KR102410052B1 - Apparatus and method for managing cloud-based microgrid distributed power supply - Google Patents

Abstract

When the device for managing the microgrid distributed power supply receives the measurement data from the microgrid, the device for managing the microgrid distributed power generates a measurement vector by mapping the received measurement data to a predetermined vector space, and the generated measurement A data processing unit for inputting a vector into a deep learning model and an anomaly detection model including an encoder and a decoder, and when the anomaly detection model simulates the measurement vector to calculate a simulated measurement vector, the difference between the measurement vector and the simulated measurement vector and an abnormality detection unit that determines whether a loss indicating a value exceeds a preset threshold, and determines that there is an abnormality in the microgrid when the loss is greater than or equal to the threshold.

Description

Apparatus and method for managing cloud-based microgrid distributed power supply}

The present invention relates to a technology for managing a microgrid distributed power supply, and more particularly, to a cloud-based device for managing a microgrid distributed power supply and a method therefor.

The current electricity supply structure is centralized management, and KEPCO is supplying it to a large number of consumers. This structure makes liberalization/marketization difficult due to the lack of substitutes, and the elasticity of demand and supply is low. In order to solve this problem, new and renewable energy power generation facilities in the form of distributed power sources are being built in various places in the region through institutions or independent operators. Distributed power generation has difficulties in operation due to difficulties in maintenance and lack of a profit structure model. This is particularly the case with small independent businesses. Therefore, a system that can integrate, manage, and trade distributed power is required.

Korean Patent Publication No. 2020-0065655 (published on June 9, 2020)

An object of the present invention is to provide an apparatus for managing a cloud-based microgrid distributed power supply and a method therefor.

When receiving measurement data from the microgrid, the apparatus for managing distributed power supply in a microgrid according to a preferred embodiment of the present invention for achieving the above object maps the received measurement data to a predetermined vector space to obtain a measurement vector A data processing unit for generating a measurement vector, which is a deep learning model and inputting the generated measurement vector to an anomaly detection model including an encoder and a decoder, and when the anomaly detection model simulates the measurement vector to calculate a simulated measurement vector, the measurement vector and an abnormality detection unit that determines whether a loss representing a difference between and the simulated measurement vector exceeds a preset threshold, and determines that there is an abnormality in the microgrid when the loss is greater than or equal to the threshold.

The device prepares a measurement vector for learning from measurement data measured in a steady state of the microgrid, and when the abnormal detection model calculates a simulated measurement vector for learning by performing an operation on the measurement vector for learning, the measurement vector for learning and the learning The method further includes a model generator for learning the anomaly detection model by calculating a loss based on the simulated measurement vector and performing optimization to update the parameters of the anomaly detection model so that the loss is minimized.

에 따라 손실을 산출하고, 상기

은 손실을 나타내고,

이고,

이고, 상기

는 인코더의 가중치 행렬이고, 상기

는 인코더의 바이어스 벡터이고, 상기

는 인코더의 활성화 함수이고, 상기

는 디코더의 가중치 행렬이고, 상기

는 디코더의 바이어스 벡터이고, 상기

는 디코더의 활성화 함수인 것을 특징으로 한다. The model generating unit is

Calculate the loss according to the above

represents the loss,

ego,

and said

is the weight matrix of the encoder,

is the bias vector of the encoder,

is the activation function of the encoder,

is the weight matrix of the decoder,

is the bias vector of the decoder,

is an activation function of the decoder.

After the learning is completed, the model generator sets a cost function for a first error for determining normal data, which is measurement data in a normal state, as abnormal, and a second error for determining abnormal data, which is measurement data in an abnormal state, as normal. , it is characterized in that a threshold value at which the cost function is minimized is set.

A method for managing a microgrid distributed power supply according to a preferred embodiment of the present invention for achieving the object as described above includes the steps of: a data processing unit receiving measurement data from the microgrid; Creating a measurement vector by mapping to a predetermined vector space, inputting the generated measurement vector to an anomaly detection model that is a deep learning model, and calculating a simulated measurement vector by simulating the measurement vector by the anomaly detection model; , determining whether the loss indicating the difference between the measurement vector and the simulated measurement vector by the anomaly detection unit exceeds a preset threshold, and if the loss is greater than or equal to the threshold value by the abnormality detection unit, there is an abnormality in the microgrid It includes the step of determining that

The method includes the steps of: before the data processing unit receives the measurement data from the microgrid, the model generation unit prepares a measurement vector for learning from the measurement data measured in a steady state of the microgrid; Calculating a simulation measurement vector for learning by performing an operation for, the step of a model generation unit calculating a loss based on the measurement vector for learning and the simulation measurement vector for training, and the model generation unit detecting the abnormality so that the loss is minimized The method further includes training an anomaly detection model by performing optimization to update parameters of the model.

에 따라 손실을 산출하고, 상기

은 손실을 나타내고,

이고,

이고, 상기

는 인코더의 가중치 행렬이고, 상기

는 인코더의 바이어스 벡터이고, 상기

는 인코더의 활성화 함수이고, 상기

는 디코더의 가중치 행렬이고, 상기

는 디코더의 바이어스 벡터이고, 상기

는 디코더의 활성화 함수인 것을 특징으로 한다. In the step of calculating the loss, the model generation unit is

Calculate the loss according to the above

represents the loss,

ego,

and said

is the weight matrix of the encoder,

is the bias vector of the encoder,

is the activation function of the encoder,

is the weight matrix of the decoder,

is the bias vector of the decoder,

is an activation function of the decoder.

In the method, after the step of learning the abnormality detection model, a type 1 error for determining normal data, which is measurement data in a normal state, as abnormal, and a type 2 error for determining abnormal data, which is measurement data in an abnormal state, as normal. The method further includes setting a cost function for , and setting a threshold at which the cost function is minimized.

According to the present invention, it is possible to efficiently manage the cloud-based microgrid distributed power supply.

1 is a diagram for explaining the configuration of a system for managing a cloud-based microgrid distributed power supply according to an embodiment of the present invention.
2 is a diagram for explaining a method for a user device to purchase power according to an embodiment of the present invention.
3 is a view for explaining a method for a management server to manage distributed power including a plurality of microgrids according to an embodiment of the present invention.
4 and 5 are diagrams for explaining a method of managing a microgrid by the grid management unit of the management server according to an embodiment of the present invention.
6 and 7 are diagrams for explaining a method of performing a prepaid power transaction by the prepaid transaction unit of the management server according to an embodiment of the present invention.
8 is a view for explaining the detailed configuration of the grid management unit 100 according to an embodiment of the present invention.
9 is a diagram for explaining the configuration of an anomaly detection model (FDM) according to an embodiment of the present invention.
10 is a flowchart illustrating a method of generating an anomaly detection model (FDM) according to an embodiment of the present invention.
11 is a flowchart illustrating a method for detecting an abnormality in a state of a microgrid using a plurality of sensors according to an embodiment of the present invention.
12 is a diagram for explaining the configuration of an anomaly detection model (FDM) according to another embodiment of the present invention.
13 is a flowchart illustrating a method of generating an anomaly detection model (FDM) according to another embodiment of the present invention.
14 is a flowchart illustrating a method for detecting an abnormality in a state of a microgrid using a plurality of sensors according to another embodiment of the present invention.
15 is a diagram illustrating a computing device according to an embodiment of the present invention.

Prior to the detailed description of the present invention, the terms or words used in the present specification and claims described below should not be construed as being limited to their ordinary or dictionary meanings, and the inventors should develop their own inventions in the best way. It should be interpreted as meaning and concept consistent with the technical idea of the present invention based on the principle that it can be appropriately defined as a concept of a term for explanation. Therefore, the embodiments described in the present specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all the technical spirit of the present invention, so various equivalents that can be substituted for them at the time of the present application It should be understood that there may be water and variations.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this case, it should be noted that in the accompanying drawings, the same components are denoted by the same reference numerals as much as possible. In addition, detailed descriptions of well-known functions and configurations that may obscure the gist of the present invention will be omitted. For the same reason, some components are exaggerated, omitted, or schematically illustrated in the accompanying drawings, and the size of each component does not fully reflect the actual size.

First, a system for managing a cloud-based microgrid distributed power supply according to an embodiment of the present invention will be described. 1 is a diagram for explaining the configuration of a system for managing a cloud-based microgrid distributed power supply according to an embodiment of the present invention. 2 is a diagram for explaining a method for a user device to purchase power according to an embodiment of the present invention. 3 is a view for explaining a method for a management server to manage distributed power including a plurality of microgrids according to an embodiment of the present invention. 4 and 5 are diagrams for explaining a method of managing a microgrid by the grid management unit of the management server according to an embodiment of the present invention. 6 and 7 are diagrams for explaining a method of performing a prepaid power transaction by the prepaid transaction unit of the management server according to an embodiment of the present invention.

Referring to FIG. 1 , a system for managing the microgrid distributed power supply includes a management server 10 , a microgrid 20 , a business operator device 30 , a smart meter 40 , and a user device 50 .

The management server 10 is a cloud server, and is for managing distributed power including a plurality of microgrids 20 . A plurality of microgrids 20 are provided by different operators, and the operator device 30 is a device used by the operator. In addition, each of the plurality of consumers is provided with a smart meter 40 for reading the power consumption, and the user device 50 is a device used by the user of each consumer.

The management server 10 includes a grid management unit 100 and a prepaid transaction unit 200 . The grid management unit 100 is for managing power supply by connecting with a plurality of microgrids 20 and a plurality of operator devices 30 corresponding to the plurality of microgrids 20 . According to an embodiment, the grid management unit 100 may be implemented as a virtual machine. When a business operator subscribes to the microgrid 20 management and power sales service provided by the management server 10, the management server 10 operates the grid management unit 100 corresponding to the microgrid 20 as a virtual machine. can create In addition, the prepaid transaction unit 200 is for managing power transaction by connecting the plurality of smart meters 40 and the plurality of user devices 50 corresponding to the plurality of smart meters 40 .

As shown in FIG. 2 , the user device 50 may purchase power points from the management server 10 according to a user's manipulation.

On the other hand, it is assumed that the plurality of distributed power sources include, for example, the first to third microgrids 21 , 22 , and 23 . First to third microgrids (21, 22, 23) Each operator, for example, through the first to third operator devices (31, 32, 33) first to third microgrids (21, 22, 23) You can run each of them. The generation amount of the first microgrid 21 is the highest, the generation amount of the third microgrid 23 is the lowest, the generation amount of the second microgrid 22 is lower than the generation amount of the first microgrid 21, and the third It is assumed that the power generation amount of the microgrid 23 is higher. In this case, relatively, the power price of the first microgrid 21 with the highest amount of generation is the lowest, and the power price of the third microgrid 23 with the lowest amount of generation is the highest, and the second microgrid has the medium amount of generation. (22) can be priced in the middle. In this way, the price of each of the plurality of microgrids 21 , 22 , and 23 may be different, and the user device 50 may selectively purchase power through the purchased power point after comparing the prices.

As shown in Figure 3, the management server 10 is a plurality of microgrids (21, 22, 23) and a plurality of operator devices (31, 32, 33) corresponding to the plurality of microgrids (21, 22, 23) ) to manage distributed power including a plurality of microgrids (21, 22, 23).

Referring to FIG. 4 , the microgrid 20 includes a renewable energy generator, an auxiliary generator, an inverter, an Energy Storage System (ESS), and the like. Renewable energy can be solar power generation, wind power generation, geothermal power generation, etc. for power generation. The generator is for emergency operation in preparation for the lack of new and renewable energy generation, and a diesel engine can be representatively exemplified. ESS (Energy Storage System) is for storing the generated electricity. The inverter is for converting the generated electricity into DC/AC.

In addition, the microgrid 20 includes an operation module for controlling the edge EMS (SCADA) and the edge EMS. The edge EMS controls data collection and functions of facilities installed in the microgrid 20 , and has a gateway function that communicates with the management server 10 . At this time, communication is made in a manner such as HTTP, Socket, and the like, and the management server 10 may control the edge EMS to manage the entire microgrid 20 .

The grid management unit 100 of the management server 10 manages information data including various types of information of the microgrid 20 through the edge EMS of the microgrid 20, and the microgrid 20 in the operator device 30 ) of information data can be provided. Then, the operator device 30 provides so that the operator or the manager can read the information data of the microgrid (20). For example, as shown in FIG. 5 , the operator device 30 may receive information data from the grid management unit 100 of the management server 10 and display it. Such information data includes, for example, a unit price of a set power point, a remaining power point, an amount of power remaining in the ESS, an amount of power load per hour (amount of demand for power), and the like. In addition, the grid management unit 100 may have a security and authentication function for providing functions for each authority to an administrator or a business operator using the operator device 30 . In particular, the grid management unit 100 may have an AI-based analysis function according to an embodiment of the present invention. It includes a solar power generation amount prediction function, a power generation prediction-based ESS charging/discharging scheduling function, and an abnormality detection function. Among these, the anomaly detection function will be described in more detail below.

Referring to FIG. 6 , smart meters 40 : 41 , 42 , 43 are provided to measure and manage power consumption for each consumer. The smart meters 40 : 41 , 42 , 43 may be connected to the management server 10 through the user devices 50 : 51 , 52 , 53 for each customer or may be directly connected to the management server 10 .

The smart meter 40 is for consumer-side power management analysis. The smart meter 40 may automatically control power-on permission and cut-off based on the customer's remaining power purchase amount, measure power consumption, and collect and analyze related data. In addition, the smart meter 40 has an AI-based power load prediction and charging time notification function, and has a power measurement-based electric leakage and conduction detection function.

The prepaid transaction unit 200 of the management server 10 provides meter reading, billing and analysis functions for each customer. In addition, various types of information related thereto are provided to the operator device 30 or the user device 50 . An example of a screen for providing information data to the user device 50 is illustrated in FIG. 7 . As shown, this information data includes, for example, the remaining power usage (calculated value) according to the unit price of the microgrid that is currently being traded, the remaining power point, the remaining power consumption using the average of the user power consumption pattern. number of days, user power consumption analysis data, and the like.

In addition, the prepaid transaction unit 200 provides an AI analysis function for predicting an AI-based power load amount and notifying the charging time, and detecting electric leakage and conduction based on power measurement.

In addition, the prepaid transaction unit 200 provides an O&M (Operation & Maintenance) function such as remote power meter reading, data collection and analysis through the smart meter 40 .

In particular, the prepaid transaction unit 200 is connected to the user device 40 to sell power points to the user device 40 , and sells power of the microgrid 20 of the operator selected by the user device 40 . In addition, the prepaid transaction unit 200 is connected to the operator device 30 to perform the settlement of the power purchased by the user device 40 as a power point. In this case, the prepaid transaction unit 200 may provide a prepaid transaction service in connection with a card company, a bank company, a telecommunication company, and the like.

The user device 40 may access the management server 10 , purchase power points through the prepaid transaction unit 200 , and purchase power using the purchased power points. In addition, the user device 40 may monitor various types of data read by the smart meter 50 in conjunction with the smart meter 50 .

On the other hand, according to the embodiment of the present invention, the grid management unit 100 through the measurement data collected from the microgrid 20 through an anomaly detection model (FDM), which is a deep learning model, the microgrid 20 Detect whether there is an abnormality in This will be described in more detail. 8 is a view for explaining the detailed configuration of the grid management unit 100 according to an embodiment of the present invention. Referring to FIG. 8 , the grid management unit 100 includes a model generation unit 110 , a data processing unit 120 , and an abnormality detection unit 130 .

The model generator 110 is to generate an anomaly detection model (FDM) that is a deep learning model according to an embodiment of the present invention through deep learning.

According to an embodiment, the model generation unit 110 uses the measurement data measured from the renewable energy generator and the ESS in a steady state as learning data to determine whether there is an abnormality in the renewable energy generator (FDM) , and using the measurement data measured from the ESS in a steady state as learning data, an anomaly detection model (FDM) for determining whether the ESS is abnormal can be created. Measurement data measured from a renewable energy generator in a steady state is, for example, when the renewable energy generator is a photovoltaic generator, the amount of power generated by each panel, temperature by panel, inverter temperature, voltage, current and temperature of the output terminal of the solar generator, etc. can be exemplified. Measurement data measured from the ESS in a steady state may exemplify, for example, voltage, current, temperature, humidity, etc. of each cell, module, pack, and rack of the battery of the ESS.

According to another embodiment, the model generation unit 110 uses both the measurement data measured from the renewable energy generator in the steady state and the measurement data measured from the ESS in the steady state as learning data, the renewable energy generator and the ESS. An anomaly detection model (FDM) for determining whether at least one of the above is abnormal may be generated. A method of generating the anomaly detection model (FDM) as described above will be described in more detail below.

The data processing unit 120 may receive measurement data from the microgrid 20 . When the data processing unit 120 receives the measurement data, it maps the measurement data to a predetermined vector space to generate a measurement vector that is a tensor of a predetermined dimension. The dimension of the tensor may be determined according to the type of measurement data.

The abnormality detection unit 130 determines whether the difference between the measurement vector and the simulated measurement vector exceeds a preset threshold when the learned abnormality detection model (FDM) simulates the measurement vector and calculates a pseudo-measured vector. to judge As a result of this determination, if the difference between the measurement vector and the simulated measurement vector is greater than or equal to the threshold value, it is determined that there is an abnormality in the state of the microgrid 20 , that is, at least one of the renewable energy generator and the ESS. When it is detected that there is an abnormality in the state of the microgrid 20 , the abnormality detection unit 130 may transmit an inspection message to the corresponding operator device 30 to inspect the microgrid 20 . For example, such a check message may be made through a push notification.

In this case, the abnormality detection unit 130 may directly control the microgrid 20 when the state abnormality of the microgrid 20 is urgent.

The operation of the control unit 13 including the model generating unit 110 , the data processing unit 120 , the abnormality detecting unit 130 , and the notification unit 400 will be described in more detail below.

Next, a configuration of an anomaly detection model (FDM) for managing a microgrid distributed power supply using a plurality of sensors according to an embodiment of the present invention will be described. 9 is a diagram for explaining the configuration of an anomaly detection model (FDM) according to an embodiment of the present invention. Referring to FIG. 9 , the anomaly detection model (FDM) includes an encoder (EN), a latent layer (LL), and a decoder (DE). According to an embodiment of the present invention, during learning, an encoder for learning rEN connected to the output terminal of the decoder DE and having the same structure as the encoder EN is additionally used.

The encoder EN includes a plurality of convolution layers (CL) including a convolution operation and an operation by an activation function. Also, the encoder EN may optionally further include a pooling layer (PL) connected to each of the plurality of convolutional layers CL. This pooling layer PL is for down-sampling.

The decoder DE includes a plurality of deconvolution layers (DLs) including a deconvolution operation and an operation by an activation function. Also, the decoder DE may optionally further include an unpooling layer (UL) connected to each of the plurality of deconvolution layers DL. The unpooling layer UL is for up-sampling.

The activation functions used in the plurality of convolutional layers (CL) or the plurality of deconvolution layers (DL) described above are Sigmoid, Hyperbolic tangent (tanh), Exponential Linear Unit (ELU), and ReLU ( Rectified Linear Unit), Leakly ReLU, Maxout, Minout, Softmax, etc. can be exemplified.

As described above, the anomaly detection model (FDM) includes a plurality of layers, and the plurality of layers includes a plurality of operations. In addition, a plurality of layers are connected by a weight (w: weight). The calculation result of one layer is weighted and becomes the input of the node of the next layer. That is, one layer of the anomaly detection model (FDM) receives a weighted value from the previous layer, performs an operation on it, and transfers the operation result to the input of the next layer.

When the measurement vector (m) is input, the encoder (EN) performs a plurality of operations in which a plurality of inter-layer weights are applied to the input measurement vector (or the measurement vector m for learning), and a latent vector (or latent vector for learning, Latent Vector: z) is calculated, and a latent layer (LL) is created with this latent vector. In addition, according to an embodiment of the present invention, during learning, the learning encoder (rEN) may receive a simulation measurement vector (m') for learning. In this way, when the simulated measurement vector (m') for learning is input, the learning encoder (rEN) performs a plurality of operations in which a plurality of inter-layer weights are applied to the input simulated measurement vector (m') to perform a simulation potential for learning. The vector z' can be calculated and output.

When the latent vector (or latent vector for learning: z) output from the encoder (EN) is input, the decoder DE performs a plurality of operations in which a plurality of inter-layer weights are applied to the latent vector (or latent vector for learning: z). A simulation measurement vector (or a simulation measurement vector for learning: m') that simulates a measurement vector (or a measurement vector for learning: m) is generated by performing it.

Next, a method for generating an anomaly detection model (FDM) according to an embodiment of the present invention will be described. 10 is a flowchart illustrating a method of generating an anomaly detection model (FDM) according to an embodiment of the present invention.

9 and 10 , the model generator 110 initializes the abnormality detection model FDM in step S110. This initialization means initializing the parameters of the anomaly detection model (FDM), that is, the weight w. When the initialization is completed, the model generation unit 110 prepares the measurement vector m for learning used for learning in the anomaly detection model FDM initialized in step S120 . The measurement vector (m) for learning is collected from the microgrid 20, and at least one of the measurement data measured from the renewable energy generator in the steady state and the measurement data measured from the ESS in the steady state is mapped to a predetermined vector space. it will be created

Next, the model generating unit 110 inputs the learning measurement vector m to the abnormality detection model FDM in step S130 . Accordingly, the anomaly detection model (FDM) performs calculations by the encoder (EN), the decoder (DE), and the learning encoder (EN) on the learning measurement vector (m) in step S140 to obtain the learning latent vector (z), for learning The simulation measurement vector (m') and the simulation latent vector for learning (z') are sequentially calculated. Hereinafter, the calculation process will be described in more detail.

First, the encoder (EN) performs a plurality of operations in which a plurality of inter-layer weights are applied on the measurement vector for learning (m) to obtain a latent vector (z) for learning in which the characteristics of the measurement vector for learning (m) are latent. Calculate. The latent layer (LL) consists of such a learning latent vector (z). Next, the decoder DE calculates a simulation measurement vector m' for learning that simulates the measurement vector m for learning by performing a plurality of operations in which a plurality of inter-layer weights are applied to the learning latent vector z. And the learning encoder (rEN) performs a plurality of operations in which a plurality of inter-layer weights are applied to the simulation measurement vector for learning (m') to simulate the learning latent vector (z). do.

Next, the model generator 110 calculates a pseudo latent loss (PLL) representing the difference between the simulation latent vector for learning (z') and the latent vector for learning (z) in step S150, and a simulated measurement vector for learning After calculating the pseudo loss (PL) representing the difference between (m') and the measurement vector for training (m), the total loss including the simulation potential loss (PLL) and the simulation loss (PL) is calculated.

At this time, the model generation unit 110 may calculate the simulation potential loss (PLL) according to the following equation (1).

Here, PLL is a simulation potential loss, z is a learning latent vector, and z' is a simulation latent vector for learning. n is the number of batches of measurement vectors (m) for training.

In addition, the model generator 110 may calculate the simulation loss (PL) according to the following Equation (2).

In Equation 2, PL is a simulation loss, m is a measurement vector for learning, and m' denotes a simulation measurement vector for learning. n is the number of batches of measurement vectors (m) for training.

Then, the model generation unit 110 is an anomaly detection model (FDM) through a back-propagation algorithm to minimize the total loss including the simulation loss (PL) and the simulation potential loss (Et) in step S160. Optimization to update the weight w is performed.

The above-described steps S120 to S160 are repeated using a plurality of different training measurement vectors (m) or a plurality of different training measurement vectors (m) batches to determine the weight w of the anomaly detection model (FDM). update And these repetitions are made until each loss and total loss of the simulation loss (PL) and the simulation potential loss (Et) are less than or equal to a preset target value.

Therefore, the model generator 110 determines whether each loss and total loss of the simulation loss (PL) and the simulation potential loss (Et) in step S170 are all less than or equal to a preset target value, and the simulation loss (PL) and the simulation potential loss (Et) If each loss and the total loss are all less than or equal to a preset target value, learning is completed.

When the learning is completed, the model generating unit 110 sets a difference between the measurement vector m and the simulated measurement vector m' in step S180, that is, the threshold Lth of the simulation loss PL. Specifically, the model generation unit 110 re-enters a plurality of error learning measurement vectors (m), that is, measurement vectors generated from measurement data measured in a normal state, to the anomaly detection model (FDM) that has been trained. After generating a plurality of simulation measurement vectors (m') for error learning, a threshold value (Lth) is calculated according to the following Equation (3).

In Equation 3, Lth represents a threshold. In addition, μ represents the average of the simulation loss (PL) representing the difference between the plurality of error learning simulation measurement vectors (m') and the plurality of error learning measurement vectors (m). σ is the standard deviation of the simulation loss (PL) representing the difference between the plurality of measurement vectors for error learning (m') and the plurality of measurement vectors for error learning (m). λ is a weight for the standard deviation of the simulation loss (PL), and is a preset value.

When the learning of the anomaly detection model (FDM) is completed according to the procedure as in FIG. 10 and a threshold is set, the abnormality of the state of the microgrid 20 using the abnormality detection model (FDM), that is, renewable energy It is possible to detect an abnormality in at least one of the generator and the ESS. These methods will be described. 11 is a flowchart illustrating a method for detecting an abnormality in a state of a microgrid using a plurality of sensors according to an embodiment of the present invention.

Referring to FIG. 11 , the data processing unit 120 may receive measurement data from the microgrid 20 in step S210 . Such measurement data includes measurement data measured from the renewable energy generator and measurement data measured from the ESS.

Then, the data processing unit 120 maps the measurement data to a predetermined vector space in step S220 to generate the measurement vector m. Then, the data processing unit 120 inputs the measurement vector (m) to the abnormality detection model (FDM) that has been trained in step S230. The anomaly detection model (FDM) has been trained as described above with reference to FIG. 10 .

When the measurement vector (m) is input, the anomaly detection model (FDM) simulates the measurement vector (m) in step S240 to calculate a pseudo-sensor vector. The step S240 will be described in more detail as follows.

First, the encoder EN of the anomaly detection model FDM calculates a latent vector z by performing a plurality of operations in which a plurality of inter-layer weights are applied to the measurement vector m. Next, the decoder DE of the anomaly detection model FDM performs a plurality of operations in which a plurality of inter-layer weights are applied to the latent vector z to simulate the measurement vector m by performing a simulated measurement vector (m') to calculate

The abnormality detection unit 130 determines whether the simulation loss (PL) representing the difference between the simulated measurement vector (m') and the measurement vector (m) in step S250 exceeds the threshold value (Lth) calculated during learning (S180) do.

As a result of the determination in step S250, if the simulation loss PL is greater than or equal to the threshold Lth, the abnormality detection unit 130 determines that an abnormality has occurred in the corresponding microgrid 20 in step S260, and checks the microgrid 20 The inspection message may be transmitted to the corresponding operator device 30 . In this step S260, the abnormality detection unit 130 determines that the state abnormality of the microgrid 20 is urgent and determines that the simulation loss PL is greater than or equal to the threshold Lth and the difference from the threshold Lth is greater than or equal to a predetermined value. 20) can be directly controlled. For example, the abnormality detection unit 130 may stop the renewable energy generator and the ESS of the microgrid 20 .

Next, a configuration of an anomaly detection model (FDM) for managing a microgrid distributed power supply using a plurality of sensors according to another embodiment of the present invention will be described. 12 is a diagram for explaining the configuration of an anomaly detection model (FDM) according to another embodiment of the present invention. 12 , the anomaly detection model (FDM) includes an encoder (EN), a hidden layer (HL), and a decoder (DE).

The encoder EN and the decoder DE include at least one layer including a plurality of nodes. Here, the layers and hidden layers HL included in the encoder EN and the decoder DE are fully connected layers. Each of the plurality of nodes included in the fully-connected layer performs an operation by an activation function. The activation function may be exemplified by Sigmoid, Hyperbolic tangent (tanh), Exponential Linear Unit (ELU), Rectified Linear Unit (ReLU), Leakly ReLU, Maxout, Minout, Softmax, and the like.

As described above, the anomaly detection model (FDM) includes a plurality of layers, and the plurality of layers includes a plurality of operations, that is, an operation by an activation function. In addition, a plurality of layers are connected by a weight (w: weight). The calculation result of one layer is weighted and becomes the input of the node of the next layer. That is, any one layer of the anomaly detection model (FDM) receives a weighted value from the previous layer, performs an operation on it, that is, an operation by an activation function, and delivers the operation result to the input of the next layer. .

When a measurement vector (m) is input, the encoder (EN) performs a plurality of operations in which a plurality of inter-layer weights are applied to the input measurement vector (or a measurement vector for learning: m) to perform a plurality of operations to apply a hidden vector (or a hidden vector for learning). , Hidden Vector: H) is calculated, and a hidden layer (HL) is created with this hidden vector.

When the hidden vector (or latent vector for learning: H) output from the encoder (EN) is input, the decoder DE performs a plurality of operations in which a plurality of inter-layer weights are applied to the hidden vector (or hidden vector for learning: H). A simulation measurement vector (or a simulation measurement vector for learning: m') that simulates a measurement vector (or a measurement vector for learning: m) is generated by performing it.

Next, a method for generating an anomaly detection model (FDM) according to another embodiment of the present invention will be described. 13 is a flowchart illustrating a method of generating an anomaly detection model (FDM) according to another embodiment of the present invention.

12 and 13 , the model generator 110 initializes the abnormality detection model FDM in step S310 . This initialization means to initialize a parameter, that is, the weight w, of the anomaly detection model FDM according to the second embodiment of FIG. 12 . When the initialization is completed, the model generating unit 110 prepares a learning measurement vector m used for learning in the FDM initialized in step S320 . The measurement vector (m) for learning is collected from the microgrid 20, and at least one of the measurement data measured from the renewable energy generator in the steady state and the measurement data measured from the ESS in the steady state is mapped to a predetermined vector space. it will be created

Next, the model generation unit 110 inputs the measurement vector (m) for learning to the abnormality detection model (FDM) in step S330, and the abnormality detection model (FDM) is an encoder ( EN), the hidden layer (HL), and the decoder (DE) perform calculations to calculate the simulation measurement vector (m') for learning. These operations will be described in more detail as follows. First, the encoder EN performs a plurality of operations in which a plurality of inter-layer weights are applied on the measurement vector for learning (m) to calculate a hidden vector (H) for learning indicating the characteristics of the measurement vector for learning (m) do. The hidden layer (HL) is made of such a hidden vector (H, for example, h1, h2, h3) for learning. Next, the decoder DE calculates a simulation measurement vector m' for learning that simulates the measurement vector m for learning by performing a plurality of operations to which a plurality of inter-layer weights are applied on the hidden vector H for learning.

Next, the model generation unit 110 calculates a loss representing the difference between the training simulation measurement vector (m') and the training measurement vector (m) in step S350. In this case, the model generator 110 may calculate the loss according to Equation 4 below.

In Equation 4, f(m) is the same as Equation 4 below.

는 인코더(EN)의 가중치 행렬이고,

는 인코더(EN)의 바이어스(bias) 벡터를 의미한다. 또한,

는 인코더(EN)의 활성화 함수를 나타낸다. in Equation 5

is the weight matrix of the encoder (EN),

denotes a bias vector of the encoder EN. In addition,

denotes the activation function of the encoder EN.

Also, in Equation 4, g(f(m)) is the same as in Equation 5 below.

는 디코더(DE)의 가중치 행렬이고,

는 디코더(DE)의 바이어스(bias) 벡터를 의미한다. 또한,

는 디코더(DE)의 활성화 함수를 나타낸다. in Equation 6

is the weight matrix of the decoder (DE),

denotes a bias vector of the decoder DE. In addition,

denotes the activation function of the decoder DE.

및

는 예컨대, 시그모이드(Sigmoid), 하이퍼볼릭탄젠트(tanh: Hyperbolic tangent), ELU(Exponential Linear Unit), ReLU(Rectified Linear Unit), Leakly ReLU, Maxout, Minout, Softmax 등이 될 수 있다. As described above, the activation function

and

For example, sigmoid (Sigmoid), hyperbolic tangent (tanh: Hyperbolic tangent), ELU (Exponential Linear Unit), ReLU (Rectified Linear Unit), Leakly ReLU, Maxout, Minout, Softmax, etc. may be.

이 최소화되도록 역전파(Back-propagation) 알고리즘을 통해 이상검출모델(FDM)의 가중치 및 바이어스 벡터를 포함하는 매개변수를 갱신하는 최적화를 수행한다. Then, the model generation unit 110 is lost in step S360

Optimization is performed to update parameters including weights and bias vectors of the anomaly detection model (FDM) through a back-propagation algorithm so that this is minimized.

The above-described steps S320 to S360 are repeated using a plurality of different training measurement vectors (m) or a plurality of different training measurement vectors (m) batches to update the parameters of the anomaly detection model (FDM). . And this repetition is made until the loss (Equation 4) becomes less than a preset target value. Therefore, the model generator 110 determines whether the loss is less than or equal to a preset target value in step S370, and if all losses are less than or equal to the preset target value, learning is completed.

)와, 비정상 상태의 계측 데이터인 비정상 데이터를 정상으로 판단하는 제2종 오류(Type II error,

)에 대한 비용함수를 설정한다. 이러한 비용함수는 다음의 수학식 7과 같다. When the learning is completed, the model generator 110 sets a threshold for classifying normal data and abnormal data in step S380 . At this time, the model generation unit 110 determines the normal data, which is the measurement data in the normal state, as abnormal, a Type I error (Type I error,

) and Type II error (Type II error,

) to set the cost function. This cost function is expressed in Equation 7 below.

는 특정 임계점(

)값에 따라 계산되는 제1종 오류 및 제2종 오류 전체의 비용함수를 나타낸다. 여기서,

는 특정 임계점(

)값을 기준으로 한 이상검출모델(FDM)에 정상 데이터가 입력되었을 때 비정상 데이터인 것으로 예측하는 경우의 오차제곱합평균(MSE: Mean Squared Error)인 제1종 오류의 오차 값이고,

는 제1종 오류에 대한 가중치이다.

is a certain critical point (

) represents the cost function of the total type 1 error and type 2 error calculated according to the value. here,

is a certain critical point (

) is the error value of the type I error, which is the mean squared error (MSE) when normal data is input to the anomaly detection model (FDM) based on the value and is predicted to be abnormal data,

is the weight for the type I error.

는 특정 임계점(

)값을 기준으로 한 이상검출모델(FDM)에 비정상 데이터가 입력되었을 때 정상 데이터인 것으로 예측하는 경우의 오차제곱합평균(MSE)인 제2종 오류의 오차 값이고,

는 제2종 오류에 대한 가중치이다. In addition,

is a certain critical point (

) is the error value of the type 2 error, which is the mean sum of squares error (MSE) when abnormal data is input to the anomaly detection model (FDM) based on the value and is predicted to be normal data,

is the weight for the type II error.

를 최소로 하는

를 결정하여 임계치를 구할 수 있다. 또한 가중치(

)의 값을 조정하여 제1종 오류 와 제 2종 오류중 사용자가 원하는 중요도에 따라 가중치를 조정 할 수 있다.The model generation unit 110 is a cost function

to minimize

can be determined to obtain a threshold. Also, the weight (

), the weight can be adjusted according to the level of importance desired by the user among the type 1 error and type 2 error.

When the learning of the abnormality detection model (FDM) is completed according to the procedure as in FIG. 13 and the threshold is set, the abnormality of the state of the microgrid 20 using the abnormality detection model (FDM), that is, a renewable energy generator and at least one state abnormality of the ESS may be detected. These methods will be described. 14 is a flowchart illustrating a method for detecting an abnormality in a state of a microgrid using a plurality of sensors according to another embodiment of the present invention.

Referring to FIG. 14 , the data processing unit 120 may receive measurement data from the microgrid 20 in step S210 . Such measurement data includes measurement data measured from the renewable energy generator and measurement data measured from the ESS.

Then, the data processing unit 120 maps the measurement data to a predetermined vector space in step S420 to generate the measurement vector m. Then, the data processing unit 120 inputs the measurement vector (m) to the abnormality detection model (FDM) that has been trained in step S430. This anomaly detection model (FDM) has been trained as described above with reference to FIG. 13 .

When the measurement vector (m) is input, the anomaly detection model (FDM) simulates the measurement vector (m) in step S240 to calculate a pseudo-sensor vector. The step S240 will be described in more detail as follows.

First, the encoder EN of the anomaly detection model FDM calculates a latent vector z by performing a plurality of operations in which a plurality of inter-layer weights are applied to the measurement vector m. Next, the decoder DE of the anomaly detection model FDM performs a plurality of operations in which a plurality of inter-layer weights are applied to the latent vector z to simulate the measurement vector m by performing a simulated measurement vector (m') to calculate

The anomaly detection unit 130 determines whether the loss representing the difference between the simulated measurement vector (m') and the measurement vector (m) exceeds the previously derived threshold ( S380 ) in step S450 . As a result of the determination in step S450, if the loss is greater than or equal to the threshold, the abnormality detection unit 130 determines that an abnormality has occurred in the corresponding microgrid 20 in step S460, and the corresponding operator device 30 to check the microgrid 20. A check message can be sent to In this step S460, when the loss is greater than or equal to the threshold and the difference from the threshold is greater than or equal to a predetermined value, the abnormality of the microgrid 20 is determined to be urgent, and the microgrid 20 can be directly controlled. For example, the abnormality detection unit 130 may stop the renewable energy generator and the ESS of the microgrid 20 .

15 is a diagram illustrating a computing device according to an embodiment of the present invention. The computing device TN100 of FIG. 15 may be a device described herein, for example, the management server 10 , the operator device 30 , the user device 50 , and the like.

In the embodiment of FIG. 15 , the computing device TN100 may include at least one processor TN110 , a transceiver device TN120 , and a memory TN130 . In addition, the computing device TN100 may further include a storage device TN140 , an input interface device TN150 , an output interface device TN160 , and the like. Components included in the computing device TN100 may be connected by a bus TN170 to communicate with each other.

The processor TN110 may execute a program command stored in at least one of the memory TN130 and the storage device TN140. The processor TN110 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to an embodiment of the present invention are performed. The processor TN110 may be configured to implement procedures, functions, and methods described in connection with an embodiment of the present invention. The processor TN110 may control each component of the computing device TN100.

Each of the memory TN130 and the storage device TN140 may store various information related to the operation of the processor TN110. Each of the memory TN130 and the storage device TN140 may be configured as at least one of a volatile storage medium and a nonvolatile storage medium. For example, the memory TN130 may include at least one of a read only memory (ROM) and a random access memory (RAM).

The transceiver TN120 may transmit or receive a wired signal or a wireless signal. The transceiver TN120 may be connected to a network to perform communication.

Meanwhile, the method according to the embodiment of the present invention described above may be implemented in the form of a program readable by various computer means and recorded in a computer readable recording medium. Here, the recording medium may include a program command, a data file, a data structure, etc. alone or in combination. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floppy disks ( magneto-optical media) and hardware devices specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions may include high-level languages that can be executed by a computer using an interpreter or the like as well as machine language such as generated by a compiler. Such hardware devices may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

Although the present invention has been described above using several preferred embodiments, these examples are illustrative and not restrictive. As such, those of ordinary skill in the art to which the present invention pertains will understand that various changes and modifications can be made in accordance with the doctrine of equivalents without departing from the spirit of the present invention and the scope of rights set forth in the appended claims.

10: management server
20: microgrid
30: operator device
40: smart meter
50: user device
100: grid management unit
110: model generation unit
120: data processing unit
130: abnormality detection unit
200: prepaid transaction department

Claims (8)

An apparatus for managing a microgrid distributed power supply, comprising:
When the measurement data is received from the microgrid, the received measurement data is mapped to a predetermined vector space to generate a measurement vector, and a data processing unit that inputs the generated measurement vector to an anomaly detection model that is a deep learning model and includes an encoder and a decoder ; and
When the anomaly detection model calculates a simulated measurement vector by simulating the measurement vector, it is determined whether a loss representing a difference between the measurement vector and the simulated measurement vector exceeds a preset threshold, and if the loss is greater than or equal to the threshold, , an abnormality detection unit that determines that there is an abnormality in the microgrid;
includes,
Prepare a measurement vector for learning from the measurement data measured in the steady state of the microgrid,
When the anomaly detection model calculates the measurement vector for learning by performing an operation on the measurement vector for learning,
a model generator for learning an anomaly detection model by calculating a loss based on the measurement vector for training and the simulated measurement vector for training, and performing optimization to update parameters of the anomaly detection model so that the loss is minimized;
further comprising,
The model generation unit
formula

Calculate the loss according to
remind

represents the loss,

ego,

ego,
remind

is the weight matrix of the encoder,
remind

is the bias vector of the encoder,
remind

is the activation function of the encoder,
remind

is the weight matrix of the decoder,
remind

is the bias vector of the decoder,
remind

is the activation function of the decoder, characterized in that
A device for managing microgrid distributed power. delete delete According to claim 1,
The model generation unit
After learning is complete,
Setting a cost function for a first error for determining normal data that is measurement data in a normal state as abnormal and a second error for determining abnormal data that is measurement data in an abnormal state as normal,
characterized by setting a threshold at which the cost function is minimized
A device for managing microgrid distributed power. A method for managing a microgrid distributed power supply, comprising:
receiving, by the data processing unit, measurement data from the microgrid;
generating a measurement vector by mapping the received measurement data to a predetermined vector space by the data processing unit, and inputting the generated measurement vector into an anomaly detection model that is a deep learning model;
calculating, by the anomaly detection model, a simulated measurement vector by simulating the measurement vector;
determining, by an anomaly detection unit, whether a loss representing a difference between the measurement vector and the simulated measurement vector exceeds a preset threshold; and
determining, by the abnormality detection unit, that there is an abnormality in the microgrid when the loss is greater than or equal to the threshold;
includes,
Before the data processing unit receiving the measurement data from the microgrid,
preparing, by the model generation unit, a measurement vector for learning from measurement data measured in a steady state of the microgrid;
calculating, by an anomaly detection model, a simulated measurement vector for learning by performing an operation on the measurement vector for learning;
calculating, by a model generator, a loss based on the training measurement vector and the training simulation measurement vector; and
learning the anomaly detection model by performing, by a model generator, optimization of updating the parameters of the anomaly detection model so that the loss is minimized;
further comprising,
The step of calculating the loss is
The model generation unit
formula

Calculate the loss according to
remind

represents the loss,

ego,

ego,
remind

is the weight matrix of the encoder,
remind

is the bias vector of the encoder,
remind

is the activation function of the encoder,
remind

is the weight matrix of the decoder,
remind

is the bias vector of the decoder,
remind

is the activation function of the decoder, characterized in that
A method for managing microgrid distributed power. delete delete 6. The method of claim 5,
After the step of learning the anomaly detection model,
Setting a cost function for a type 1 error that determines normal data that is measurement data in a normal state as abnormal and a type 2 error that determines abnormal data that is measurement data in an abnormal state as normal,
setting a threshold at which the cost function is minimized;
characterized in that it further comprises
A method for managing microgrid distributed power.

Images