KR20200089450A - Device and method for generating image to monitor state of power plant - Google Patents

Abstract

Embodiments relate to an image generating device for monitoring a situation of a power generation facility and to a method thereof. Provided are the image generating device, which comprises a data obtaining unit obtaining sensing data of a plurality of sensors installed in a power generation facility, and an image generating unit generating a situation image representing a situation of the power generation facility based on a situation score in a monitoring time range, and the method thereof.

Description

DEVICE AND METHOD FOR GENERATING IMAGE TO MONITOR STATE OF POWER PLANT}

The present invention relates to a technology for monitoring the status of a power generation facility, and more particularly, to an apparatus and method for generating an image representing the status of a power generation facility based on multidimensional sensing data obtained through a plurality of sensors installed in the power generation facility. will be.

The most necessary operation in driving the power generation facility is to monitor the operation status of the power generation facility and/or the abnormality of the surrounding situation. To this end, a number of and various types of sensors are used to monitor the operation of the power plant.

1 is a view showing sensing data of a power generation facility according to an exemplary embodiment.

Referring to FIG. 1, air pressure data for each air pressure sensor may be acquired through a plurality of air pressure sensors installed in a power generation facility. Based on the air pressure data value, monitoring is performed to determine whether the situation of the power generation facility is a normal situation or a failure situation in terms of air pressure. It is common to monitor the situation in terms of air pressure for each sensor. For example, it is determined whether the air pressure surrounding the first air pressure sensor is normal based on the air pressure data of the first air pressure sensor, and individually the air pressure situation around the second air pressure sensor is based on the air pressure data of the second air pressure sensor. It is judged whether it is normal.

However, as the power generation facilities and sensor technologies have been advanced, the types of sensors required for monitoring the power generation facilities have been diversified, and the number of sensors has been installed with as few as tens and tens of thousands. In this case, when the amount of sensing data for each sensor is large and the sensing data itself is provided to the user, it is almost impossible for the user to make an accurate judgment on the situation of the power generation facility.

In addition, it may take a long time to determine the situation of the power generation facility based on the sensor's sensing data for each sensor. Furthermore, there is a limit to efficiently providing information capable of comprehensively determining the situation of the entire power generation facility.

Patent Publication No. 10-2016-0055622

According to an aspect of the present invention, it is possible to provide an apparatus for generating an image representing a situation of a power generation facility based on multidimensional sensing data obtained through a plurality of sensors installed in the power generation facility.

An image generation method performed by an image generation device for monitoring according to an aspect of the present invention includes obtaining sensing data of a plurality of sensors installed in the power generation facility; Calculating a situation score by applying the sensing data to a score model; And generating a situation image representing the situation of the power generation facility based on the situation score in the monitoring time range. Here, the score model includes a statistical feature extraction layer configured to extract statistical feature values from the sensing data, and a score calculating layer configured to calculate a situation score based on the statistical feature values extracted from the sensing data.

In one embodiment, the score model extracts statistical feature values for each sample by applying a sample in a normal situation to a pre-modeled statistical feature extraction layer-the sample receives sensing data of a plurality of sensors in a normal situation. Included as sample data; Calculating an average and a covariance for sample data for each sensor based on the statistical characteristic values of the sample data; Modeling a score calculation layer for calculating a situation score based on the average and covariance of the sample data for each sensor; And generating a score model including the statistical feature extraction layer and the score calculation layer.

In one embodiment, the statistical feature value is one or more of an average value, median value, maximum value, minimum value, standard deviation, root mean squre (RMS), skewness, kurtosis, and combinations thereof. It may include.

In one embodiment, the statistical feature extraction layer may be configured to extract statistical feature values for input sensing data.

In one embodiment, the statistical feature extraction layer may be configured to extract a statistical feature value for the variation data of the input sensing data.

In one embodiment, the score calculation layer may be modeled by Equation 1 below.

Here, F _k is a statistical characteristic value set including a statistical feature values extracted from the sensed data of the plurality of sensor at time (t _k) F _k comprises a statistical feature values extracted from the sensed data of the i-th sensor, n is the number of sensors, fn is the number of types of statistical feature values, m _i is the mean for the statistical characteristic values of the i-th sensor, cov _i is the covariance for the statistical characteristics of the i-th sensor, T is a linear transformation of the matrix Shows.

In one embodiment, when the sensing data is multi-dimensional sensing data obtained from a plurality of sensors, the multi-dimensional sensing data is converted into one-dimensional time series data and scored.

In one embodiment, the generating of the context image may include: determining, by the image generator, the number of pixels forming the context image based on the monitoring time range; And determining a color based on a situation score associated with the pixel.

In an embodiment, determining the number of pixels may include determining the number of horizontal pixels and the number of vertical pixels of the context image based on the monitoring time range.

In one embodiment, the position of the pixel may be expressed as a value corresponding to a time in the monitoring time range as horizontal coordinates, and a value corresponding to a time in the monitoring time range as vertical coordinates.

In one embodiment, the monitoring time range is a time range between the first time and the second time, and the second time may be a previous time from the first time.

In one embodiment, the first time may be real time.

In one embodiment, the step of determining the color comprises: calculating a difference between situation scores in time related to the pixel as an image color value; Generating an image color table by setting a maximum value and a minimum value of the calculated image color value to a numerical range of a color set including at least some of the pre-stored colors; And determining the color of the image color table corresponding to the calculated image color value as the color of each pixel.

In one embodiment, the image color value represents a difference between a situation score at a time (t _v ) corresponding to the vertical coordinate (v) and a situation score at a time (t _h ) corresponding to the horizontal coordinate (h). , The image color value may be calculated by the following equation.

Here, the image (v, h) is an image generation method characterized in that the image color value of the pixel having the vertical coordinates (v) and horizontal coordinates (h).

In an embodiment, the method may further include determining a situation of the power generation facility based on the situation image.

In one embodiment, the step of determining the situation of the power generation facility may include determining the situation of the power generation facility by applying the situation image to the situation determination model. The situation determination model is a machine learning model based on a convolution neural network (CNN), and is machine-learned to classify a situation image in a normal situation and a situation image in an abnormal situation.

The computer-readable recording medium according to another aspect of the present invention may store program instructions readable by a computing device and operable by the computing device. Here, when the program instruction is executed by a processor of the computing device, the processor may perform an image generation method according to the above-described embodiment.

An image generating apparatus for monitoring the situation of a power generation facility according to another aspect of the present invention includes: a data acquisition unit that acquires sensing data of a plurality of sensors installed in the power generation facility; A storage unit for storing the pre-modeled score model; A score calculating unit that calculates a situation score by applying the sensing data to the score model; And an image generation unit that generates a situation image representing the situation of the power generation facility based on the situation score in the monitoring time range. Here, the score model includes a statistical feature extraction layer configured to extract statistical feature values from the sensing data, and a score calculating layer configured to calculate a situation score based on the statistical feature values extracted from the sensing data.

The image generating apparatus according to an aspect of the present invention may generate a situation image comprehensively representing a situation of a power generation facility and provide the user with a multidimensional data of a plurality of sensors installed in the power generation facility. Therefore, the user can more intuitively determine the situation of the power generation facility through the situation image, which can help to improve the user's situation judgment ability.

The image generating device may provide the user with comprehensive and intuitive information on the situation of the power generation facility through the situation image. That is, the user is provided with a non-numeric monitoring result, so that the numerical result of the sensor itself is provided and there is no need to interpret it again, thus increasing user convenience.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

BRIEF DESCRIPTION OF DRAWINGS To describe the technical solutions in the embodiments of the present invention or in the prior art more clearly, the drawings required in the description of the embodiments are briefly introduced below. It should be understood that the drawings below are for the purpose of describing the embodiments of the present specification and not for the purpose of limitation. In addition, some elements to which various modifications such as exaggeration and omission are applied may be illustrated in the drawings below for clarity.
1 is a view showing sensing data of a power generation facility according to an exemplary embodiment.
2 is a schematic block diagram of an image generating device for monitoring the situation of a power generation facility, according to an embodiment of the present invention.
3 is a diagram for schematically explaining the operation of the image generating apparatus according to an embodiment of the present invention.
4 is a flowchart illustrating a process of generating a score model according to an embodiment of the present invention.
5A to 5C are diagrams for explaining a process of generating a context image according to an embodiment of the present invention.
6 is a view for explaining the meaning of a situation image according to a situation of a power generation facility according to an embodiment of the present invention.
7 is a diagram schematically showing a structure of a situation determination model according to an embodiment of the present invention.
8 is a schematic flowchart of an image generating method showing a situation of a power generation facility according to an embodiment of the present invention.
9 is a diagram schematically showing step-by-step results of the method of FIG. 8 according to an embodiment of the present invention.

The terminology used herein is only to refer to a specific embodiment and is not intended to limit the invention. The singular forms used herein also include plural forms unless the phrases clearly indicate the opposite. As used herein, the meaning of “comprising” embodies a particular property, region, integer, step, action, element, and/or component, and the presence or presence of another property, region, integer, step, action, element, and/or component. It does not exclude addition.

Although not defined differently, all terms including technical terms and scientific terms used herein have the same meaning as those generally understood by those skilled in the art to which the present invention pertains. Commonly used dictionary-defined terms are additionally interpreted as having meanings consistent with related technical documents and currently disclosed contents, and are not interpreted in an ideal or very formal meaning unless defined.

In the present specification, the abnormal situation is very likely to cause all the situations affecting the driving result of the preset power generation facility, such as failure of the power generation facility, operation interruption, etc., and/or the above-described situation such as failure, operation interruption, etc. It includes. For example, abnormal situations may include failure situations and dangerous situations. The normal situation is a situation other than the abnormal situation, and refers to a situation in which an operation result of a preset power generation facility can be obtained. For example, an error in the rated range may be referred to as a normal situation.

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

2 is a diagram for schematically explaining the operation of the image generating apparatus according to an embodiment of the present invention.

Referring to FIG. 2, the image generating apparatus 10 converts sensing data acquired through a plurality of sensors (eg, sensing data of FIG. 1) to the situation scores of FIG. 2, and monitors time ranges M ₁ and M _A situation image I ₁ , I ₂ is generated based on the situation score in ₂ ). When the sensing data obtained through a plurality of sensors is multidimensional sensing data, the situation score is data that is multidimensionally sensed and converted into one-dimensional time series data. The image generating device 10 may provide the user with comprehensive and intuitive information on the situation of the power generation facility through the situation image. That is, the user is provided with a non-numeric monitoring result, so that the sensor's numerical result itself is provided and there is no need to interpret it again. As such, the burden on the user is reduced, and eventually the user's monitoring convenience is increased.

3 is a schematic block diagram of an apparatus for generating a situation image of a power generation facility, according to an embodiment of the present invention.

Referring to FIG. 3, an apparatus for generating a situation image representing a situation of a power generation facility (hereinafter referred to as an “image generation device”) 10 for monitoring the power generation facility includes: a data acquisition unit 100 that acquires sensing data; It includes a score calculating unit 300 for calculating the situation score of the power generation facility based on the obtained sensing data and an image generating unit 400 for generating a situation image based on the situation score in the monitoring time range. In addition, the image generating device 10 may further include a situation determination unit 500 for determining the situation of the power generation facility based on the situation image. In some embodiments, the image generating apparatus 10 may further include a model generator 200 for modeling a score model for calculating a situation score and/or a prediction model for determining the situation.

Embodiments may have aspects that are entirely hardware, entirely software, or partially hardware and partially software. For example, the image generating apparatus 10 may collectively refer to hardware equipped with data processing capability and operating software for driving the same. In this specification, the terms “unit”, “module”, “device”, or “system” are intended to refer to a combination of hardware and software driven by the hardware. For example, the hardware may be a data processing device including a central processing unit (CPU), a graphics processing unit (GPU), or other processor. Also, the software may refer to a running process, an object, an executable, a thread of execution, a program, or the like.

The data acquisition unit 100 acquires sensing data related to the situation of the power generation facility for monitoring the power generation facility.

Sensing data is obtained through a plurality of sensors installed in a power generation facility. In one embodiment, sensing data for monitoring may be real-time sensing data. Or it may be sensing data at a user set time that the user wants to monitor.

The sensor includes various sensors capable of obtaining information on the driving situation and/or the surrounding situation of the power generation facility (ie, the situation of the power generation facility). For example, sensors include temperature sensor, pressure sensor, moisture sensor, gravity sensor, geomagnetic sensor, motion sensor, gyro sensor, acceleration sensor, tilt sensor, brightness sensor, olfactory sensor, depth sensor, banding sensor, audio sensor, image sensor , And combinations thereof.

The sensing data is a result data of a plurality of sensors installed in a power generation facility detecting a situation of the power generation facility, and is data composed of a detection time and a detection result. The detection result may include various information depending on the sensor. For example, the sensing results can be heat, temperature, humidity, pressure, image, voice, speed, location information, or a combination thereof.

The sensing data includes sensing data in a normal situation and/or sensing data in an abnormal situation (eg, a dangerous situation, a failure situation, etc.). In addition, the sensing data obtained from a plurality of sensors includes sensing data sensed by each sensor. The sensing data may be multi-dimensional sensing data when there is more than one type of sensor pre-installed in the power generation facility.

The obtained sensing data may be represented as a graph over time for each sensor, as shown in FIG. 1. In addition, the sensing data may be represented by an l×n matrix, where l is time and n is the number of sensors installed in the power generation facility. The matrix of sensing data is composed of the sensing results detected by each sensor in time.

In one embodiment, the time at which the sensing data is acquired may be expressed based on the time at which the monitoring operation is started. For example, 1000 seconds in FIG. 2 may represent 1000 seconds after starting monitoring. However, the present invention is not limited thereto, and may be expressed by various criteria such as time from the start time of driving the power generation facility. In addition, the sensing data acquisition time is not limited to seconds, and may be expressed by various units such as minutes and milliseconds.

Additionally, the data acquisition unit 100 may further acquire data related to sensing data. For example, the data related to the sensing data may include power plant information (eg, an identifier assigned to the power plant).

The score calculation unit 300 calculates the situation score by applying the monitoring sensing data to the score model. The score model includes a statistical feature extraction layer configured to extract statistical feature values from monitoring sensing data and a score calculating layer configured to calculate a situation score based on the statistical feature values extracted from the monitoring sensing data.

The image generating apparatus 10 according to an embodiment of the present invention generates a context image using a pre-generated model. For example, a score model for calculating a situation score may be modeled and generated by the model generator 200.

The model generation unit 200 may be configured to configure one or more models when the image generation device 10 uses one or more models. In one embodiment, the model generator 200 may be configured to generate a score model for calculating a situation score and/or a judgment model for situation judgment. For example, the model generation unit 200 may include a score model generation unit 230 and a situation determination model generation unit 250.

The score model generation unit 230 generates a score model showing a probabilistic model for a normal situation for each sensor. The probabilistic model can be modeled based on a statistical distribution of sensing data for a normal situation for each sensor.

4 is a flowchart illustrating a process of generating a score model according to an embodiment of the present invention.

Referring to FIG. 4, the score model generation unit 230 extracts a statistical characteristic value of the sensing data of the sample using a normal situation sample including sensing data in the normal situation (S231); In the extracted statistical feature value, calculating the mean and covariance for the statistical feature value (S233); Modeling a score calculation layer for calculating a situation score based on the mean and covariance of the statistical feature value (S235); And generating a score model including a score calculation layer (S237), thereby generating a score model.

The score model generation unit 230 uses sensing data in a normal situation as a sample for generating a score model. The sample includes sensing data in a normal situation, and when there are multiple sensors installed in the power generation facility, the sensing data of the sample includes sensing data for each sensor.

The acquisition of a sample including sensing data in a normal situation may be obtained by input by a user, reception from a sensor pre-installed in a power generation facility, reception through wired/wireless communication, and the like.

When the sensing data of a sample is obtained by a plurality of sensors installed in a power generation facility, it can be expressed in a matrix like the sensing data to be used for monitoring. For example, the sample may be expressed as Equation 1 below.

[Equation 1]

Here, S _i is the i-th sample, and N is the number of samples. Each sample, like the sensing data described above, may be represented by an l×n matrix, where l is time and n is the number of sensors installed in the power generation facility. The matrix of samples represents sensing data according to the time of each sensor.

For example, in the sample S ₁ containing the sensing data for 1 to 300 seconds, the sample S ₁ is the sensing data of the first sensor (eg, pressure sensor) at 1 to 300 seconds, 1 to 300 It includes the sensing data of the second sensor (eg, temperature) in seconds, and the sensing data of the nth sensor (eg, moisture) in 1,300 seconds.

In step S231, statistical characteristics are extracted for each sample from sensing data included in the plurality of samples. Statistical feature values are extracted using a pre-modeled statistical feature extraction layer. The process of extracting a statistical feature value from a sample including sensing data in a normal situation can be schematically expressed through Equation (2).

[Equation 2]

Here, extFunc represents a statistical feature extraction layer, and F _i represents a statistical feature value extracted from the i-th sample (S _i ). The extracted feature F _i may be represented by an fn×n matrix, fn denotes the number of types of statistical feature values extracted, and n denotes the number of sensors. The matrix of statistical feature values consists of statistical feature values for sensing data calculated for each sensor.

Statistical feature values may include various statistical values that may indicate characteristics of the data distribution. For example, statistical characteristic values include mean, median, maximum, minimum, standard deviation, root mean squre (RMS), skewness, and kurtosis values. It may include. However, the present invention is not limited thereto, and other values capable of representing characteristics of distribution of sensing data may be extracted as statistical characteristic values. The process of calculating the above-described statistical feature values is known to those skilled in the art, and thus detailed descriptions thereof will be omitted.

In one embodiment, the statistical characteristic value extracted in step S231 may be a statistical characteristic value for the sensing data of the sample. For example, it may be an average value of temperature sensing data included in the sample.

Additionally, the statistical characteristic values extracted in step S231 may include statistical characteristic values for the two-dimensional data converted based on the sensing data of the sample. For example, the secondary data may be change data obtained by converting raw sensing data based on a change amount over time. In this case, the statistical characteristic values extracted in step S231 may include statistical characteristic values for sensing data and statistical characteristic values for variation data. Here, the statistical characteristic values (eg, kurtosis) related to the variation amount are excluded from the statistical characteristic values for the variation amount data.

As a result of performing step S231, one or more statistical feature values may be extracted from the statistical feature extraction layer extFunc, and may be expressed as a set of statistical feature values for each sample as follows.

[Equation 3]

, Where N is a natural number

In one example, when the average value, the maximum value, the minimum value, and the RMS value are extracted as the statistical feature values, the set of statistical feature values F ₁ extracted from the sample S ₁ is the first included in the sample S1. 1 for the second sensor (e.g., temperature) included in the sample S1, including the average value, the maximum value, the minimum value, and the RMS of the sensing data for 1 to 300 seconds of the sensor (e.g., pressure sensor) Contains the average value, maximum value, minimum value, and RMS of sensing data for ~300 seconds, ..., the average value, maximum value of sensing data for 1 to 300 seconds of the nth sensor (e.g., moisture) , Minimum value, and RMS. In this case, the statistical feature value layer is pre-modeled to calculate the average value, maximum value, minimum value, and RMS value of the sensing data.

That is, the statistical feature extraction layer indicated by extFunc in Equation 2 is only a symbol for indicating that each statistical feature value can be extracted, and one mathematical expression (eg, a function) composed of a specific variable and/or constant ).

In step S233, the score model generation unit 230 calculates the average and covariance of the statistical feature values for each sensor based on the statistical feature values extracted in step S231.

For example, when the set of statistical feature values is N (F ₁ to F _N ) as in Equation 3, the feature values (eg, average values) of the first sensor are also N. Based on the set of statistical feature values for each sensor, the average and covariance of the statistical feature values for the normal situation for each sensor is calculated (S233). This is because, under normal circumstances, the distribution tendency of statistical feature values extracted from a plurality of samples is the same or similar to the Gaussian distribution.

In step S235, the score model generator 230 models a score calculation layer that calculates a situation score based on the average and covariance for each sensor calculated in step S233. The situation score is a score normalized as an average for each sensor, and the score calculation layer may be expressed by the following equation (4).

[Equation 4]

Here, F _k is a set of statistical feature values extracted from sensing data acquired by a plurality of sensors (eg, the first sensor to the n-th sensor) at time t _k , and F _k is extracted from the sensing data of the i-th sensor Includes statistical feature values. m _i and cov _i represent the mean and covariance of statistical feature values for the i-th sensor calculated in step S233. T represents the linear transformation of the matrix.

The score model generated through the above-described process converts various types of sensing data pre-installed in the power generation facility into one-dimensional time series data. That is, multi-dimensional sensing data such as temperature data, wind data, speed data, and the like are converted into the same category called a situation score. In addition, the size included in the detection result included in the sensing data is reflected in the situation score, so that the correlation with the actual detection result is high. In one embodiment, the model generator 200 may be a component included in the image generating device 10, as shown in FIG. 2. However, the present invention is not limited to this. In another embodiment, the model generator 200 may be a component remotely located in the image generating device 10. In this case, the image generation device 10 may receive and store the score model previously generated by the remotely located model generation unit 200 before monitoring and use it to calculate the situation score. In this case, the image generating device 10 may further include a storage device (not shown) for storing the score model. The storage device may include at least one of a volatile memory (not shown), a non-volatile memory (not shown), and an external memory (not shown). For example, the volatile memory may be any of dynamic random access memory such as read-only memory (RAM) or synchronous dynamic read-only memory (synchronous DRAM, SDRAM), double-speed (DDR) SDRAM, or Rambus DRAM (RDRAM). The non-volatile memory may be any one of a read-only memory (ROM) and a flash memory, and the external memory may be any of the MMC type, SD type, and CF type memory cards.

In one embodiment, the image generating device 10 is non-amplified so that a model previously learned in a non-volatile memory in which stored data is not lost even after a power supply is started again is restored even when a reset is performed due to a power supply interruption or the like. The pre-trained model can be stored in the sex memory.

The image generator 400 may generate a situation image indicating the situation of the power generation facility during the specific time based on the sensing data in the monitoring time range.

In one embodiment, the image generator 400 may determine the number of pixels based on the unit time of the monitoring time range. For example, the unit time in the monitoring time range is in seconds, and the seconds and pixels may correspond. In some embodiments, the number of horizontal pixels and number of vertical pixels in the context image is determined based on a monitoring time range.

5A to 5D are diagrams for explaining a process of generating a context image according to an embodiment of the present invention. In the embodiment of FIG. 5, the monitoring time range is 7 seconds, and a situation image is generated using the situation score for 7 seconds.

However, 7 seconds as the monitoring time range is merely exemplary, and may be variously set based on the type of power generation facility, driving characteristics, and the like. For example, in the case of a power generation facility that requires very precise monitoring that requires immediate response, the time range for image generation may be set to a range of less than 7 seconds. On the other hand, when simply monitoring the current status of power generation facilities for a long time, it may be set to a longer time range (for example, 100 seconds).

As shown in Figure 5a when the situation scores from 1 to 7 seconds are calculated. The image generator 400 is configured to determine the number of pixels forming the context image based on the monitoring time range 7 seconds.

For example, in the case of the monitoring time range of FIG. 5A, since the monitoring time range is 7 seconds, the number of horizontal pixels in the situation image is determined to be 7 and the number of vertical pixels is 7. That is, when receiving the situation score for 7 seconds of FIG. 5A, the image generator 400 may generate a situation image having 7×7 pixels in size.

The image generator 400 determines the number of pixels, and determines the color of each pixel to give a corresponding color to the pixel.

In one embodiment, the image generator 400 associates the position of each pixel with time, and calculates the difference between the situation scores at the time associated with the pixel as an image color value. Then, the color of each pixel is assigned based on the image color value.

The image generator 400 may associate the horizontal and vertical coordinates of the pixel with the monitoring time. For example, when the number of 7×7 pixels is determined by the monitoring time of 7 seconds, horizontal coordinates of 1 to 7 and vertical coordinates of 1 to 7 may be set. As such, time associated with a pixel includes time associated with the location of the pixel.

As a result, the image generator 400 may determine the number of pixels based on the unit time of the monitoring time range, and associate the monitoring time with each pixel, thereby expressing the position of the pixel as the monitoring time.

Then, the image generator 400 calculates the image color value for each pixel by Equation 5 below.

[Equation 5]

Here, image(v, h) is an image color value of a pixel having vertical coordinates (v) and horizontal coordinates (h), and the image color value is a situation at a time (t _v ) corresponding to the vertical coordinate (v). It represents the difference between the score and the situation score at the time (t _h ) corresponding to the horizontal coordinate (h).

The result of calculating the image color value for each pixel using the situation score of FIG. 5A is shown as shown in FIG. 5B. 5B, in the situation image Ix, the horizontal coordinate corresponding to 1 second is 1, ??. The horizontal coordinate corresponding to 7 seconds is represented by 7. In this case, since the vertical coordinate (v) of the pixel P1 is 3 and the horizontal coordinate (h) is 1, the image color value of the pixel P1 is calculated as follows: image(3, 1) = abs(score (3)-score(1)) = abs(3-2.5) = 0.5. Also, since the vertical coordinate (v) of the pixel P2 is 2 and the horizontal coordinate (h) is 4, the image color value of the pixel P2 is calculated as follows: image(2, 4) = abs(score( 2)-score(4)) = abs(4.5-3) = 1.5.

Through the above process, the image generating unit 400 calculates the image color value of the pixel, and then generates a context image including the color by applying the color corresponding to the image color value to the pixel.

In one embodiment, the image generator 400 is configured to generate an image color table quantifying colors included in a pre-stored color set. Then, a color corresponding to the calculated image color value is determined using the image color table.

The color set includes a plurality of colors. In one example, the color set may include at least some colors of an RGB color model (ie, RGB color series), as shown in FIG. 5C. However, it is not limited thereto and may include at least some colors of various color models (eg, HSV, YCbCr, CMY, etc.). In another example, the color set may include all of the colors of the gray color model composed of white and black, and the like.

In some embodiments, the colors of the color set may be represented as continuous colors. Here, continuity is expressed as a criterion for classifying color models. For example, when the color set includes at least some colors of the gray color model, whether the colors are continuous or not is expressed as brightness.

The image generator 400 quantizes the colors included in the color set based on the calculated image color values. The digitization represents an operation of setting one color to one numerical range.

In one embodiment, the image generator 400 quantizes the colors included in the color set using the maximum and minimum values of the calculated image color values. Referring back to FIG. 5B, the minimum value of the calculated image color value is 0 and the maximum value of the calculated image color value is 2.5. The image generating unit 400 may set a numerical range of the color set to 0 and 2.5, and numerically color the numerical value within the range of 0 and 2.5.

Digitization of color can be performed in various ways. In one embodiment, the image generator 400 colors the colors of the color set according to the classification criteria (eg, intensity, hue, saturation, color difference information (Cr, Cb), etc.). You can arrange and colorize the arranged colors based on the ratio of the numerical range.

For example, if the color set contains the color of the gray color model, the color with 100% brightness (i.e. white) is the maximum, and the color with 0% brightness (i.e. black) is the minimum. The color whose value and brightness correspond to 50% can be quantified as the middle value between the maximum value and the minimum value.

However, the present invention is not limited thereto, and may be variously set based on whether the colors included in the color set are continuous, discontinuous, or the number of colors. For example, if a set of colors is continuous, digitization can be performed in a one-to-many manner that can represent multiple similar colors with a single value, or if a set of colors is discontinuous, multiple values are set for a single color. Numericalization may be performed in a one-to-one manner.

Thereafter, the image generator 400 assigns a color corresponding to the image color value calculated for each pixel to the corresponding pixel by using the digitized image color table. Referring to FIGS. 5B and 5C, a dark blue color corresponding to 0 is assigned to a pixel having 0 as an image color value, and a blue color corresponding to 0.5 is assigned to a pixel having 0.5 as an image color value, Pixels having an image color value of 1 are assigned a light blue color corresponding to 1, and pixels having an image color value of 1.5 correspond to dark green, a color corresponding to 1.5, and pixels having an image color value of 2. A color corresponding to 2 is dark yellow, and an image pixel having 2.5 is yellow.

As a result, the image generator 400 may generate the situation image of FIG. 5D from the situation score for 7 seconds in FIG. 5A.

As such, the color is determined based on the image pixel values. When the values of the vertical coordinate (v) and the horizontal coordinate (h) of the pixel are the same, since the situation score at the same time is used to calculate the image color value, as shown in FIG. 5C, the image color of the pixel. The value is 0. Therefore, the situation image is constructed symmetrically with respect to the diagonal.

6 is a view for explaining the meaning of a situation image according to a situation of a power generation facility according to an embodiment of the present invention.

The situation image generated by the image generator 400 indicates whether the situation of the power generation facility in the monitoring time range is normal or abnormal.

It is based on the complexity of the image, that is, the image entropy, that the situation image indicates whether the situation of the power generation facility is normal or likely to be normal.

6 illustrates a set of colors used by the image generator 400 to generate images. The color of the color set on the left side of FIG. 6 is still in the pre-digitization stage. The image generator 400 quantizes the color of the color set based on the color value for each pixel. In one embodiment, the image generator 400 may set the numerical range to quantify the color of the color set.

According to Equation (4), the situation score tends to increase as the sample has a data distribution different from the data distribution in the normal situation.

Therefore, when an abnormal situation for the power generation facility does not occur during the monitoring time, the situation score at the monitoring time is calculated to have a value within a narrow range with little variation between each other. For this reason, the maximum value of the image color value for each pixel representing the difference in the situation score over time is calculated as a low value close to 0 (for example, 10 or less), and the color of the color set is quantified as a narrow range of values.

On the other hand, when an abnormal situation for the power generation facility occurs during the monitoring time, the situation score in the abnormal situation has a larger value than the situation score in the normal situation. This is because the data distribution in the abnormal situation is different from the data distribution in the sample in the normal situation. In general, the more the abnormal situation, the greater the difference in data distribution, so the situation score in the abnormal situation has a very large value compared to the situation score in the normal situation. For this reason, the image color value for each pixel according to the time calculated by Equation 5 is much larger than the maximum value in the absence of an abnormal situation. As a result, when an abnormal situation occurs, the colors of the color set are numerically measured with a relatively wide range of values.

That is, even if the image generating unit 400 uses the same color set, the colors of the color set are numerically divided into different numerical ranges depending on whether an abnormal situation occurs during the monitoring time.

In one example, when a normal situation is maintained for a monitoring time for a specific power generation facility, as illustrated in FIG. 5, maximum and minimum values of image color values for each pixel may be calculated as 0 and 2.5, respectively. Therefore, the color of the color set in FIG. 6 is numerically valued between 0 and 2.5, and the image color table shown at the bottom of FIG. 6 is used for image generation. When the image color table at the bottom of FIG. 6 is used, a situation image I ₃ having a complex color distribution at the bottom of FIG. 6 is generated.

On the other hand, in the same example, when an abnormal situation occurs during the monitoring time, the maximum value of the image color value for each pixel may be 1000 due to a sudden change in sensing data in the abnormal situation. Accordingly, the same color set of FIG. 6 is numerically valued from 0 to 1000, and the image color table shown at the top of FIG. 6 is used for image generation.

In this case, when the time associated with the pixel is a normal situation, the image color value for the pixel may be calculated as a value of approximately 0 to 2.5. Since the colors corresponding to values of 0 to 2.5 in the image color table at the top of FIG. 6 are the same or similar colors, the colors of pixels related to the normal situation are the same or similar. In this way, when an abnormal situation occurs, the situation image I ₄ is generated to have a simple color distribution tendency as illustrated in the upper part of FIG. 6.

Due to the differences in the numerical range, the situation image is configured to have a complex color distribution when an abnormal situation does not occur during the monitoring time range. And, when an abnormal situation occurs during the monitoring time range, the situation image is configured to have a simple color distribution.

The complexity of the color distribution of the image can be expressed by entropy through Equation 6 below.

[Equation 6]

Here, pi represents the probability that the i-th image color value appears in the image color value distribution. For example, in FIG. 5B, the image color value distribution is between 0 and 2.5. In this case, the probability that the image pixel value is 0 is 9/49.

When calculating the image entropy for the context image of FIG. 6, the image entropy of the context image I ₃ has 6.32, and the image entropy of the context image I ₄ has 2.75.

As described above, the context image I ₃ is a relatively complex image when compared to the context image I ₄ , and has a relatively high entropy. After all, the lower the entropy, the more complex the color distribution of the image is, which provides information that the situation of the power plant represents a normal situation.

The image generating device 10 provides this situation image to the user, so that the user can monitor the situation of the power generation facility.

Additionally, the image generating device 10 may continuously provide the user with a situational image between predetermined times in real time.

Additionally, the image generating device 10 may be further configured to determine the situation of the power generation facility indicated by the situation image. Furthermore, the image generating device 10 may provide a user with a determination result based on the context image.

Referring back to FIG. 2, in one embodiment, the image generating apparatus 10 may further include a situation determination unit 500.

In one embodiment, the situation determination unit 500 may calculate the entropy of the situation image, and determine that the situation is normal if the entropy of the situation image is less than or equal to a predetermined threshold. The predetermined threshold may be determined based on the performance of the power generation facility, the type of sensor, the performance of the sensor, and the like.

In another embodiment, the situation determination unit 500 determines the situation of the power generation facility by applying the situation image to the situation determination model. The situation determination model is a machine learning model to classify a situation image in a normal situation and a situation image in an abnormal situation.

7 is a diagram schematically showing a structure of a situation determination model according to an embodiment of the present invention.

In one embodiment, the situation determination model may be a machine learning model based on a convolution neural network (CNN). The situation determination model is a machine trained model that classifies an input image into at least two groups (eg, a normal situation group and an abnormal situation group, or a normal situation group, a crisis situation group, and a failure situation group) using a convolution filter.

In one embodiment, the situation determination model is an input layer (L ₁ ) to which the situation image is input, a feature extraction layer (L ₂ ) to extract features for determining whether the situation is normal, and whether the input situation image is a normal situation, It includes a classification layer (L ₃ ) for classifying whether the situation is abnormal.

The input layer L ₁ may be further configured to perform augmentation processing. For example, in order to derive a stronger learning result for the situation image, random flip, random rotation, random brightness adjustment, random contrast adjustment, etc. Can be performed.

The feature extraction layer L ₂ may include one or more convolutional layers. Further, the feature extraction layer L ₂ may further include a pooling layer.

The classification layer L ₃ may include a full-connected layer responsible for final determination. In addition, the classification layer L ₃ may further include a probability layer configured to calculate a probability that the input situation image belongs to the normal situation group and/or the abnormal situation group based on the output value output from the fully connected air. . The probability layer may be configured to calculate the probability by, for example, a softmax algorithm.

In one embodiment, the situation determination model may be generated by the determination model generation unit 250 of the model generation unit 200 illustrated in FIG. 2. The judgment model generation unit 250 is a situation image (eg, I ₁ , I ₃ ) representing a normal situation generated in a normal situation and/or an image (eg, I ₂ , I ₄ ) representing an abnormal situation generated in an abnormal situation. Use as a machine learning sample to learn the parameters of the situation judgment model. Specifically, the parameters are learned by reducing the error by comparing the classification result (ie, the model application result) of the input situation image with the actual result of the sample. To this end, the machine learning sample may include contextual images and labeling data representing actual results.

The situation determination unit 500 may determine whether the situation image generated by the image generation unit 400 represents a normal situation using a previously learned situation determination model. The pre-trained situation determination model may be stored in a storage device (not shown) of the image generating device 10.

It will be apparent to one skilled in the art that the image generating device 10 may include other components not described herein. For example, the image generating device 10 includes other hardware elements necessary for the operations described herein, including a network interface, an input device for data entry, and an output device for display, printing or other data display. It may include.

8 is a schematic flowchart of an image generating method showing a situation of a power generation facility according to an embodiment of the present invention, and FIG. 9 is a schematic step-by-step result of the method of FIG. 8 according to an embodiment of the present invention It is a drawing shown.

8 and 9, the image generation method performed by the image generation device 10 acquires sensing data for monitoring through a plurality of sensors installed in a power generation facility (S810). Thereafter, the situation data is calculated by applying the sensing data to the score model (S830).

For example, as illustrated in FIG. 9, sensing data between 1000 and 1600 seconds is obtained by six air pressure sensors (S810 ). The score generator 300 calculates the situation score by applying the sensing data to the score model (S830). When the sensing data is applied to the score model, a plurality of statistical feature values are extracted for each of the six air pressure sensors through the feature extraction layer of the score model. That is, the statistical feature values extracted from the sensing data on the left side of FIG. 9 include statistical feature values for the first to sixth air pressure sensors, and multidimensional sensing data obtained from the first to sixth air pressure sensors through the score model It is converted to situation score, which is one-dimensional time series data. That is, the situation of the power generation facility is comprehensively considered and quantified by the situation score.

Then, the image generating device 10 generates a situation image based on the situation score in the monitoring time range (S840). When the monitoring time range is 100 seconds, the image generating device 10 generates a situation image I ₁ having 100×100 pixels using a situation score between 1000 seconds and 1100 seconds. In addition, the image generating apparatus 10 generates a situation image I ₂ having 100×100 pixels by using a situation score between 1200 seconds and 1300 seconds.

Since the complexity of the situation image I ₂ of FIG. 9, that is, the image entropy is relatively low, the situation image I ₂ indicates that the situation of the power generation facility is abnormal for 1200 seconds to 1300 seconds. The user can intuitively and conveniently monitor the situation of the power generation facility through the situation image I ₂ , and furthermore, determine that the situation is abnormal.

Additionally, when the image generating apparatus 10 includes the situation determination unit 500, the situation in the monitoring time range may be determined by applying the situation image generated in step S840 to the situation determination model.

According to the embodiments described above The operation by the image generating apparatus and method for monitoring the situation of the power generation facility may be implemented at least partially as a computer program, and recorded in a computer-readable recording medium. For example, implemented with a program product consisting of a computer-readable medium comprising program code, which may be executed by a processor for performing any or all of the steps, operations, or procedures described.

The computer may be a desktop computer, a laptop computer, a laptop, a smart phone, or a computing device such as the like, or any device that may be integrated. A computer is a device that has one or more alternative and special purpose processors, memory, storage space, and networking components (either wireless or wired). The computer may, for example, run an operating system such as Microsoft's Windows compatible operating system, Apple OS X or iOS, Linux distribution, or Google's Android OS.

The computer-readable recording medium includes all types of record identification devices that store data that can be read by a computer. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tapes, floppy disks, and optical data storage identification devices. In addition, the computer readable recording medium may be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present embodiment will be readily understood by those skilled in the art to which this embodiment belongs.

The present invention described above has been described with reference to the embodiments shown in the drawings, but this is only exemplary, and those skilled in the art will understand that various modifications and modifications of the embodiments are possible therefrom. However, it should be considered that such modifications are within the technical protection scope of the present invention. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

An image generating apparatus according to an aspect of the present invention merely transmits a sensor detection result to a user numerically or transmits whether a detection result exceeds a specific criterion (eg, a preset threshold). Unlike conventional monitoring technology, it is possible to maximize user convenience by minimizing the user's judgment burden through a comprehensive and intuitive situation image. Furthermore, it is possible to determine whether the situation of the power generation facility is abnormal using machine learning, which is one of the 4th industrial technologies.

Claims (17)

In the image generation method performed by a computing device including a processor,
Obtaining sensing data of a plurality of sensors installed in the power generation facility;
Calculating a situation score by applying the sensing data to a score model; And
Generating a situation image representing the situation of the power generation facility based on the situation score in the monitoring time range,
The score model includes a statistical feature extraction layer configured to extract statistical feature values from the sensing data, and a score calculating layer configured to calculate a situation score based on the statistical feature values extracted from the sensing data. Way.
According to claim 1, The score model,
Applying a sample in a normal situation to a pre-modeled statistical feature extraction layer to extract statistical feature values for each sample, wherein the sample includes sensing data of a plurality of sensors in a normal situation as sample data;
Calculating an average and a covariance for sample data for each sensor based on the statistical characteristic values of the sample data;
Modeling a score calculation layer for calculating a situation score based on the average and covariance of the sample data for each sensor; And
And generating a score model including the statistical feature extraction layer and the score calculation layer.
According to claim 1, The statistical feature value,
An image generation method comprising one or more of mean, median, maximum, minimum, standard deviation, root mean squre (RMS), skewness, kurtosis, and combinations thereof.
According to claim 1,
The statistical feature extraction layer is configured to extract statistical feature values for input sensing data.
According to claim 1,
The statistical feature extraction layer is configured to extract a statistical feature value for the variation data of the input sensing data.
According to claim 2, The score calculation layer is modeled by the following equation (1),
[Equation 1]

Here, F _k is a statistical characteristic value set including a statistical feature values extracted from the sensed data of the plurality of sensor at time (t _k) F _k comprises a statistical feature values extracted from the sensed data of the i-th sensor, n is the number of sensors, fn is the number of types of statistical feature values, m _i is the mean for the statistical characteristic values of the i-th sensor, cov _i is the covariance for the statistical characteristics of the i-th sensor, T is a linear transformation of the matrix Method for generating an image, characterized in that it represents.
The method of claim 1, wherein the generating of the context image comprises:
Determining, by the image generator, the number of pixels forming a context image based on the monitoring time range; And
And determining a color based on the context score associated with the pixel.
The method of claim 7, wherein the step of determining the number of pixels,
And determining the number of horizontal pixels and the number of vertical pixels of the context image based on the monitoring time range.
The method of claim 7,
The position of the pixel can be expressed as a value corresponding to a time in the monitoring time range as horizontal coordinates, and a value corresponding to a time in the monitoring time range as vertical coordinates.
The method of claim 7,
The monitoring time range is a time range between the first time and the second time,
The second time is an image generation method, characterized in that the previous time from the first time.
The method of claim 10,
The first time is an image generation method, characterized in that in real time.
The method of claim 7, wherein the step of determining the color,
Calculating a difference between situation scores at the time associated with the pixel as an image color value;
Generating an image color table by setting a maximum value and a minimum value of the calculated image color value to a numerical range of a color set including at least some of the pre-stored colors; And
And determining the color of the image color table corresponding to the calculated image color value as the color of each pixel.
The method of claim 12,
The image color value represents the difference between the situation score at the time t _v corresponding to the vertical coordinate v and the situation score at the time t _h corresponding to the horizontal coordinate h, and the image color value Is calculated by the following equation,
[Equation 2]

Here, the image (v, h) is an image generation method characterized in that the image color value of the pixel having the vertical coordinates (v) and horizontal coordinates (h).
According to claim 1,
And determining a situation of the power generation facility based on the situation image.
The method of claim 14, wherein the step of determining the situation of the power generation facility,
Including the step of determining the situation of the power generation equipment by applying the situation image to the situation determination model,
The situation determination model is a CNN (Convolution Neural Network)-based machine learning model, the image generation method characterized in that the machine learning to classify the situation image in the normal situation and the situation image in the abnormal situation.
A computer-readable recording medium that stores program instructions readable by a computing device and operable by the computing device, wherein the processor is executed when the program instructions are executed by a processor of the computing device. A computer-readable recording medium for performing the image generation method according to any one of the preceding claims.
In the image generating device for monitoring the situation of the power generation equipment,
A data acquisition unit that acquires sensing data of a plurality of sensors installed in the power generation facility;
A storage unit for storing the pre-modeled score model;
A score calculating unit that calculates a situation score by applying the sensing data to the score model; And
Including the image generating unit for generating a situation image representing the situation of the power generation facility based on the situation score in the monitoring time range,
The score model includes a statistical feature extraction layer configured to extract statistical feature values from the sensing data, and a score calculating layer configured to calculate a situation score based on the statistical feature values extracted from the sensing data. Device.

Images