KR102533023B1 - Apparatus for measuring solar power generation based on Deep Leaning Model and method therefor - Google Patents

Abstract

A method for measuring solar power generation based on a deep learning model is provided. The method includes generating a thermal image including a plurality of frames captured by a thermal image camera by capturing a photovoltaic cell array in units of a predetermined time, and generating a power generation prediction network using a weight learned between a plurality of layers for the thermal image. Calculating a power generation vector that predicts the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of applied operations, and an efficiency estimation network to which a weight learned between a plurality of layers is applied to the thermal image Calculating a power generation efficiency vector that predicts the power generation efficiency representing the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of operations; and calculating an effective power generation vector, which predicts the amount of power generated by the photovoltaic power generation of the photovoltaic cell array and provided to the load, from the power generation efficiency vector.

Description

Apparatus for measuring solar power generation based on Deep Leaning Model and method therefor}

The present invention relates to solar measurement technology, and more particularly, to a device for measuring the amount of solar power actually delivered to a load based on a deep learning model and a method therefor.

Photovoltaic system (solar power system) is a power generation method that produces electricity by converting sunlight into direct current electricity. Solar power generation uses a photovoltaic panel in which several solar cells are attached. As the demand for renewable energy increases, the production of solar cells and photovoltaic arrays is also increasing significantly.

Photovoltaic power generation refers to a power generation method that converts light from the sun into electrical energy using a photovoltaic effect. It is different from solar power generation that uses the thermal energy of light in that it directly converts light energy into electrical energy.

The photovoltaic effect is almost similar to the photoelectric effect, but the situation in which it is explained is slightly different. If the photoelectric effect generally refers to an effect in which a certain material receives light and emits electrons, the photovoltaic effect refers to the movement of electrons and holes resulting from the photoelectric effect inside a material to create a potential difference. That is, the photovoltaic effect can be regarded as a side effect that can be considered as one result of the photoelectric effect.

Korean Registered Patent No. 2030925 Registered on October 2, 2019 (Name: Solar power generation monitoring system)

An object of the present invention is to provide an apparatus and method for measuring solar power generation based on a deep learning model.

In order to achieve the above object, a method for measuring solar power generation based on a deep learning model according to a preferred embodiment of the present invention captures a solar cell array with a thermal imaging camera and captures a plurality of images taken at predetermined time units. The step of generating a thermal image including a frame, and the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of calculations to which weights learned between a plurality of layers are applied to the thermal image by the power generation prediction network. Calculating the predicted power generation vector, and the efficiency estimation network calculates the load versus the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of calculations in which weights learned between a plurality of layers are applied to the thermal image. Calculating a power generation efficiency vector that predicts power generation efficiency representing a ratio of the amount of power supplied; and an integrated prediction network is generated by the photovoltaic power generation of the photovoltaic cell array from the power generation amount vector and the power generation efficiency vector and provided to the load. and calculating an effective power generation vector by predicting the amount of power to be generated.

The step of calculating the effective power generation vector includes calculating a weight vector by the integrated prediction network multiplying the power generation vector and the power generation efficiency vector, and the integrated prediction network calculating the power generation vector, the power generation efficiency vector, and the weight vector. and calculating the effective power generation vector by performing an operation to which a learned weight is applied.

Before the step of generating the thermal image, the learning unit calculates the power generation amount predicting the amount of power generated by the photovoltaic power generation of the photovoltaic cell array from the thermal image obtained by the thermal image camera unit. The power generation efficiency vector predicting the power generation efficiency representing the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array from the same thermal image in which the network is individually learned and the power generation prediction network is learned individually learning the efficiency estimation network to calculate; and integrating and learning the efficiency estimation network and the integrated prediction network.

The integrating and learning step includes the step of preparing the thermal image for learning by the learning unit, the generation amount, which is the amount of power output from the photovoltaic cell array through the sensor unit during a predetermined unit period during which the learning unit took the learning thermal image, and the input to the load. Measuring effective power generation, which is the amount of power to be generated, and setting the measured power generation and effective power generation as labels for a learning thermal image; setting as the actual power generation efficiency vector, calculating the power generation vector through a plurality of calculations in which weights of a plurality of layers are applied to the learning thermal image by the power generation prediction network, and the integrated prediction network calculating the power generation vector and the actual measurement Calculating a weight vector by multiplying the power generation efficiency vector and calculating an effective power generation prediction vector from the power generation vector, the weight vector, and the actually measured power generation efficiency vector; and optimizing weights of the power generation prediction network, the efficiency estimation network, and the integrated prediction network so that a total loss including an effective power generation loss, which is a difference between the effective power generation vector and the effective power generation, is minimized.

를 통해 상기 전체 손실이 최소가 되도록 상기 발전량예측망, 상기 효율추정망 및 상기 통합예측망의 가중치를 최적화하며, 상기 L은 상기 발전량벡터와 상기 발전량과의 차이를 나타내는 평균 제곱 오차 손실인 발전량 손실과, 상기 실효발전량벡터와 상기 실효발전량과의 차이를 나타내는 평균 제곱 오차 손실인 실효 발전량 손실을 포함하는 전체 손실이고, 상기 GC는 상기 발전량벡터이고, 상기 gc는 상기 발전량이고, 상기 EGC는 상기 실효발전량벡터이고, 상기 egc는 상기 실효발전량이고, 상기 N은 학습용 열화상의 프레임의 수이며, 상기 α는 발전량 손실에 대한 가중치이고, 상기 β는 실효 발전량 손실에 대한 가중치인 것을 특징으로 한다. Optimizing the weights is performed by the learning unit using a loss function

Weights of the power generation prediction network, the efficiency estimation network, and the integrated prediction network are optimized so that the total loss is minimized through and an effective power generation loss, which is a mean square error loss representing a difference between the effective power generation vector and the effective power generation, wherein GC is the power generation vector, gc is the power generation, and EGC is the effective A power generation vector, egc is the effective power generation, N is the number of frames of a thermal image for learning, α is a weight for power generation loss, and β is a weight for effective power generation loss.

After the step of calculating the effective power generation vector, the state information providing unit receiving the amount of power generation and the effective power generation measured by the sensor unit, the step of calculating the power generation efficiency from the power generation amount and the effective power generation amount by the state information providing unit, and The information providing unit may further include comparing the power generation amount vector, the power generation efficiency vector, and the effective power generation amount vector with the power generation amount, the power generation efficiency, and the effective power generation amount, and checking whether or not the photovoltaic power generation system is abnormal.

An apparatus for measuring solar power generation based on a deep learning model according to a preferred embodiment of the present invention for achieving the above object is to photograph a solar cell array and include a plurality of frames photographed in units of predetermined time A power generation vector obtained by predicting the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a thermal imaging camera unit that generates a thermal image and a plurality of calculations to which weights learned between a plurality of layers are applied to the thermal image. It represents the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of calculations in which the calculated generation amount prediction network and the weight learned between the plurality of layers are applied to the thermal image. An efficiency estimation network that calculates a power generation efficiency vector that predicts power generation efficiency, and an effective power generation vector that predicts the amount of power generated by the photovoltaic power generation of the photovoltaic cell array and provided to the load from the power generation vector and the power generation efficiency vector Includes an integrated prediction network that calculates

According to the present invention, after optimizing the weight of the deep neural network through deep learning, it is possible to predict the amount of solar power generation through the deep neural network. Accordingly, it is possible to check whether or not the photovoltaic system has an abnormality by comparing the measured value with the actual measured value.

1 is a diagram for explaining the configuration of a photovoltaic power generation system according to an embodiment of the present invention.
2 is a diagram for explaining the configuration of a measuring device for measuring sunlight according to an embodiment of the present invention.
3 is a diagram for explaining a detailed configuration of a thermal imaging camera unit of a measuring device according to an embodiment of the present invention.
4 is a diagram for explaining a detailed configuration of a measuring unit of a measuring device according to an embodiment of the present invention.
5 is a diagram for explaining a detailed configuration of an integrated prediction network of a measurement unit according to an embodiment of the present invention.
6 is a flowchart for explaining a method for learning a deep neural network according to an embodiment of the present invention.
7 is a flowchart illustrating integrated learning for a deep neural network according to an embodiment of the present invention.
8 is a flowchart illustrating a method of measuring solar power generation using a trained deep neural network according to an embodiment of the present invention.
9 is a diagram illustrating a computing device that performs a computing operation according to an embodiment of the present invention.

Prior to the detailed description of the present invention, the terms or words used in this specification and claims described below should not be construed as being limited to a common or dictionary meaning, and the inventors should use their own invention in the best way. It should be interpreted as a meaning and concept corresponding to the technical idea of the present invention based on the principle that it can be properly defined as a concept of a term for explanation. Therefore, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all of the technical ideas of the present invention, so various equivalents that can replace 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. At this time, it should be noted that the same components in the accompanying drawings are indicated 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 in the accompanying drawings are exaggerated, omitted, or schematically illustrated, and the size of each component does not entirely reflect the actual size.

First, the configuration of a photovoltaic system according to an embodiment of the present invention will be described. 1 is a diagram for explaining the configuration of a photovoltaic power generation system according to an embodiment of the present invention.

Referring to FIG. 1, a photovoltaic power generation system according to an embodiment of the present invention includes a solar cell array (SCA), a junction box (JB), an inverter (IV: inverter), a load, and It includes a measuring device (10).

The solar cell array (SCA) includes a plurality of solar cells (SC: Solar Cell). Each of the plurality of photovoltaic cells SC produces electrical energy from light coming from the sun by using a photovoltaic effect.

The junction box JB manages a plurality of photovoltaic cells SC and transmits direct current, which is electrical energy produced by the plurality of photovoltaic cells SC.

The inverter (IV) converts the current flowing in from the junction box into an alternating current and outputs it to a load (L), for example, a customer.

The measurement device 10 continuously takes thermal images of the solar cell array (SCA) and predicts the amount of power generated through photovoltaic power generation for a unit period through a thermal image including a plurality of frames captured during a unit period. .

Next, the configuration of the measuring device 10 for measuring sunlight according to an embodiment of the present invention will be described in detail. 2 is a diagram for explaining the configuration of a measuring device for measuring sunlight according to an embodiment of the present invention. 3 is a diagram for explaining a detailed configuration of a thermal imaging camera unit of a measuring device according to an embodiment of the present invention. 4 is a diagram for explaining a detailed configuration of a measuring unit of a measuring device according to an embodiment of the present invention. 5 is a diagram for explaining a detailed configuration of an integrated prediction network of a measurement unit according to an embodiment of the present invention.

First, referring to FIG. 2 , the measurement device 10 for measuring sunlight according to an embodiment of the present invention includes a thermal imaging camera unit 100, a measurement unit 200, a learning unit 300, and a sensor unit 400. and a state information providing unit 500 .

Referring to FIG. 3 , the thermal imaging camera unit 100 captures a solar cell array (SCA) and outputs a thermal image. The thermal imaging camera unit 100 includes a thermal imaging lens 101, a thermal imaging sensor 102, and an A/D converter (Analog to Digital Converter 103). In addition, a predetermined filter may be further included in the configuration of the thermal imaging camera unit 100, and mechanically, the thermal imaging lens 101, the thermal imaging sensor 102, and the A/D converter 103 are actuators A driver or the like mounted in a housing including an actuator and driving the actuator may be included in the thermal imaging camera unit 100, but illustration and description thereof are omitted to clarify the subject matter of the present invention. However, those skilled in the art will be able to repeatedly practice the present invention without explanation of non-described configurations.

The thermal imaging lens 101 focuses infrared rays incident on the thermal imaging camera unit 100 on the thermal imaging sensor 102 . Infrared light has the same optical properties as visible light, such as reflection, refraction, and transmission. However, glass used for a general optical device of a camera, that is, a lens, cannot be used because its material does not pass infrared rays well. Conversely, materials that pass infrared rays well do not pass visible rays well. Therefore, it is preferable to use silicon (Si) and germanium (Ge) as the material of the thermal imaging lens 101. Silicon is suitable for the mid-wavelength range, and germanium is suitable for the long-wavelength infrared. An antireflection coating may be applied to the thermal imaging lens 101 .

The thermal image sensor 102 detects the amount of infrared energy and outputs the value. The thermal image sensor 102 can be classified into two types: a thermal detector and a quantum detector. Representative examples of the thermal sensor include an uncooled microbolometer made of a metal or semiconductor material. Quantum sensors are made of InSb, InGaAs, PtSi, HgCdTe (MCT), etc., and a GaAs/AlGaAs layer is formed to form a QWIP (Quantum Well Infrared Photon) sensor. Quantum sensors are based on the phenomenon that the state of an electron in a crystal is changed by an incident photon. The quantum sensor must be cooled to extremely low temperatures using liquid nitrogen or a small Stirling freezer, including such a cooler. The thermal image sensor 102 is a component corresponding to an image sensor of a general color image camera, for example, a charge-coupled device (CCD) or a complementary metal-oxide semiconductor (CMOS). While general image sensors detect visible light, the thermal image sensor 102 is a focal plane array (FPA) composed of micrometer-sized pixels made of a material capable of detecting infrared wavelengths. . The resolution of the FPA, which is the thermal image sensor 102, preferably ranges from 160×120 pixels to 1024×1024 pixels, but is not limited thereto.

In other words, the thermal image sensor 102 detects electromagnetic radiation in an infrared wavelength range, derives the magnitude of infrared energy, and outputs the value. At this time, the thermal image sensor 102 outputs the value for each pixel determined according to the resolution of the FPA.

The field of view (FOV) of the thermal imaging camera unit 100 is determined according to specifications of the thermal imaging lens 101 and the thermal imaging sensor 102 . Here, the field of view means a vertical and horizontal range of a subject or scene that can be photographed by the thermal imaging camera unit 100 . In addition, an Instantaneous Field of View (IFOV) of each pixel is determined according to the specifications of the thermal image sensor 102, that is, the resolution. Here, the pixel field of view (IFOV) means a vertical and horizontal range of a subject or scene that can be sensed by one pixel of the thermal image sensor 102 . In an embodiment of the present invention, one pixel of the thermal image sensor 102 will be referred to as a 'thermal pixel'. In addition, a value representing the amount of infrared energy output from each 'thermal pixel' of the thermal image sensor 152 will be referred to as a 'thermal pixel value', and an image formed through these thermal pixel values will be referred to as a thermal image. . In addition, the thermal pixel values output by the thermal image sensor 152 are analog signals, and the A/D converter 103 converts the thermal pixel values, which are analog signals, into digital signals to have the thermal pixel values converted into digital signals. Thermal infrared image (TII) is output.

Referring to FIG. 4 , the measurement unit 200 analyzes a thermal image taken by the thermal imaging camera unit 100 to predict the amount of power generation. To this end, the measurement unit 200 includes a power generation prediction network 210, an efficiency estimation network 220, and an integrated prediction network 230. Here, each of the power generation prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230 is a deep learning model and is composed of a deep neural network (DNN). A deep neural network (DNN) may include a convolutional neural network (CNN), a recurrent neural network (RNN), a long short term memory model (LTSM), and the like. Each of the power generation prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230 is composed of a plurality of layers, and the plurality of layers are connected by weights, and each of the plurality of layers includes a plurality of operations. .

The power generation prediction network 210 generates a power generation vector (GC) for estimating the solar power generation for a corresponding period from a thermal image including a plurality of frames in which the thermal imaging camera unit 100 captures the solar cell array (SCA) for a predetermined period. yields The generation amount vector (GC) is generated by photovoltaic power generation of the solar cell array (SCA) and predicts the amount of power input to the junction box (JB).

The efficiency estimation network 220 estimates the power generation efficiency of photovoltaic power generation for a corresponding period from a thermal image including a plurality of frames captured by the thermal imaging camera unit 100 of the solar cell array (SCA) for a predetermined period of time. Calculate the vector (EF).

The integrated forecasting network 230 is based on the power generation vector (GC) input from the power generation prediction network 210 and the power generation efficiency vector (EF) input from the efficiency estimation network 220. An effective power generation vector (EGC) predicting an effective value of power generation is calculated. The effective power generation vector (EGC) is generated by photovoltaic power generation of the photovoltaic cell array (SCA) and predicts the amount of power provided to the load (L) via the junction box (JB) and the inverter (IV).

Accordingly, power generation efficiency may be calculated according to Equation 1 below.

In Equation 1, ef denotes power generation efficiency, gc denotes power generation amount, and egc denotes effective power generation amount.

In other words, the power generation efficiency vector (EF) according to an embodiment of the present invention is an estimation of the power generation efficiency according to Equation 1.

Meanwhile, referring to FIG. 5 , the integrated predictive network 230 includes a multiplier 231 , a merging layer 232 and an output layer 233 .

The multiplier 231 calculates a weight vector (WV) by multiplying the generation amount vector (GC) and the generation efficiency vector (EF) (GC × EF = WV).

Accordingly, the merging layer 232 may receive the power generation amount vector GC, the weight vector WV, and the power generation efficiency vector EF. Accordingly, the output layer 233 performs a plurality of calculations in which weights learned between a plurality of layers are applied to the power generation amount vector (GC), weight vector (WV), and power generation efficiency vector (EF), thereby generating the solar cell array (SCA). An effective power generation vector (EGC) is calculated by estimating the amount of power generated by photovoltaic power generation and supplied to the load (L) through the junction box (JB) and the inverter (IV).

Again, referring to FIG. 2 , the learning unit 300 is for deep learning the generation amount prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230. To this end, the learning unit 300 first performs individual learning to individually learn each of the power generation prediction network 210 and the efficiency estimation network 220, and then the power generation prediction network 210 and the efficiency estimation network 220 and integrated learning to integrate and learn the entire integrated prediction network 230.

During individual learning, the learning unit 300 predicts the amount of power generated by the photovoltaic power generation of the solar cell array (SCA) from the thermal image captured by the thermal imaging camera unit 100 of the solar cell array (SCA), a power generation amount vector. The power generation prediction network 210 may be trained to calculate (GC). In addition, during individual learning, the learning unit 300 compares the actual load to the amount of power generated by the photovoltaic power generation of the solar cell array (SCA) from the thermal image taken by the thermal imaging camera unit 100 of the solar cell array (SCA). The efficiency estimation network 220 may be trained to calculate a power generation efficiency vector EF, which predicts the power generation efficiency representing the ratio of the amount of power supplied to (L).

At the time of integrated learning, the learning unit 300 uses the power generation prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230 to generate the actual load ( The power generation prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230 may be integrated and trained to calculate the effective generation amount (EGC) by predicting the amount of power supplied to L).

The sensor unit 400 is for measuring the amount of power. To this end, the sensor unit 400 includes a first sensor 401 and a second sensor 402 . As shown in FIG. 1, the first sensor 401 is installed at the output terminal of the solar cell array (SCA), and measures the power generation amount (gc), which is the amount of power generated by the photovoltaic power generation of the solar cell array (SCA). can In addition, the second sensor 402 is installed at the input terminal of the load (L), and the amount of power generated by photovoltaic power generation of the solar cell array (SCA) is the amount of power actually supplied to the load (L), which is the effective power generation amount (egc) can be measured.

The status information providing unit 500 is for checking whether or not the photovoltaic system is abnormal. The state information providing unit 500 receives the amount of power generation gc and the amount of effective power generation egc measured by the sensor unit 400, and calculates the power generation efficiency ef based on the amount of power generation gc and the amount of effective power generation egc. After that, the power generation vector (GC), power generation efficiency vector (EF), and effective power generation vector (EGC) derived by the measurement unit 200, the measured power generation (gc), the calculated power generation efficiency (ef), and the measured effective power generation (egc) can be compared to check whether the photovoltaic system is abnormal.

Next, an apparatus and method for measuring solar power generation based on a deep learning model according to an embodiment of the present invention will be described. In order to measure solar power generation based on this deep learning model, deep learning of the deep neural networks 210, 220, and 230 must be preceded. This learning method (deep leaning) will be described. 6 is a flowchart for explaining a method of learning deep neural networks 210, 220, and 230 according to an embodiment of the present invention.

Referring to FIG. 6 , the learning unit 300 performs individual learning in step S110 in which each of the power generation prediction network 210 and the efficiency estimation network 220 is separately trained. During this individual learning, the learning unit 300 predicts the amount of electricity generated by the solar power of the solar cell array (SCA) from the thermal image captured by the thermal imaging camera unit 100. The power generation prediction network 210 is individually trained to calculate the vector GC, and the actual load compared to the amount of power generated by the photovoltaic power generation of the solar cell array (SCA) from the same thermal image on which the power generation prediction network 210 is trained. The efficiency estimation network 220 may be trained to calculate a power generation efficiency vector EF, which predicts the power generation efficiency representing the ratio of the amount of power supplied to (L).

Next, the learning unit 300 predicts the amount of power generation to calculate the effective power generation amount (EGC) by predicting the amount of power supplied to the actual load (L) among the amount of power generated by photovoltaic power generation of the solar cell array (SCA) in step S120. The network 210, the efficiency estimation network 220, and the integrated prediction network 230 can be integrated and trained.

Then, the above-described integrated learning (S120) will be described in more detail. 7 is a flowchart illustrating integrated learning for a deep neural network according to an embodiment of the present invention.

Referring to FIG. 7 , the learning unit 300 prepares a thermal image for learning in step S210. The thermal image for learning includes a plurality of frames obtained by photographing the solar cell array (SCA) for a predetermined unit period through the thermal imaging camera unit 100 . Next, the learning unit 300 determines the amount of electricity generated from the solar cell array (SCA) through the first sensor 401 and the second sensor 402 during a predetermined unit period during which the thermal image for learning was taken in step S220, The amount of power input to the load (L), the effective power generation, is measured, and the measured power generation (gc) and effective power generation (egc) are set as labels for the thermal image for training. That is, the measured generation amount gc is set as the generation amount label, and the measured effective generation amount egc is set as the effective generation amount label. Then, the learning unit 300 calculates the power generation efficiency (ef) according to Equation 1 based on the power generation amount (gc) and the effective power generation amount (egc) previously measured in step S230, and calculates the power generation efficiency (ef) Set as the actual power generation efficiency vector (GTEF).

Then, the learning unit 300 inputs the thermal image for learning to the measurement unit 200 in step S240. Then, the power generation prediction network 210 of the measuring unit 200 calculates the power generation vector GC through a plurality of calculations to which weights of a plurality of layers are applied to the training thermal image in step S250.

Subsequently, the learning unit 300 inputs the measured power generation efficiency vector (GTEF) to the integrated prediction network 230 in step S260. Then, the integrated prediction network 230 calculates the weight vector (WV) by multiplying the power generation vector (GC) and the actual power generation efficiency vector (GTEF) in step S270, and generates the power vector (GC) and weight vector (WV) in step S280 and an effective power generation prediction vector (EGC) is calculated from the actually measured power generation efficiency vector (GTEF).

Then, the learning unit 300 calculates the power generation loss, which is the difference between the power generation vector (GC) calculated in step S290 and the measured power generation (gc), the calculated effective power generation vector (EGC) and the measured effective power generation (egc) The weights of the power generation prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230 are optimized so that the total loss including the effective power generation loss, which is the difference between , is minimized.

At this time, the learning unit 300 optimizes the weights of the generation amount prediction network 210, the efficiency estimation network 220, and the integrated prediction network 230 so that the total loss is minimized through a loss function such as Equation 2 below. .

In Equation 2, L is the power generation loss, which is the mean square error loss representing the difference between the calculated power generation vector (GC) and the measured power generation (gc), and the calculated effective power generation vector (EGC) and the measured effective power generation (egc ) represents the total loss including the effective power generation loss, which is the mean square error loss representing the difference from In Equation 2, GC is the calculated generation amount vector, and gc is the measured generation amount. In addition, EGC is the calculated effective power generation vector, and egc is the measured effective power generation amount. Here, N is the number of frames of the thermal image for training, α is a weight for loss in power generation, and β is a weight for loss in effective power generation. The weight α for the loss of power generation and the weight β for the loss of effective power generation may be set in advance.

As described above, when learning is completed, the amount of photovoltaic power generation can be measured using the learned deep neural network. 8 is a flowchart illustrating a method of measuring solar power generation using a trained deep neural network according to an embodiment of the present invention.

Referring to FIG. 8 , the thermal imaging camera unit 100 photographs a Solar Cell Array (SCA) including a plurality of solar cells (SC) in step S310, and takes a predetermined time unit in step S320. A thermal image (TII) including a plurality of frames photographed is input to the measuring unit 200 .

The power generation prediction network 210 of the measurement unit 200 measures the amount of power generated by photovoltaic power generation of the solar cell array (SCA) through a plurality of calculations to which weights learned between a plurality of layers are applied to the thermal image in step S330. Calculate the predicted power generation vector (GC). In parallel with this, the efficiency estimation network 220 of the measurement unit 200 performs a plurality of calculations in which weights learned between a plurality of layers are applied to the thermal image in step S340 to generate solar power of the solar cell array (SCA). Calculate a power generation efficiency vector (EF) predicting the power generation efficiency representing the ratio of the amount of power supplied to the actual load (L) to the amount of power generated by

Next, the integrated prediction network 230 of the measuring unit 200 multiplies the power generation vector (GC) and the power generation efficiency vector (EF) through the multiplier 231 in step S350 to calculate the weight vector (WV) (GC × EF =WV). Subsequently, the integrated prediction network 230 performs an operation in which the learned weights are applied to the generation amount vector (GC), the generation efficiency vector (EF) and the weight vector (WV) in step S360 to obtain the sunlight of the solar cell array (SCA). An effective generation amount vector (EGC) estimating the amount of power generated by power generation and provided to the load (L) is calculated.

At this time, the merging layer 232 of the integrated prediction network 230 receives the power generation vector (GC), power generation efficiency vector (EF) and weight vector (WV), and generates the power generation vector (GC), power generation efficiency vector (EF) and The learned weight is applied to the weight vector (WV) and inputted to the output layer 233. Then, the merging layer 232 of the integrated prediction network 230 performs an operation through an activation function on the power generation vector (GC), the power generation efficiency vector (EF), and the weight vector (WV) to which the learned weights are applied, thereby calculating the effective power generation Calculate the vector (EGC).

Accordingly, the measuring unit 200 provides the power generation vector (GC), power generation efficiency vector (EF), and effective power generation vector (EGC) calculated according to the above method to the state information providing unit 500, and the state information providing unit 500 receives the amount of power generation (gc) and the amount of effective power generation (egc) measured during the period in which the photovoltaic cell array (SCA) was photographed from the sensor unit 400 in step S370, and generates power generation efficiency (ef) according to Equation 1 After calculating , the power generation vector (GC), power generation efficiency vector (EF), and effective power generation vector (EGC) input from the measurement unit 200, the measured power generation amount (gc), the calculated power generation efficiency (ef), and the measured Compare the effective power generation (egc) to check whether there is an abnormality in the photovoltaic power generation system.

9 is a diagram illustrating a computing device that performs a computing operation according to an embodiment of the present invention. The computing device TN100 of FIG. 9 may be a device described in this specification (eg, the measuring device 10, etc.).

In the embodiment of FIG. 9 , the computing device TN100 may include at least one processor TN110, a transceiver TN120, and a memory TN130. In addition, the computing device TN100 may further include a storage device TN140, an input interface device TN150, and an output interface device TN160. Elements included in the computing device TN100 may communicate with each other by being connected by a bus TN170.

The processor TN110 may execute program commands 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 embodiments of the present invention are performed. Processor TN110 may be configured to implement procedures, functions, methods, and the like described in relation to embodiments 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 include at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory TN130 may include at least one of read only memory (ROM) and random access memory (RAM).

The transmitting/receiving device TN120 may transmit or receive a wired signal or a wireless signal. The transmitting/receiving device TN120 may perform communication by being connected to a network.

On the other hand, 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 on a computer-readable recording medium. Here, the recording medium may include program commands, data files, data structures, etc. alone or in combination. Program instructions recorded on the recording medium may be those specially designed and configured for the present invention, or those known and usable to those skilled in computer software. For example, recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks ( It includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media and ROM, RAM, flash memory, etc. Examples of the program command may include a high-level language that can be executed by a computer using an interpreter, as well as a machine language generated by a compiler. These hardware devices may be configured to act as one or more software modules to perform the operations of the present invention, and vice versa.

The present invention has been described above using several preferred examples, but these examples are illustrative and not limiting. As such, those skilled in the art to which the present invention belongs will understand that various changes and modifications can be made according to 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: measuring device
100: thermal imaging camera unit
101: thermal imaging lens
102: thermal image sensor
103: A/D converter
200: measurement unit
210: power generation prediction network
220: efficiency estimation network
230: integrated prediction network
231: multiplier
232 merge layer
233: output layer
300: learning unit
400: sensor unit
401: first sensor
402: second sensor
500: state information provision unit

Claims (7)

In the method for measuring solar power generation based on a deep learning model,
generating a thermal image including a plurality of frames photographed at predetermined time units by photographing the solar cell array by a thermal imaging camera;
Calculating a power generation vector predicting the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of calculations to which weights learned between a plurality of layers are applied to the thermal image by a power generation prediction network;
The efficiency estimation network calculates the power generation efficiency representing the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of operations in which weights learned between a plurality of layers are applied to the thermal image. Calculating a predicted generation efficiency vector; and
calculating, by an integrated prediction network, an effective power generation vector that predicts the amount of power generated by the photovoltaic power generation of the photovoltaic cell array and provided to the load from the power generation vector and the power generation efficiency vector;
Including,
Before the step of generating the thermal image,
The learning unit individually learns the power generation prediction network so that the thermal image camera unit calculates a power generation vector predicting the amount of power generated by the photovoltaic power generation of the photovoltaic cell array from the thermal image taken by the photovoltaic cell array,
The efficiency estimation network calculates a power generation efficiency vector that predicts the power generation efficiency representing the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array from the same thermal image on which the power generation prediction network has been trained. learning individually; and
Integrating and learning the power generation prediction network, the efficiency estimation network, and the integrated prediction network so that the learning unit calculates the effective power generation by predicting the amount of power supplied to the load among the amount of power generated by the photovoltaic power generation of the photovoltaic cell array ;
characterized in that it includes
A method for measuring solar power generation. According to claim 1,
The step of calculating the effective power generation vector
calculating, by the integrated prediction network, a weight vector by multiplying the power generation amount vector and the power generation efficiency vector; and
calculating, by the integrated prediction network, the effective power generation vector by performing an operation in which learned weights are applied to the power generation vector, the power generation efficiency vector, and the weight vector;
characterized in that it includes
A method for measuring solar power generation. delete According to claim 1,
The step of integrating and learning
preparing a thermal image for learning by the learning unit;
The learning unit measures the amount of power generation, which is the amount of power output from the solar cell array, and the amount of effective power generation, which is the amount of power input to the load, through the sensor unit during a predetermined unit period when the learning thermal image is captured, and the measured power generation amount and effective power generation amount are converted into thermal images for learning. Setting as a label for;
calculating power generation efficiency based on the amount of power generation and the amount of effective power generation by the learning unit, and setting the calculated power generation efficiency as an actual power generation efficiency vector;
Calculating a power generation vector through a plurality of calculations to which weights of a plurality of layers are applied to a training thermal image by a power generation prediction network;
calculating a weight vector by multiplying the power generation vector and the measured power generation efficiency vector by an integrated prediction network, and calculating an effective power generation prediction vector from the power generation vector, the weight vector, and the actually measured power generation efficiency vector;
The power generation prediction network and the efficiency estimation network such that the learning unit minimizes total loss including power generation loss, which is a difference between the power generation vector and the power generation amount, and effective power generation loss, which is a difference between the effective power generation vector and the effective power generation amount. and optimizing weights of the integrated prediction network.
characterized in that it includes
A method for measuring solar power generation. According to claim 4,
Optimizing the weights
the learning section
loss function

Optimizing the weights of the power generation prediction network, the efficiency estimation network, and the integrated prediction network so that the total loss is minimized through
L is a total loss including power generation loss, which is a mean square error loss representing a difference between the power generation vector and the power generation amount, and effective power generation loss, which is a mean square error loss representing a difference between the effective power generation vector and the effective power generation amount; ,
The GC is the generation amount vector, the gc is the generation amount,
The EGC is the effective power generation vector,
The egc is the effective power generation amount,
N is the number of frames of the thermal image for learning,
α is a weight for power loss,
The β is characterized in that the weight for the effective power generation loss
A method for measuring solar power generation. According to claim 1,
After calculating the effective power generation vector,
Receiving input of the amount of power generation and the amount of effective power generation measured by the sensor unit by the state information providing unit;
Calculating power generation efficiency from the power generation amount and the effective power generation amount by the state information providing unit;
comparing the power generation amount vector, the power generation efficiency vector, and the effective power generation amount vector with the power generation amount, the power generation efficiency, and the effective power amount, by the status information providing unit, to check whether or not the photovoltaic power generation system is abnormal;
characterized in that it further comprises
A method for measuring solar power generation. In the apparatus for measuring solar power generation based on a deep learning model,
a thermal imaging camera unit that photographs the photovoltaic cell array and generates a thermal image including a plurality of frames photographed at predetermined time intervals;
A power generation prediction network that calculates a power generation vector that predicts the power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of calculations to which weights learned between a plurality of layers are applied to the thermal image;
Power generation efficiency that represents the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array through a plurality of calculations to which weights learned between a plurality of layers are applied to the thermal image. Efficiency estimation network that calculates vectors; and
an integrated prediction network for calculating an effective power generation vector that predicts the amount of power generated by the photovoltaic power generation of the photovoltaic cell array and provided to the load from the power generation vector and the power generation efficiency vector;
Including,
Learning the power generation prediction network separately so that the thermal imaging camera unit calculates a power generation vector estimating the amount of power generated by the photovoltaic power generation of the photovoltaic cell array from the thermal image taken by the photovoltaic cell array,
The efficiency estimation network calculates a power generation efficiency vector that predicts the power generation efficiency representing the ratio of the amount of power supplied to the load to the amount of power generated by the photovoltaic power generation of the photovoltaic cell array from the same thermal image on which the power generation prediction network has been trained. learn individually,
A learning unit for integrating and learning the power generation prediction network, the efficiency estimation network, and the integrated prediction network to calculate the effective power generation by predicting the amount of power supplied to the load among the amount of power generated by the photovoltaic power generation of the photovoltaic cell array;
characterized in that it further comprises
A device for measuring solar power generation.

Images