KR102272573B1

Abstract

A nonintrusive load monitoring method using energy consumption data is provided. A nonintrusive load monitoring method using energy consumption data according to an embodiment of the present invention comprises the steps of: storing multiple learning models; and inferring a current state by inputting real-time energy consumption data on the basis of the stored multiple learning models. Accordingly, it is possible to perform nonintrusive load monitoring (NILM) using only the main power consumption data of an energy consumption space, and thus it is possible to recognize what is happening in the energy consumption space and use the recognition result as an important parameter for intelligent control.

Description

Unsupervised learning-based load monitoring method of energy usage data

The present invention relates to an unsupervised learning-based load monitoring method of energy usage data, and more particularly, by classifying the state of past energy usage data and generating learning data necessary for artificial intelligence learning based on the classified data, It relates to a method for developing a recognition model that identifies a load by training a network, and monitoring a load based on power usage data using the developed recognition model.

In the conventional power energy measurement technology field, although a means for measuring the power usage for each load is provided, there is a problem that a large amount of cost is generated because each measurement means must be provided.

In order to improve this, in the field of power energy measurement-based load identification technology, various methods such as active power, reactive power, frequency analysis of power usage, and inference according to phase change have been tried to identify, but various power Since the energy use pattern of each device consuming is different, there is a limit to classifying them all in one way.

Therefore, it is required to find a way to identify the load using only the main power usage data of the energy consumption space.

The present invention has been devised to solve the above problems, and an object of the present invention is to provide a method of performing unsupervised learning-based load monitoring (NILM) using only main power usage data of an energy consumption space. .

According to an embodiment of the present invention for achieving the above object, an unsupervised learning-based load monitoring method of energy usage data includes the steps of: storing a multi-learning model; and inferring the current state by inputting real-time energy usage data based on the stored multi-learning model.

In this case, the inference step may infer a current state by identifying a load based on past energy use data.

In addition, according to an embodiment of the present invention, an unsupervised learning-based load monitoring method of energy usage data includes: securing a data set required for neuron learning by using a cluster classification technique of energy usage data; generating training data by adjusting the training sample data set in units of clusters classified for neuron learning; And by using the generated learning data, the step of learning a multi-learning model that outputs a result of the current state as an input of real-time energy usage data; may further include.

And the learning data may be classification identification data.

In addition, in the generating of the learning data, when the actual energy usage data and the classification identification data are secured, the learning data and the test data for learning the neurons may be classified.

And, in the step of generating the training data, the output unit of the neural network may be modified based on the number of identification data classified by the cluster classification technique.

In addition, in the learning step of the multi-learning model, by loading the stored multi-model and using real-time energy usage data and learning data, neuron backpropagation learning may be performed.

And according to an embodiment of the present invention, an unsupervised learning-based load monitoring method of energy usage data includes: verifying the performance of the entire multi-learning model confirmed by using test data; and, as a result of the verification, separating and storing the multi-learning model having the highest classification accuracy among all the multi-learning models.

In addition, the verification step may verify the performance of the entire multi-learning model stored or learned at every preset period, respectively.

On the other hand, according to another embodiment of the present invention, a computer-readable recording medium containing a computer program for performing an unsupervised learning-based load monitoring method of energy usage data includes the steps of: storing a multiple learning model; and inferring the current state by inputting real-time energy usage data based on the stored multiple learning model.

In addition, according to another embodiment of the present invention, an unsupervised learning-based load monitoring system of energy usage data includes: a learning model storage for storing multiple learning models; and a processor for inferring a current state by inputting real-time energy usage data based on the stored multi-learning model.

And according to another embodiment of the present invention, an unsupervised learning-based load monitoring method of energy usage data includes the steps of: storing a multi-learning model; verifying the performance of the entire multi-learning model stored at every preset period; Separating and storing a multi-learning model having the highest classification accuracy among all multi-learning models as a result of the verification; and inferring the current state by inputting real-time data based on the separately stored multi-learning model.

As described above, according to the embodiments of the present invention, unsupervised learning-based load monitoring (NILM) can be performed using only the main power usage data of the energy consumption space, and through this, what happens in the energy consumption space can be recognized, and the recognition result can be used as an important parameter for intelligent control.

1 is a flowchart provided for explaining an unsupervised learning-based load monitoring method of energy usage data according to an embodiment of the present invention;

2 to 3 are diagrams provided for explaining an unsupervised learning-based load monitoring method of energy usage data according to an embodiment of the present invention;

4 is a diagram provided to explain an unsupervised learning-based load monitoring system of energy usage data according to an embodiment of the present invention.

Hereinafter, the present invention will be described in more detail with reference to the drawings.

1 is a flowchart provided to explain an unsupervised learning-based load monitoring method (hereinafter, collectively referred to as a 'load monitoring method') of energy usage data according to an embodiment of the present invention, and FIGS. 2 to 3 are the present invention It is a view provided for the description of the load monitoring method according to an embodiment of the.

The load monitoring method according to the present embodiment may perform unsupervised learning-based load monitoring using only main power usage data of the energy consumption space to identify a load that is consuming energy in the energy consumption space.

Specifically, this load monitoring method does not know in advance the characteristics seen in individual facilities, and utilizes clustering technology of energy consumption status based on patterns of past energy consumption data to measure energy consumption without various equipment data attached to the load. By classifying the state change according to the increase/decrease, it is possible to identify the load that is consuming energy.

For example, this load monitoring method secures a basic data set necessary for learning through classification of past energy use data, and performs neural network learning based on this data set to complete a load identification recognition model. .

To this end, this load monitoring method utilizes a cluster classification technique of energy usage data to obtain a data set required for neuron learning ( S110 ), and a cluster unit training sample data set classified for neuron learning. A learning data generation step (S120) of generating learning data by adjusting the , a learning step (S130) of learning a multi-learning model that uses the generated learning data to output real-time energy usage data as input and output results of the current state (S130) ), a storage step (S140) of storing the multi-learning model, and an inference step (S150) of inferring the current state by inputting real-time energy usage data based on the stored multi-learning model.

Here, the learning data is classification identification data, and the cluster classification technique may be used to classify energy usage data by clusters as illustrated in FIG. 3 . 3 illustrates a state in which energy usage data is classified into first, second, and third clusters.

Meanwhile, in the learning data generation step ( S120 ), when the actual energy usage data and the classification identification data are secured, learning data and test data for neuron learning may be classified.

In addition, in the training data generation step ( S120 ), the output unit of the neural network may be modified based on the number of identification data classified by the cluster classification technique.

Specifically, for example, in the training data generation step ( S120 ), when there is a cluster in which the number of identification data classified by the cluster classification technique is equal to or less than a predetermined value, the number of energy samples of the corresponding cluster may be adjusted to increase.

And in the training data generation step (S120), when the number of samples of the corresponding cluster is increased to be greater than or equal to a preset value, the number of identification data classified by the cluster classification technique is adjusted so that all clusters are the same, thereby generating a sufficient number of training data. At the same time, it is possible to easily classify the training data and the test data.

In the learning step ( S130 ), neuron backpropagation learning may be performed by loading the stored multi-model and using real-time energy usage data and learning data.

In addition, this load monitoring method, after the learning step (S130) or the storage step (140), using test data to verify the performance of the confirmed or stored entire multi-learning model, and the verification result, among the entire multi-learning model The method may further include a separate storage step of separating and storing the multi-learning model having the highest classification accuracy.

That is, in this load monitoring method, after the learning step ( S130 ) or the storage step ( 140 ), the performance of the confirmed or stored entire multi-learning model is verified using test data, and classified among all the multi-learning models according to the verification result. By separately storing the multi-learning model with the highest accuracy, it is possible to infer the current state with high accuracy based on the separately-stored multi-learning model when inferring the current state.

For example, in the verification step, the performance of the entire multi-learning model stored or trained at each preset cycle is verified, and the current state can be inferred based on the multi-learning model with the highest classification accuracy according to the current situation. have.

At this time, the method of separately storing the multi-learning model with the highest classification accuracy is to assign identification classification tag information to each stored multi-learning model, and separate and store the identification classification tag information of the multi-learning model with the highest classification accuracy. , a method of calling a multi-learning model with the highest classification accuracy can be used when inferring the current state.

Through this, in the inference step ( S150 ), the current state may be inferred by identifying the load based on the past energy use data.

In addition, this load monitoring method does not only identify the load with energy (power) data, but can be utilized based on consumption data of various energy sources, and is also used for analysis that recognizes the state by monitoring changes in the values of various sensor data. It can be applied and used.

4 is a diagram provided for explanation of an unsupervised learning-based load monitoring system (hereinafter, collectively referred to as a 'monitoring system') of energy usage data according to an embodiment of the present invention.

Referring to FIG. 4 , the monitoring system includes a data collection unit 110 , a processor 120 , and a learning model storage 130 .

The data collection unit 110 is provided to collect real-time energy usage data.

The processor 120 may infer a current state by inputting real-time energy usage data based on the stored multi-learning model.

Specifically, for example, the processor 120 utilizes a cluster classification technique of energy usage data to secure a data set required for neuron learning ( S110 ), and learn in units of clusters classified for neuron learning. Learning to train a multi-learning model that outputs a result of the current state by inputting real-time energy usage data as an input, using the generated training data, by adjusting the sample data set to generate training data (S120) Step (S130), the storage step (S140) of storing the multi-learning model in the learning model storage, and the inference step (S150) of inferring the current state by inputting real-time energy usage data based on the stored multi-learning model are sequentially performed. can

In addition, after the learning step ( S130 ) or the storing step ( 140 ), the processor 120 verifies the performance of the confirmed or stored entire multi-learning model by using the test data, and classifies among all the multi-learning models according to the verification result. By separately storing the multi-learning model with the highest accuracy, the current state can be inferred with high accuracy based on the separately-stored multi-learning model when inferring the current state.

The learning model storage 130 is provided to store multiple learning models.

Specifically, for example, the learning model storage 120 gives identification classification tag information to each of the stored multiple learning models, and separates and stores the identification classification tag information of the multiple learning model with the highest classification accuracy, and the current state In inference, a multi-learning model with the highest classification accuracy can be called.

On the other hand, it goes without saying that the technical idea of the present invention can also be applied to a computer-readable recording medium containing a computer program for performing the functions of the apparatus and method according to the present embodiment. In addition, the technical ideas according to various embodiments of the present invention may be implemented in the form of computer-readable codes recorded on a computer-readable recording medium. The computer-readable recording medium may be any data storage device readable by the computer and capable of storing data. For example, the computer-readable recording medium may be a ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, hard disk drive, or the like. In addition, the computer-readable code or program stored in the computer-readable recording medium may be transmitted through a network connected between computers.

In addition, although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and the technical field to which the present invention belongs without departing from the gist of the present invention as claimed in the claims Various modifications are possible by those of ordinary skill in the art, and these modifications should not be individually understood from the technical spirit or prospect of the present invention.

Claims

storing multiple learning models; and

Based on the stored multi-learning model, inputting real-time energy usage data to infer the current state; an unsupervised learning-based load monitoring method of energy usage data comprising a.
The method according to claim 1,

The reasoning step is

A load monitoring method based on unsupervised learning of energy usage data, characterized in that the current state is inferred by identifying the load based on the past energy usage data.
The method according to claim 1,

obtaining a data set required for neuron learning by utilizing a cluster classification technique of energy usage data;

generating training data by adjusting the training sample data set in units of clusters classified for neuron learning; and

Unsupervised learning-based load monitoring method of energy usage data, characterized in that it further comprises; using the generated learning data, learning a multi-learning model that outputs a result of the current state as input to real-time energy usage data as an input .
4. The method according to claim 3,

learning data,

An unsupervised learning-based load monitoring method of energy usage data, characterized in that it is classification identification data.
5. The method according to claim 4,

The step of generating the training data is,

When the actual energy usage data and classification identification data are secured, the unsupervised learning-based load monitoring method of energy usage data, characterized in that the training data and test data for neuron learning are classified.
6. The method of claim 5,

The step of generating the training data is,

An unsupervised learning-based load monitoring method of energy usage data, characterized in that the output unit of the neural network is modified based on the number of identification data classified by the cluster classification technique.
7. The method of claim 6,

The learning steps of a multi-learning model are:

An unsupervised learning-based load monitoring method of energy usage data, characterized in that by loading multiple stored models and using real-time energy usage data and training data, neuron backpropagation learning is performed.
8. The method of claim 7,

verifying the performance of the trained or stored entire multi-learning model using the test data; and

As a result of the verification, separating and storing the multi-learning model having the highest classification accuracy among all the multi-learning models; the unsupervised learning-based load monitoring method of energy usage data, further comprising: a.
9. The method of claim 8,

The verification step is

An unsupervised learning-based load monitoring method of energy usage data, characterized in that each of the performance of the stored or learned multi-learning model is verified at every preset cycle.
storing multiple learning models; and

A computer-readable recording medium containing a computer program for performing an unsupervised learning-based load monitoring method of energy usage data including; inputting real-time energy usage data based on the stored multiple learning model to infer the current state.
a learning model repository for storing multiple learning models; and

An unsupervised learning-based load monitoring system of energy usage data comprising a; processor for inferring a current state by inputting real-time energy usage data based on the stored multiple learning model.
storing multiple learning models;

verifying the performance of the entire multi-learning model stored at every preset period;

Separating and storing a multi-learning model having the highest classification accuracy among all multi-learning models as a result of the verification; and

An unsupervised learning-based load monitoring method of energy usage data comprising a; inferring a current state by inputting real-time data based on a multi-learning model stored separately.