KR102558609B1 - Method for evaluating wind speed patterns to ensure structural integrity of buildings, and computing apparatus for performing the method - Google Patents

Abstract

The present invention discloses a wind speed pattern evaluation method and computing device for a building.
More specifically, the wind speed pattern evaluation method guarantees the structural integrity of buildings and the safety of pedestrians by utilizing a generative adversarial imputation network (GAIN) to estimate unknown wind speed values in the process of collecting wind speed data due to environmental winds occurring instantaneously between buildings.

Description

Wind velocity pattern evaluation method for ensuring structural integrity of a building and computing device performing the method

The present invention relates to a wind speed pattern evaluation method and a computing device, and more particularly, to a method and device for evaluating a wind speed pattern of wind instantaneously generated between buildings to ensure safe walking of a user and soundness of a building.

In general, densely populated areas have their own velocities, resulting in very complex and varied wind flow patterns. This wind flow not only affects people passing on the street, but also affects the air flow between buildings. In other words, the effect of convection coefficients, heat transfer between buildings and the environment, rainfall and weather changes are all affected by air currents, and changes in wind speed cause many incidents as they contribute to irritating and harmful conditions for pedestrians and residents.

Accordingly, in major cities, wind flow patterns are necessarily analyzed when urban design is started in order to prevent negative effects caused by wind speed. The wind flow pattern is evaluated using the cross-sectional shape based on factors such as the height and design of the building and the characteristics of the building cluster. Aerodynamic building design adjustments, such as cross-sectional geometry, are suitable design strategies that can significantly reduce the effects of high wind flow. Accordingly, the cross section of the building becomes thinner and has a round shape, and the wind stress applied to the structure is lowered compared to a square cross-sectional structure. Additionally, the design of lift-up buildings primarily improves air circulation, provides shelter and facilitates recreational activities.

Channeling effects and suppression of horseshoe vortices should be investigated when two buildings are built adjacently, and where there is a significant space between the buildings, the overall aerodynamics of a group of buildings and the aerodynamics of a single building appear similar, and full-scale assessment of these airflow patterns becomes increasingly difficult.

Conventionally, computational fluid dynamics (CFD) methods have been used to evaluate the aerodynamics of airflow patterns. Computational fluid dynamics (CFD) plays a more important role than tunnel testing in the entire flow field, although it has relatively low precision. However, while computational fluid dynamics presents the most important observations of frequency response, it limits the scope of wind velocity evaluation. In addition, computational fluid dynamics is difficult to directly evaluate the wind speed as there are limitations in the detectable range.

Recently, PIV, which provides higher precision over a wider area than electrohydrodynamics, has been provided. PIV can capture wind flow types (flow field and image data) in real time by directly calculating the speed of the tracking particle. After that, PIV divides each photo data into small sections and estimates the actual wind speed using the distance vector according to the time interval of lighting. However, although PIV has many advantages, it cannot accurately measure the wind flow at a specific location due to the laser light shielding effect. Accordingly, PIV has been used to predict missing values due to laser beam shielding, and in previous studies, machine learning (ML) and deep learning (DL) are widely used to analyze wind pressure patterns and effects around structures.

However, in predicting missing values, machine learning causes overfitting problems between data due to the increased number of parameters, and deep learning has difficulties in measuring the wind effect of buildings as wind flow patterns cannot be measured.

The present invention provides a wind speed pattern evaluation method that comprehensively determines the flow pattern along which guarantees the structural integrity of a building and the safety of pedestrians on all floors by evaluating the wind speed pattern of wind generated instantaneously between a plurality of buildings.

The present invention provides a wind speed pattern evaluation method for predicting a wind speed value substituted with zero or NaN in wind speed data through a generative adversarial network model of a machine learning approach that predicts unobserved data based on wind speed data.

The present invention provides a wind speed pattern evaluation method that evaluates the wind speed pattern according to the visualized heat map by visualizing wind speed values of predicted wind speed data as a heat map.

A wind speed pattern evaluation method according to an embodiment of the present invention includes measuring wind speed data by environmental wind generated instantaneously between buildings using a particle image velocimetry (PIV); setting a projection area blocking a light source of the particle image speedometer in a target area including buildings where the wind speed data is measured; replacing wind speed values of the wind speed data belonging to the projection area with zero (0) or NaN (Not a Number); predicting a wind speed value replaced with zero or NaN by analyzing a wind speed value other than zero or NaN of the wind speed data through imputation learning; and visualizing the predicted wind speed value as a heat map and evaluating a wind speed pattern according to the visualized heat map.

In the step of measuring wind speed data according to an embodiment of the present invention, the light source of the particle image speedometer may emit light at a specific time period to measure wind speed data representing the speed of air moving horizontally between the buildings.

In the setting step according to an embodiment of the present invention, among the wind speed values of the wind speed data measured in the target area, a projection area having a missing value as the light source of the particle image speedometer is blocked may be set.

In the replacing step according to an embodiment of the present invention, the missing value of the wind speed data may be replaced with zero or NaN in consideration of a connection characteristic between wind speed values of the wind speed data.

In the predicting step according to an embodiment of the present invention, the wind speed value substituted with zero or NaN may be predicted by applying wind speed data to the generative network model and the discriminatory network model trained in an adversarial manner for the imputation learning.

The generative network model according to an embodiment of the present invention distinguishes between the wind speed value of the wind speed data and a spoofed value to perform a function of improving a prediction ratio of the discriminant network model, and the discriminant network model can generate a spoofed value to be used in the generative network model and perform a function of reducing error loss of the generative network model.

The predicting step according to an embodiment of the present invention may include receiving a wind speed value other than zero or NaN and a spoofed value of the wind speed data through the generated network model; discriminating a real component and a false component according to a vector format of wind speed data from the inputted wind speed value other than zero or NaN and the spoofed value through the discrimination network model; and estimating the wind speed value replaced with zero or NaN in consideration of the distinguished real component and false component.

In the evaluating step according to an embodiment of the present invention, a wind speed pattern of wind generated instantaneously between the buildings may be evaluated from an average wind speed of the predicted wind speed values according to the visualized heat map.

In a computing device that performs a wind speed pattern evaluation method according to an embodiment of the present invention, the computing device includes a processor, wherein the processor measures wind speed data due to environmental wind generated instantaneously between buildings using a particle image velocimetry, sets a projected area in which a light source of the particle image velocimetry is blocked in a target area including the buildings where the wind speed data is measured, replaces a wind speed value of the wind speed data belonging to the projected area with zero or NaN, and replaces the wind speed value of the wind speed data belonging to the projected area with zero or NaN. Wind speed values that are not zero or NaN of the wind speed data are analyzed to predict wind speed values substituted with the zero or NaN values, and the predicted wind speed values are visualized as a heat map to evaluate a wind speed pattern according to the visualized heat map.

The processor according to an embodiment of the present invention may measure wind speed data representing the speed of horizontal air moving between the buildings by emitting light from the light source of the particle image velocity meter at a specific time period.

The processor according to an embodiment of the present invention may set a projection area having a missing value among the wind speed values of the wind speed data measured in the target area as the light source of the particle image speedometer is blocked.

A processor according to an embodiment of the present invention replaces a missing value of the wind speed data with zero or NaN in consideration of a connection characteristic between wind speed values of the wind speed data.

The processor according to an embodiment of the present invention may predict the wind speed value substituted with the zero or NaN for applying the wind speed data to the generative network model and the discriminatory network model trained in an adversarial manner for the imputation learning.

The processor according to the embodiment of the present invention is to distinguish between the actual components and false components according to the vector format of the wind speed data for zero, or the value of the input through the discriminating network model, or the values that are not NAN, or the value of the wind speed and the spoiled value of the wind speed data through the generating network model. In consideration of the distinctive components and false components that are distinct, the zero or wind speed value confronted with the nan may be predicted.

According to one embodiment of the present invention, the wind speed pattern evaluation method evaluates the wind speed pattern of wind generated instantaneously between a plurality of buildings, thereby ensuring the structural integrity of the building and the safety of pedestrians on all floors. It can comprehensively determine the flow pattern.

According to an embodiment of the present invention, the method for evaluating wind speed patterns can predict wind speed values substituted with zero or NaN in wind speed data through a generative adversarial network model of a machine learning approach that predicts missing unobserved values based on wind speed data.

According to an embodiment of the present invention, the wind speed pattern evaluation method may evaluate the wind speed pattern according to the visualized heat map by visualizing wind speed values of the predicted wind speed data as a heat map.

1 is a diagram illustrating an overall system for ensuring the soundness of a building according to an embodiment of the present invention.
2 is a flowchart illustrating a wind speed pattern evaluation method according to an embodiment of the present invention.
3 is a graph for setting a projection area including an unknown wind speed value in a target area of a building according to an embodiment of the present invention.
4 is a diagram for explaining a GAIN model with an ML algorithm according to an embodiment of the present invention.
5 is a graph for confirming performance and square error according to a GAIN model according to an embodiment of the present invention.
6 is a graph showing the average wind speed through the actual average wind speed and the GAIN model by wind generated instantaneously between buildings according to an embodiment of the present invention.
7 is a graph visualizing instantaneous wind speed using wind speed values of wind speed data and substitution values derived through a GAIN model according to an embodiment of the present invention.

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

1 is a diagram illustrating an overall system for ensuring the soundness of a building according to an embodiment of the present invention.

Referring to FIG. 1 , the computing device 101 may evaluate a wind speed pattern of environmental wind generated instantaneously between buildings 102 through particle image velocimetry (PIV). The computing device 101 may record the horizontal speed of air moving between the buildings 102 by using a camera capable of interlocking with the particle image velocimetry in the buildings 102 . The particle image speedometer may visualize scattered particles by using wind generated between the buildings 102 as a tracking particle and irradiating a laser light on the tracking particle. The computing device 101 may measure wind speed data due to wind from particles visualized by the particle image velocimetry. Here, the wind speed data may include the length of the buildings 102 and the height at which the wind speed is measured, and may include a frictional speed between the buildings 102 and wind and surface roughness.

The computing device 101 may set a projection area in which a light source of the particle image speedometer is blocked in a target area including the buildings 102 where wind speed data is measured. In other words, although the particle image speedometer can predict the instantaneous wind speed between the buildings 102 , it may not be possible to collect wind speed data in a specific area as laser light is used. Accordingly, the computing device 101 may perform an in-depth investigation of wind characteristics generated by laser light blocking by replacing the instantaneous wind loss measurement using a machine learning approach (ML). Computing device 101 may utilize a generative adversarial imputation network (GAIN) model of a machine learning approach to estimate missing values of wind speed data.

Generative Adversarial Imputation Network (GAIN) models may include generative network models and discriminant network models that are trained in an adversarial manner for imputation learning. The generative network model distinguishes between the wind speed value of the wind speed data and the spoofed value, and performs a function of improving the prediction ratio of the discriminant network model. The discriminant network model can generate a spoofed value to be used in the generative network model, and can perform a function of reducing error loss of the generative network model.

Computing device 101 applies the wind speed data to a generative adversarial imputation network (GAIN) model, thereby estimating missing values of wind speed data not measured as a result of laser occlusion by the particle image velocimetry at the nearest wide location around buildings 102.

As a result, the computing device 101 integrates the particle image velocimetry and the generative adversarial imputation network model to estimate the unknown wind speed in a large area of the buildings 102 to obtain complete aerodynamic characteristics by the buildings 102, thereby comprehensively ensuring the structural integrity of the buildings and the safety of pedestrians.

2 is a flowchart illustrating a wind speed pattern evaluation method according to an embodiment of the present invention.

In step 201, the computing device may measure wind speed data by environmental wind generated instantaneously between buildings using a particle image velocimetry (PIV). The computing device may measure wind speed data indicating the speed of horizontal air moving between buildings by emitting light of the particle image velociometer at a specific time period.

In operation 202, the computing device may set a projection area blocking a light source of the particle image velocimetry in a target area including buildings where wind speed data is measured. The computing device may set a projection area having missing values among wind speed values of the wind speed data measured in the target area because the light source of the particle image speedometer is blocked.

In step 203, the computing device may replace the wind speed value of the wind speed data belonging to the projection area with zero (0) or NaN (Not a Number). The computing device may replace the missing value of the wind speed data with zero or NaN in consideration of a connection characteristic between wind speed values of the wind speed data.

In step 204 , the computing device may predict a wind speed value replaced with zero or NaN by analyzing a wind speed value other than zero or NaN of the wind speed data through imputation learning. For imputation learning, the computing device may predict a wind speed value substituted with zero or NaN for applying wind speed data to a generative network model and a discriminatory network model trained in an adversarial manner. Here, the generating network model may perform a function of improving a prediction rate of the discriminating network model by discriminating between a wind speed value and a spoofed value of the wind speed data. The discriminant network model can generate spoofed values to be used in the generative network model, and can perform a function of reducing error loss in the generative network model.

In addition, the computing device may receive a wind speed value other than zero or NaN and a spoofed value of the wind speed data through the generated network model. The computing device may discriminate between real and false components according to the vector format of wind speed data, targeting wind speed values that are not zero or non-NaN input and spoofed values through the discriminant network model. The computing device may estimate the wind speed value substituted with zero or NaN considering the real and false components.

In step 205, the computing device may visualize wind speed values as a heat map and evaluate a wind speed pattern according to the visualized heat map. The computing device may evaluate a wind speed pattern of wind instantaneously generated between buildings from an average wind speed of predicted wind speed values according to the visualized heat map.

3 is a graph for setting a projection area including an unknown wind speed value in a target area of a building according to an embodiment of the present invention.

Referring to (a) of FIG. 3 , the computing device may collect wind speed data due to wind using a particle image velocimetry. Particle image velocimetry can image particles scattered by laser light to evaluate the speed and pattern that occur instantaneously around buildings. The computing device may measure wind speed data according to a result of scattering particles being imaged.

Accordingly, the computing device may set the projection area 301 blocking the light source of the particle image speedometer in the target area including the buildings for which the wind speed data is measured according to (b) of FIG. 3 . To this end, the computing device may extract a wind speed value other than '0' by analyzing a value of the wind speed data measured through the particle image velocimetry. The computing device may set the projection area 301 having an unknown wind speed value according to the extracted wind speed value as a value other than '0'.

More specifically, the instantaneous wind speed generated in the entire area where the building exists, that is, the target area, may have a rectangular portion blocking the laser light of the particle image speedometer, and the rectangular portion may interfere with wind speed measurement due to the blocking of the laser light. Accordingly, the computing device may set an unknown wind speed value, which is not measured due to shielding of the laser light, as '0' for the instantaneous flow condition among the wind speed data.

Computing devices can utilize generative adversarial imputation network models to estimate missing values that are not directly recorded through particle image velocimetry to accurately analyze wind characteristics at specific locations on buildings. In developing the generative adversarial imputation network model, the computing device may designate an instantaneous wind speed other than '0' to replace a missing value of the wind speed data.

Accordingly, the computing device may apply wind speed data having a non-zero wind speed value and an unknown wind speed value to the generative adversarial imputation network model. The computing device may replace the wind speed value of the wind speed data belonging to the projection area with zero (0) or NaN (Not a Number). The computing device may accurately analyze the wind pressure at a selected location around the building by imputing the observed wind speed according to the particle image velocimetry using the generative adversarial imputation network model.

Finally, the computing device may utilize the machine learning approach's generative adversarial imputation network model to estimate missing values of unmeasured wind speeds in the wind speed data. The computing device may extract the missing value of the wind speed by applying the wind speed data to the generative adversarial imputation network model.

As an example, the computing device can specify the fully recorded wind speed at any 555 locations to create a generative adversarial imputation network model at all measurement points. The computing device may evaluate the efficiency of the generative adversarial imputation network model by performing various mathematical calculations through the generative adversarial imputation network model. Then, the computing device may analyze the wind pressure at the selected location of the building by applying the wind speed data to the generative adversarial transfer network model.

4 is a diagram for explaining a GAIN model with an ML algorithm according to an embodiment of the present invention.

Referring to FIG. 4 , the computing device may estimate a missing value (Null) based on known values of other variables and assumptions according to the purpose of the imputation approach of the present invention. The computing device may make a prediction for the missing value using the previously predicted wind speed value including other characteristics already present in the data store.

In the present invention, the computing device may use two different approaches to impute missing values of wind speed data.

① First approach

The first approach is an alternative approach, which may be an approach that classifies similar results based on a predetermined set of parameters before using the data for that result. In this case, the first approach can use the expected Null Data in the original database to replace empty values when working with unsampled data sets.

② Second approach

The second approach can predict missing values included in the wind speed data using a pre-verified computational approach based on the entire data set. A second approach can address the gap in available information.

In the present invention, the computing device may estimate the missing wind speed value around the building by using a calculation approach based on machine learning.

A computing device may impute wind speed data using a computational approach based on machine learning. More specifically, a computational approach based on machine learning may regard '0' as indicating no data for model training. In consideration of this, the computing device may replace the missing wind speed data with zero (0) or Not a Number (NaN). The computing device may consider the structure of the filled '0' values before predicting the incomplete values of the data set according to the machine learning approach. The computing device may supplement incomplete data by utilizing a connection characteristic between known information and unknown information to have a '0' value structure. Connection characteristics may be as follows.

① First connection characteristics

The computing device may consider the case where the '0' value was randomly generated and the case where a completely missing random (MFAR) value occurred without adjustment with the observed value.

② Second connection characteristics

A computing device may consider a case of Random Missing (MAR) if the blank data correlates with data found for other attributes.

③ Third connection characteristics

Computing devices may consider missing values (MNAR) where values are not random when the bin values depend on both observed and unobserved variables.

Thereafter, the computing device may analyze a wind speed value other than zero or NaN of the wind speed data through imputation learning and predict a wind speed value replaced with the zero or NaN. The computing device may visualize the predicted wind speed value as a heat map and evaluate a wind speed pattern according to the visualized heat map.

Computing devices may utilize generative adversarial network models of machine learning approaches. The generative adversarial network model can derive optimal results by using a feedback loop combining two independent neural networks. Here, the two independent neural networks may include a generative network model 401 and a discriminant network model 402 trained in an adversarial manner for imputation learning. The generative adversarial network model can reliably impute missing values by simultaneously operating the generative network model 401 and the discriminant network model 402 . After all, generative adversarial network models can be machine learning models that can handle imperfect data that can be found in any data set.

The generative adversarial network model can approximate the missing wind speed values in a specific area of a building and more efficiently derive the wind speed pattern by the missing wind speed values.

The discrimination network model 402 is used to discriminate the difference between the wind speed value of the wind speed data and the spoofed value, and can perform a function of improving the prediction rate of the generating network model 401 .

The generator network model 401 may generate a spoofed value, ie, a fake sample, to be used in the discriminant network model 402 and may perform a function of reducing error loss of the discriminant network model 402 .

As a result, in the present invention, by training the discriminant network model 402 and the generative network model 401 in an adversarial manner, the generative network model attempts to increase the erroneous prediction rate of the discriminant network model 402, and the discriminant network model 402 can perform to minimize the classification error loss of the generative network model.

In the following, we will explain sequentially how the generative adversarial network model works.

The generator (G) of the generative network model (401) collects the incomplete values using the vectors, and then in a subsequent step, the discriminator (D) of the discriminant network model (402) can provide those values to be used as inputs.

The present invention may use a hint mechanism to ensure that the generator (G) of the generative network model (401) learns appropriately using the discriminator (D) of the discriminant network model (402). Here, the hint mechanism may be information additionally provided to the generator G of the generation network model 401 to obtain accurate knowledge. The additionally provided information may be a hint vector H, and the hint vector H may help the generator G of the generating network model 401 generate missing data having a similar distribution pattern to actual data.

In addition, the workflow of the generative adversarial network model may be as follows.

The computing device may extract a missing value in the wind speed data by referring to Equation 1.

Referring to Equation 1, is a D-dimensional space, is a continuous variable that takes the value of V may represent a distribution function. is regarded as a random variable, and the random variable S is can be used as input. Here, Z may mean a data matrix and S may mean a mask matrix.

According to Equation 1, all is the new space ( ), where * indicates an arbitrary position other than V _i and may indicate an unmeasured value. this , a new random variable can be expressed as

Also, S in Equation 1 represents the measured component of Z, This realized replication of can be displayed as The main purpose of the procedure is to It may be to estimate the value of , and the imputed value is a distribution function can be extracted by applying

here, is for every I-th position to impute missing values in D-dimensional space. It can be a conditional distribution function given by , through which the uncertainty of the missing value predicted from the model distribution of wind speed data can be captured. Computational distribution functions are motivating using a generative adversarial network model. can be defined as The generative network model 401 and the discriminant network model 402 may be utilized in the imputation process of the generative adversarial network model to impute values not present in the data set.

In the generative adversarial network model, the generative network model 401 is , S, the noise variable and X values as input, and can be output. this defined as a function, can be defined as dimensional noise. Also, a random variable Can be expressed as Equations 2 to 3 below.

where denotes element-by-element multiplication, denotes the vector form of the predicted sample, may represent the vector format of the completed data.

On the other hand, the present invention can implement the discriminator D of the discriminant network model 402, which will be a challenge during training of the generator G of the generative network model 401. The discriminator D of the discriminant network model 402 can discriminate between real and false components for the purpose of predicting the mask vector s. Through this, the discriminator D of the discriminant network model 402 can identify whether the entire vector is genuine.

The function of the discriminator (D) of the discriminant network model 402 is is expressed as Each element of the ith of ( ) may coincide with the actual measured probability associated with the component.

In addition, the hint mechanism consists of a random variable denoted by H that can take a value in the space denoted by the hint vector H, and the missing value Distributing the hint mechanism may appear differently.

The estimated value of the hint vector H is transmitted to the discriminator D of the discriminant network model 402 along with other inputs, and the function of the discriminator D of the discriminant network model 402 can be finally expressed as in Equation 4.

here, the ith of ( ) element is i th of ( ) element is may be consistent with the probability of being conditionally measured in .

5 is a graph for confirming performance and square error according to a GAIN model according to an embodiment of the present invention.

Referring to FIG. 5 , the computing device may consider both the average variance and the average standard deviation of wind speed data surrounding the building in the evaluation criterion. The computing device may compare the actual value of the wind speed data with the value predicted by the generative adversarial network model. This can be seen in Table 1.

The effectiveness of wind speed data according to Table 1 can be evaluated by comparing predicted and measured average wind speeds in various areas of the rescue system. It can be confirmed that the degree of deviation of the predicted wind speed from the measured value according to the performance evaluation result is extremely insignificant. The generative adversarial network model showed AMSE and ASS of 0.073 and 0.965 predicted wind speed, respectively.

6 is a graph showing the average wind speed through the actual average wind speed and the GAIN model by wind generated instantaneously between buildings according to an embodiment of the present invention.

Referring to FIG. 6 , the computing device may visualize an actual wind speed pattern and an estimated wind speed pattern in an instantaneous time interval in order to fairly compare the efficiency of different algorithms. The computing device may visualize the predicted wind speed values as a heat map.

The computing device can visualize heat maps at the 555 locations geographically closest to the proposed building. The computing device can determine the optimal ML approach for missing data imputation areas through the model.

In this evaluation process, the heat map presentation approach can be performed at a total of 100 different IWVs ranging from 1 to 1,000. As shown in (a) of FIG. 6, this can be expressed as a heat map of actual and instantaneous average wind speeds attributed to each model. In addition, the actual (IWV) value may be displayed as an actual (IWV) value in a wide area of a building as shown in (b) of FIG. 6 .

The computing device of the present invention may manually substitute a value of 0 for each actually recorded wind speed measurement at various locations to develop and train the ML model. The computing device may use the GAIN method according to the heat map attributed to 0 to attribute the wind speed data and then visualize the generated heat map.

7 is a graph visualizing instantaneous wind speed using wind speed values of wind speed data and substitution values derived through a GAIN model according to an embodiment of the present invention.

The estimated instantaneous wind speed of the region selected using the GAIN model can be represented as shown in FIG. In the present invention, the computing device may represent the characteristics of the wind speed pattern by visualizing a heat map similar to each other.

The computing device used instantaneous average wind speed measurements over a wide area around the building, and could develop an optimal ML model to impute the missing wind speed across the vast missing area at the nearest location around two adjacent buildings.

Computing devices can provide significant imputation performance of instantaneous wind speeds in the wide area closest to the location of the building model, and by utilizing generative adversarial network models, computing devices can approximate missing wind speeds even for data sets with a higher proportion of missing values.

As a result, the generative adversarial network model can estimate wind speeds not observed in the particle image velocimetry as a result of laser light shielding both at random and in the nearest wide area around two adjacent buildings.

Meanwhile, the method according to the present invention is written as a program that can be executed on a computer and can be implemented in various recording media such as magnetic storage media, optical reading media, and digital storage media.

Implementations of the various techniques described herein may be implemented in digital electronic circuitry, or in computer hardware, firmware, software, or combinations thereof. Implementations may be implemented as a computer program product, i.e. a computer program tangibly embodied in an information carrier, e.g. a machine readable storage device (computer readable medium) or radio signal, for processing by, or for controlling the operation of, a data processing device, e.g. a programmable processor, computer, or operation of multiple computers. Computer programs, such as the computer program(s) described above, may be written in any form of programming language, including compiled or interpreted languages, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program can be deployed to be processed on one computer or multiple computers at one site or distributed across multiple sites and interconnected by a communication network.

Processors suitable for processing a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from read only memory or random access memory or both. Elements of a computer may include at least one processor that executes instructions and one or more memory devices that store instructions and data. In general, a computer may include one or more mass storage devices that store data, such as magnetic, magneto-optical disks, or optical disks, or may be coupled to receive data from or transmit data to, or both. Information carriers suitable for embodying computer program instructions and data include, for example, semiconductor memory devices, for example, magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as compact disk read only memory (CD-ROM) and digital video disks (DVD), magneto-optical media such as floptical disks, read memory only (ROM), random access memory (RAM). , flash memory, EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), and the like. The processor and memory may be supplemented by, or included in, special purpose logic circuitry.

In addition, computer readable media may be any available media that can be accessed by a computer, and may include both computer storage media and transmission media.

Although this specification contains many specific implementation details, they should not be construed as limiting on the scope of any invention or what is claimed, but rather as a description of features that may be unique to a particular embodiment of a particular invention. Certain features that are described in this specification in the context of separate embodiments may also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable subcombination. Further, while features may operate in particular combinations and be initially depicted as such claimed, one or more features from a claimed combination may in some cases be excluded from that combination, and the claimed combination may be modified as a subcombination or variation of a subcombination.

Similarly, while actions are depicted in the drawings in a particular order, it should not be construed as requiring that those actions be performed in the specific order shown or in the sequential order, or that all depicted actions must be performed to obtain desired results. In certain cases, multitasking and parallel processing can be advantageous. Further, the separation of various device components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the program components and devices described may generally be integrated together into a single software product or packaged into multiple software products.

On the other hand, the embodiments of the present invention disclosed in this specification and drawings are only presented as specific examples to aid understanding, and are not intended to limit the scope of the present invention. In addition to the embodiments disclosed herein, it is obvious to those skilled in the art that other modified examples based on the technical idea of the present invention can be implemented.

101: computing device
102: Architecture

Claims (14)

Measuring wind speed data due to environmental wind generated instantaneously between buildings using a particle image velocimetry (PIV);
setting a projection area blocking a light source of the particle image speedometer in a target area including buildings where the wind speed data is measured;
replacing wind speed values of the wind speed data belonging to the projection area with zero (0) or NaN (Not a Number);
supplementing the wind speed value substituted with the zero or NaN by utilizing a connection characteristic between known information and unknown information to have a zero (0) value structure;
predicting a wind speed value replaced with zero or NaN by analyzing a wind speed value other than zero or NaN of the wind speed data through imputation learning; and
Evaluating a wind speed pattern according to the visualized heat map by visualizing the predicted wind speed value as a heat map
including,
The connection characteristics are,
A first linkage characteristic that takes into account the case where zero values are randomly generated, or completely missing random (MFAR) values occur without adjustment with observed values;
A second connection characteristic considering the case where the wind speed value substituted with zero or NaN is randomly missing (MAR) because it is related to data found for other attributes, and
A wind speed pattern evaluation method comprising a third connection characteristic considering a case where a non-random value is missing (MNAR) because the wind speed value substituted with zero or NaN depends on both observed factors and unobserved variables.
According to claim 1,
Measuring the wind speed data,
A wind speed pattern evaluation method for measuring wind speed data representing the speed of horizontal air moving between the buildings by emitting light from a particle image speedometer at a specific time period.
According to claim 1,
In the setting step,
The wind speed pattern evaluation method of setting a projected area having a missing value as the light source of the particle image speedometer is blocked among the wind speed values of the wind speed data measured in the target area.
According to claim 3,
The replacing step is
The wind speed pattern evaluation method of replacing missing values of the wind speed data with zero (0) or NaN (Not a Number) in consideration of connection characteristics between wind speed values of the wind speed data.
According to claim 1,
The predicting step is
Wind speed pattern evaluation method for predicting the wind speed value substituted with zero or NaN, applying wind speed data to the generative network model and the discriminatory network model trained in an adversarial manner for the imputation learning. According to claim 5,
The discriminant network model,
Distinguishing between a wind speed value of the wind speed data and a spoofed value, and performing a function of improving a prediction ratio of a generated network model;
The generative network model,
A wind speed pattern evaluation method for generating a spoofed value to be used in the discriminant network model and performing a function of reducing error loss of the generation network model.
According to claim 5,
The predicting step is
receiving a wind speed value other than zero or NaN and a spoofed value of the wind speed data through the generated network model;
discriminating a real component and a false component according to a vector format of wind speed data from the inputted wind speed value other than zero or NaN and the spoofed value through the discrimination network model; and
Predicting the wind speed value substituted with the zero or NaN in consideration of the distinguished real component and false component
Wind speed pattern evaluation method comprising a.
According to claim 1,
The evaluation step is
A wind speed pattern evaluation method of evaluating a wind speed pattern of wind generated instantaneously between the buildings from the average wind speed of the predicted wind speed values according to the visualized heat map.
A computing device for performing a wind speed pattern evaluation method,
The computing device includes a processor,
the processor,
Using a particle image speedometer, measure wind speed data by environmental winds generated instantaneously between buildings,
Setting a projection area blocking a light source of the particle image speedometer in a target area including buildings where the wind speed data is measured;
Replace the wind speed value of the wind speed data belonging to the projection area with zero or NaN;
The wind speed value replaced by the zero or NaN is supplemented by utilizing the connection characteristic between known information and unknown information to have a zero (0) value structure,
Analyzing a wind speed value other than zero or NaN of the wind speed data through imputation learning to predict a wind speed value replaced with the zero or NaN,
Visualizing the predicted wind speed value as a heat map to evaluate a wind speed pattern according to the visualized heat map;
The connection characteristics are,
A first linkage characteristic that takes into account the case where zero values are randomly generated, or completely missing random (MFAR) values occur without adjustment with observed values;
A second connection characteristic considering the case where the wind speed value substituted with zero or NaN is randomly missing (MAR) because it is related to data found for other attributes, and
and a third connection characteristic considering a case where a non-random value is missing (MNAR) because the wind speed value substituted with zero or NaN depends on both an observed factor and an unobserved variable.
According to claim 9,
the processor,
A computing device for measuring wind speed data indicating the speed of horizontal air moving between the buildings by emitting light of a particle image speedometer at a specific time period.
According to claim 9,
the processor,
A computing device that sets a projection area having a missing value among wind speed values of the wind speed data measured in the target area as the light source of the particle image speedometer is blocked.
According to claim 11,
the processor,
A computing device that replaces missing values of the wind speed data with zero (0) or Not a Number (NaN) in consideration of connection characteristics between wind speed values of the wind speed data.
According to claim 10,
the processor,
A computing device for predicting wind speed values substituted with zero or NaN, applying wind speed data to a generative network model and a discriminatory network model trained in an adversarial manner for the imputation learning.
According to claim 13,
the processor,
Receive a zero or non-NaN wind speed value and a spoofed value of the wind speed data through the generation network model,
Distinguish real components and false components according to the vector format of wind speed data for the wind speed value and the spoofed value that are not zero or NaN through the discriminant network model;
A computing device for predicting the wind speed value replaced with the zero or NaN in consideration of the distinguished real component and false component.