KR102573134B1 - Apparatus for measuring pollutants using AI and method for measuring pollutants using AI - Google Patents

Abstract

A pollutant measuring device using artificial intelligence and a pollutant measuring method using artificial intelligence are provided. The pollutant measuring device provides standard data consisting of concentration measurement values measured by a plurality of sensors for pollutants in the standard gas containing pollutants and data of actual pollutant concentration values of the standard gas used in generating the standard data. A pattern formation module that derives and collects a distribution pattern of concentration measurement values representing the actual concentration of pollutants in standard gas by learning the correlation by an artificial intelligence learning algorithm, and a plurality of identical sensors used for standard gas measurement in the measurement area In a measurement module that measures the concentration of pollutants contained in the measurement gas with different measurement sensors of the measurement gas and generates and transmits measurement data consisting of measurement values of concentrations measured by a plurality of measurement sensors for pollutants in the measurement gas, and in the measurement module The concentration measurement value distribution of the transmitted measurement data is compared with the distribution patterns provided by the pattern formation module, but the correlation is learned by the artificial intelligence learning algorithm to determine the distribution pattern with the highest similarity with the concentration measurement value distribution of the measurement data and a pattern-comparison calculation module for calculating the pollutant concentration represented by the determined distribution pattern and the pollutant corresponding thereto as an actual measured value for the pollutant in the measured gas.

Description

Pollutant measuring device using artificial intelligence and pollutant measuring method using artificial intelligence {Apparatus for measuring pollutants using AI and method for measuring pollutants using AI}

The present invention relates to a measuring device and method for measuring pollutants (hereinafter, 'measuring' and 'measuring' have substantially the same meaning), and more particularly, to a pollutant contained in gas based on artificial intelligence technology. It relates to a pollutant measuring device using artificial intelligence capable of more accurately measuring the concentration of substances, and a pollutant measuring method using artificial intelligence.

Pollutants mixed in the air are emitted from various sources. For example, industrial facilities that process chemical products, power plants that generate electricity by burning fuel, and residential facilities equipped with heating facilities can all be emission sources that emit pollutants, and various vehicles operated by engines are also moving pollutants. may be an emission source. Most of the pollutants are contained in the gas produced from these pollutant sources and are emitted with the gas.

Therefore, it is essential to measure pollutants contained in gas as a basic step for removing pollutants in the air or suppressing the spread of pollutants. In addition to the examples, there may be gases produced in various ways, and various contaminants may exist in such gases. If the pollutants are accurately measured, the treatment method can be accurately determined, so that air pollution can be dealt with more efficiently.

However, conventional measurement of contaminants in gas is usually performed using a specific sensor (eg, Korean Patent No. 10-1734355, etc.), and the range of application of these sensors is often limited due to their operating principle or structure, which is a problem. For example, there are many sensors that have high sensitivity to one contaminant but not to other contaminants, and when other contaminants are mixed, the sensor's detection ability unexpectedly drops. Therefore, it is quite difficult to accurately measure the concentration of pollutants in the gas. In addition, it is not only very difficult to configure measurement equipment with only expensive sensors with high detection ability, but also expensive sensors may have reduced sensitivity when various contaminants are mixed. Since it is used only to the extent of determining the , it is difficult to accurately measure the concentration of various contaminants contained in the gas with the prior art.

On the other hand, the inventors of the present invention research (No. S0426-20-1004, IoT monitoring system development for fine dust reduction in small-scale emissions workplaces) with the support of the National IT Industry Promotion Agency with financial resources from the government (Ministry of Science and ICT in 2020) ) was performed.

Republic of Korea Patent Registration No. 10-1734355, (2017. 05. 15), specification

The technical problem of the present invention is to solve this problem, and to provide a pollutant measuring device using artificial intelligence that can more accurately measure the concentration of pollutants contained in gas based on artificial intelligence technology. In addition, it is to provide a pollutant measurement method using artificial intelligence that can more accurately measure the concentration of pollutants contained in gas based on artificial intelligence technology.

The technical problem of the present invention is not limited to the above-mentioned problems, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following description.

A pollutant measuring device using artificial intelligence according to the present invention includes standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in a standard gas containing at least one kind of pollutant, and the standard data Data of the actual pollutant concentration values of the standard gas used at the time of generation are provided, and the correlation between the distribution of measured concentration values shown in the standard data and the actual pollutant concentration values of the standard gas is learned by an artificial intelligence learning algorithm. a pattern forming module for deriving and collecting a distribution pattern of measured concentration values representing actual pollutant concentrations of the standard gas; Arranged in the measurement area, a plurality of different measurement sensors identical to the sensors used for measuring the standard gas measure the concentration of pollutants contained in the measurement gas, and a plurality of different measurement sensors for pollutants in the measurement gas. a measurement module for generating and transmitting measurement data consisting of measured concentration values; And comparing the concentration measurement value distribution of the measurement data transmitted from the measurement module with the distribution patterns provided by the pattern forming module, and artificial intelligence learning the correlation between the concentration measurement value distribution of the measurement data and the distribution pattern. The distribution pattern having the highest similarity with the concentration measurement value distribution of the measured data is determined by learning by an algorithm, and the pollutant concentration represented by the determined distribution pattern and the pollutant corresponding thereto are determined for the pollutant in the measured gas. A pattern-comparison calculation module for calculating actual values, wherein one distribution pattern consists of an array of numbers corresponding to concentration measurement values measured by the plurality of different sensors and is distinguished from other distribution patterns, and the measurement data It is characterized in that information is compared in the form of a matrix with a measurement pattern consisting of an array of numbers corresponding to the measured concentration values of .

The plurality of sensors used for measuring the standard gas and the plurality of measuring sensors of the measurement module may be formed of a combination of sensors of different types that measure pollutants.

The different types of sensors may include two or more selected from among metal oxide sensors, polymer sensors, photoacoustic spectroscopy sensors, surface plasmon resonance sensors, light scattering particle measurement sensors, and microcantilever sensors.

The artificial intelligence learning algorithm of the pattern forming module and the pattern-comparison calculation module includes an artificial neural network algorithm in which at least one hidden layer consisting of a plurality of calculation nodes is formed between an input layer and an output layer, and the artificial The neural network algorithm may include a convolutional neural network algorithm.

The number of pollutants contained in the standard gas is plural, the distribution pattern is composed of a distribution of measured concentration values representing concentrations of a plurality of pollutants, and the measured value calculated by the pattern-comparison calculation module is represented by the distribution pattern. A plurality of contaminant concentrations and corresponding contaminant types may be included.

The pattern forming module learns information about the error between the distribution of the measured concentration value shown in the standard data and the actual pollutant concentration value of the standard gas, and sets an allowable range among the plurality of sensors used for measuring the standard gas. It is possible to determine a sensor having a measurement error that exceeds

The pattern forming module transmits an alarm signal to detect a sensor having a measurement error exceeding the allowable range among the plurality of sensors used for measuring the standard gas and the same tolerance range among the plurality of measurement sensors of the measurement module. It is possible to instruct replacement of a measurement sensor with a measurement error that exceeds.

The pollutant measurement method using artificial intelligence according to the present invention includes (a) standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in a standard gas containing at least one kind of pollutant, and A computer processes the data of the actual pollutant concentration values of the standard gas used in generating the standard data, and the computer determines the correlation between the distribution of the measured concentration values shown in the standard data and the actual pollutant concentration values of the standard gas. Learning by an artificial intelligence learning algorithm to derive and collect a distribution pattern of concentration measurement values representing actual pollutant concentrations of the standard gas; (b) In the measurement area, the concentration of pollutants contained in the measurement gas is measured with a plurality of different measurement sensors identical to those used for the measurement of the standard gas in step (a), and the pollutants in the measurement gas are measured. Generating measurement data consisting of concentration measurement values measured by a plurality of different measurement sensors and transmitting the same to a computer; and (c) a computer compares the concentration measurement value distribution of the measurement data transmitted in step (b) with the collected distribution patterns, and determines the correlation between the concentration measurement value distribution of the measurement data and the distribution pattern. It is learned by an artificial intelligence learning algorithm to determine the distribution pattern having the highest similarity with the concentration measurement value distribution of the measurement data, and the pollutant concentration represented by the determined distribution pattern and the pollutant corresponding thereto Contamination in the measurement gas Comprising the step of calculating the measured value for the substance, wherein the one distribution pattern collected in the step (a) is composed of an array of numbers corresponding to the concentration measurement values measured by the plurality of different sensors, and different distribution patterns It is distinguished from, characterized in that information in the form of a matrix is compared with a measurement pattern consisting of an array of numbers corresponding to the measured concentration values of the measurement data in step (c).

The artificial intelligence learning algorithm of steps (a) and (c) includes an artificial neural network algorithm in which at least one hidden layer consisting of a plurality of calculation nodes is formed between the input layer and the output layer, and the artificial neural network The algorithm may include a convolutional neural network algorithm.

The number of contaminants contained in the standard gas in step (a) is plural, and the distribution pattern is composed of a distribution of measured concentration values representing concentrations of a plurality of contaminants, and the measured values calculated in step (c) are: A plurality of pollutant concentrations represented by the distribution pattern and a plurality of pollutant types corresponding thereto may be included.

In the step (a), the computer learns information about the error between the distribution of the concentration measurement value shown in the standard data and the actual pollutant concentration value of the standard gas, and among the plurality of sensors used to measure the standard gas A process of determining a sensor having a measurement error exceeding an acceptable range may be included.

According to the present invention, it is possible to provide more accurate measurement values by improving the accuracy of measurement values related to concentrations, types, etc. of pollutants in gas, which have not been easy to accurately measure in the past. In particular, it is possible to simultaneously provide information on the concentration values and types of various contaminants contained in gas in a method based on artificial intelligence technology, so that various tasks based on measurements such as gas analysis can be performed more efficiently. In addition, through the present invention, it is possible to utilize various types of sensors, from expensive sensors with relatively excellent detection ability to low-cost sensors with relatively low detection ability, so that the configuration of measuring equipment can be diversified, and existing sensors possessed in the field can be diversified. Even with these systems, more improved measurement values can be provided, so that significant improvements can be made to various operations or processes related to pollutant measurement.

1 is a block diagram of a pollutant measuring device using artificial intelligence according to an embodiment of the present invention.
FIG. 2 is a conceptual diagram illustrating a method of generating standard data shown in FIG. 1 .
FIG. 3 is a view conceptually showing the process of deriving the distribution pattern shown in FIG. 1 and its contents.
4 is a conceptual diagram illustrating a method of generating measurement data shown in FIG. 1 .
FIG. 5 is a diagram conceptually illustrating a process of comparing measurement data and distribution patterns of FIG. 4 and contents of the measurement data.
6 is a diagram for explaining the artificial intelligence learning algorithm of the present invention.
7 is a flowchart illustrating a pollutant measurement method using artificial intelligence according to an embodiment of the present invention.

Advantages and features of the present invention and methods for achieving them will become clear with reference to the detailed description of the following embodiments in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms, but only the present embodiments make the disclosure of the present invention complete and the common knowledge in the art to which the present invention belongs. It is provided to fully inform the possessor of the scope of the invention, and the invention is defined only by the claims. Like reference numerals designate like elements throughout the specification.

In this specification, 'measurement' is substantially the same as 'measurement'.

Hereinafter, a pollutant measuring device using artificial intelligence and a pollutant measuring method using artificial intelligence according to the present invention will be described in detail with reference to FIGS. 1 to 7 . For concise and clear description, first, the configuration and operational effects of the pollutant measuring device capable of performing the pollutant measuring method will be described in detail with reference to FIGS. 1 to 6, and then the pollutant measuring method will be described in detail based on this. let it do

1 is a block diagram of a pollutant measuring device using artificial intelligence according to an embodiment of the present invention.

Referring to FIG. 1, the pollutant measuring device 1 using artificial intelligence according to the present invention (hereinafter, referred to as a 'pollutant measuring device' has the same meaning) includes a pattern forming module 100, pattern-comparison calculation It may be configured in a form including a module 300 and a measurement module 200 . Through this configuration, the pollutant measuring device 1 derives a distribution pattern of measured concentration values that matches the actual concentration value of pollutants in the gas, and compares the derived distribution pattern with the measured data to produce more accurate measurement results. can The distribution pattern of the present invention is a combination of concentration measurement values that can be calculated from a plurality of sensors derived by learning pre-collected measurement data by an artificial intelligence learning algorithm, which operates with different measurement objects, measurement methods, and/or precision. It is also the information that shows the overall distribution of the measured value and the response pattern of the different sensors that do when the contaminant(s) in the gas is present at a specific concentration. Therefore, one distribution pattern simultaneously represents the pollutant(s) contained in the gas and their specific concentration, and by determining the same or most similar distribution pattern as the measured data obtained at the present time, not only the types of pollutants mixed in the measured gas but also their specific concentrations are determined. Concentration can be calculated with high accuracy. In addition, in the process of deriving the distribution pattern and comparing the distribution pattern with the measurement data, the result of comprehensively considering various possible correlations including non-linear correlation through data learning by an artificial intelligence learning algorithm is calculated, resulting in a plurality of different sensors. It is possible to measure pollutants with higher reliability than before through the process of deriving a distribution pattern for the measured values of the pollutants and comparing them with the measured data.

The contaminant measuring device 1 of the present invention is specifically configured as follows. The pollutant measuring device 1 generates standard data (see FIG. 2) consisting of concentration measurement values measured by a plurality of different sensors for pollutants in a standard gas containing at least one kind of pollutant (see FIG. 2), and standard data. receive data of the actual pollutant concentration value of the standard gas used in the test, and learn the correlation between the distribution of the measured concentration value shown in the standard data and the actual pollutant concentration value of the standard gas by an artificial intelligence learning algorithm. A pattern forming module 100 that derives and collects a distribution pattern (see FIG. 3) of concentration measurement values representing actual pollutant concentrations, arranged in the measurement area, and a plurality of different measurements identical to the sensors used for standard gas measurement. Measurement module (200 ), and compare the concentration measurement value distribution of the measurement data transmitted from the measurement module 200 with the distribution patterns provided by the pattern forming module 100, but artificially It is learned by an intelligent learning algorithm to determine the distribution pattern that has the highest similarity with the concentration measurement value distribution of the measured data (see Fig. 5), and the pollutant concentration represented by the determined distribution pattern and the pollutant corresponding to it are polluted in the measured gas It includes a pattern-comparison calculation module 300 that calculates the measured value for the material.

In particular, according to an embodiment of the present invention, the plurality of sensors used for measuring the standard gas and the plurality of measuring sensors of the measurement module 200 may be formed by a combination of sensors of different types that measure pollutants. Different types of sensors may include two or more selected from metal oxide sensors, polymer sensors, photoacoustic spectroscopy sensors, surface plasmon resonance sensors, light scattering particle measurement sensors, and microcantilever sensors. In addition, the number of pollutants contained in the standard gas is plural, and the distribution pattern is composed of a distribution of measured concentration values representing the concentrations of a plurality of pollutants. A plurality of pollutant concentrations represented and a plurality of corresponding pollutant types may be included. That is, the present invention not only can more accurately measure the concentration of pollutants by utilizing various sensors of different types, which have been used separately in the past, but also through a method of comprehensively informationalizing and comparing the distribution patterns of a plurality of sensors. It has the feature of being able to calculate the types of multiple pollutants and their concentration values at once. Hereinafter, each configuration and their operational effects will be described in more detail with reference to the drawings.

Referring to FIG. 1, the pattern forming module 100 includes standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in the standard gas, and actual pollutant concentrations of the standard gas used when generating the standard data. It receives data of value and processes it. The pattern forming module 100 learns a plurality of data provided by an artificial intelligence learning algorithm to determine the correlation between the standard data and the actual pollutant concentration value of the standard gas, and from this, the actual contamination of the standard gas used when generating the standard data. Derive a distribution pattern of concentration measurements representing the concentration of a substance. This process includes a series of data calculation processes performed on a computer, and the pattern forming module 100 capable of performing such calculation processes or the pattern-comparison calculation module 300 to be described later may be implemented on one or a plurality of computers. . For example, the pattern forming module 100 or the pattern-comparison calculation module 300 may be composed of one or a plurality of program modules installed in one or a plurality of computers. Each program module may include one or a plurality of operation programs, and may be formed as a kind of software in which related programs are combined, or may be formed in a form including hardware storing the same together with such software. If necessary, the pattern forming module 100 or the pattern-comparison calculation module 300 may even include a central arithmetic unit capable of driving an arithmetic program, and in that case, one or a plurality of computers themselves form the pattern forming module 100 Alternatively, a pattern-comparison calculation module 300 may be formed. However, it is not necessary to be limited in this way, and the pattern forming module 100 or the pattern-comparison calculation module 300 may be formed in the form of a program module or the like to share at least a portion of hardware with each other. The pattern formation module 100 or the pattern-comparison calculation module 300 may be implemented on a computer in various forms.

The data provided to the pattern forming module 100 is not data on the measured gas measured by the actual measurement module 200, but rather consists of data on the pre-collected standard gas. The standard gas can be artificially created, and the data obtained by generating the standard gas function as learning data for deriving the above-described distribution pattern. These data include both measurement data obtained by measuring the standard gas containing pollutants and data on the actual concentration of the standard gas. That is, the pattern forming module 100 includes standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in a standard gas containing at least one kind of pollutant, and a standard used when generating the standard data. The data of the actual pollutant concentration value of the gas is provided and calculated. A more detailed description of the standard gas is as follows.

FIG. 2 is a conceptual diagram illustrating a method of generating standard data shown in FIG. 1 .

Referring to FIG. 2 , the standard gas may be generated by, for example, the standard gas generator 10 . The standard gas generator 10 may be formed in various shapes, but may have a certain space in which gas can stay, such as a chamber, for example. The standard gas may be generated using, for example, a standard material capable of generating pollutants, and the type of pollutant may vary without particular limitation depending on the gas to be simulated. For example, if an exhaust gas containing gaseous and/or particulate pollutants is to be simulated, the gaseous and/or particulate pollutants may be contaminants and the standard substances capable of generating them (concentration as a substance corresponding to the pollutant) It is possible to generate pollutants by using a substance of known concentration, for example, a substance such as nitrogen monoxide or dust of which concentration is known. By mixing these contaminants with an inert gas (eg, nitrogen), a standard gas containing one or a plurality of contaminants at a specific concentration can be generated. The types of contaminants contained in the standard gas do not need to be limited, and the concentration of each contaminant contained can be varied. By combining various contaminants as desired and adjusting the ratio between the inert gas and the contaminant as desired, various standard gases containing various contaminants at various concentrations can be generated. Pollutants do not necessarily need to be limited to substances of a specific category, such as substances harmful to the human body, and substances that are judged to need to derive a distribution pattern can be arbitrarily combined in standard gas and included as pollutants. Such a substance may be, for example, one or more selected from “air pollutants,” and climate/ecosystem-changing substances. Air pollutants, 　 and climate/ecosystem change-inducing substances may be defined under the Air Quality Conservation Act (Enforcement 　 2020.5.27) of the Republic of Korea, respectively. That is, pollutants may be gas and/or particulate matter recognized as causes of air pollution, greenhouse gases, etc. as gaseous substances that may bring changes to the ecosystem due to global warming. More specifically, bromine, aluminum, vanadium, manganese, iron, zinc, selenium, antimony, tin, tellurium, barium, phosphorus, cadmium, lead, chromium, arsenic, mercury, copper, chlorine, nickel, phenol, beryllium or more Compounds, carbon monoxide, nitrogen monoxide, ammonia, nitrogen oxides, sulfur oxides, methyl sulfide, methyl disulfide, mercaptans, amines, carbon tetrachloride, carbon disulfide, hydrocarbons, boron compounds, aniline, benzene, styrene, acrolein, cyanides, fluorides , asbestos, vinyl chloride, dioxin, propylene oxide, polychlorinated biphenyl, chloroform, formaldehyde, benzidine, 1,3-butadiene, polycyclic aromatic hydrocarbons, ethylene oxide, dichloromethane, tetrachloroethylene, 1,2-dichloroethane ,　ethylbenzene,　trichloroethylene,　acrylonitrile,　hydrazine,　vinyl acetate,　bis(2-ethylhexyl)phthalate,　dimethylformamide,　dust, soot,　soot, volatile organic compounds, carbon dioxide, methane, nitrous oxide, hydrofluorocarbons , perfluorocarbon, sulfur hexafluoride, chlorofluorocarbon, hydrochlorofluorocarbon, and the like. Contaminants in the standard gas may be the same as contaminants in the measurement gas to be described later, and therefore, various contaminants such as gas and/or particulate contaminants can be measured by the present invention.

In this way, a standard gas containing one or a plurality of contaminants can be generated, and when the standard gas is generated, the concentration of each contaminant can be changed to obtain data on the standard gas with various properties. For example, a situation in which a standard gas that simulates a specific gas (eg, the exhaust gas described above) is changed to various concentrations by fixing the type of pollutant and changing the concentration of each pollutant to have different values. can be reproduced and data about it can be obtained with a plurality of sensors. While the standard gas generator 10 generates the standard gas in this way, the concentration is measured with a plurality of different sensors to generate standard data consisting of the measured concentration values measured by the plurality of different sensors.

The standard gas generated by the standard gas generator 10 may be brought into contact with the sensor by, for example, arranging a plurality of sensors (sensor type 1 to n, where n is a natural number) in the chamber. The number of sensors can be increased as desired and can be decreased if necessary. A plurality of different sensors (sensor types 1 to n, where n is a natural number) for measuring the concentration of the standard gas may be formed as a combination of sensors of different types, in particular, measuring pollutants. Such different types of sensors may be, but are not necessarily limited to, two or more selected from, for example, metal oxide sensors, polymer sensors, photoacoustic spectroscopy sensors, surface plasmon resonance sensors, light scattering particle measurement sensors, and microcantilever sensors. In other words, it is a concentration measurement value obtained by measuring the concentration of the pollutant in the above-mentioned standard gas by using a plurality of different types of sensors that operate in completely different ways and have different application ranges and errors, and also have different effects on a plurality of pollutants. standard data can be created. When the number of the plurality of sensors is increased, the type may be increased as well, and standard data with an increased amount of information may be generated accordingly. A plurality of sensors, for example, can be exposed to the generated standard gas at the same time for the same amount of time to calculate a measurement value, and one unit standard data calculated accordingly is, for example, the type of pollutant detected from each sensor. and their concentration measurements can be configured in a combined form. As described above, the standard data may be repeatedly generated while changing the concentration of the same type of pollutant, or may be repeatedly generated while changing not only the concentration but also the type of pollutant. In this way, a plurality of different sensors can generate and collect standard data consisting of concentration measurement values of a standard gas containing one or a plurality of contaminants.

At this time, the actual pollutant concentration value of the standard gas used when generating the standard data may be directly provided from the standard gas generator 10 or the like. For example, the type and amount of the standard material used when generating the standard gas in the standard gas generator 10, the type and amount of pollutants generated therefrom, and the amount of air mixed with the standard gas, etc. Since it is known, the pollutants and their concentrations in the generated standard gas can be accurately known through recorded information and calculations. Therefore, while generating the standard gas in the standard gas generator 10, it is possible to collect data on the actual pollutant types and the actual pollutant concentration values contained therein. The data of actual pollutant concentration values for the standard gas collected in this way are stored together with standard data consisting of concentration measurement values measured by a plurality of different sensors, for example, in a separate storage such as the data storage unit 20. It may be recorded and provided to the pattern forming module 100.

A plurality of sensors generating standard data will be described in more detail by way of example. As described above, these are composed of different types of sensors for measuring contaminants, and one of them may be, for example, a metal oxide sensor. The metal oxide sensor uses the principle that a substance in a gas reacts with oxygen on the surface of a metal oxide to change electrical resistance, and can be manufactured using a semiconductor material or the like. Also, any one of these may be, for example, a polymer sensor. Polymer sensors use the principle that electrical properties fluctuate as a material in a gas is adsorbed to a conductive polymer and a composite in which a conductive material is mixed, and there may be various types depending on the polymer. Also, any one of these may be, for example, a photoacoustic spectroscopy sensor. The photoacoustic spectroscopy sensor is based on the principle that the molecular binding energy differs depending on the substance in the gas and the wavelength of light absorbed is also different. Depending on the light source, there may be several types. Also, any one of these may be, for example, a surface plasmon resonance sensor. A surface plasmon resonance sensor is a device in which the momentum change of the quantized collective motion (surface plasmon) of the charge density oscillating with the same energy, the same momentum, and the same phase on the surface of a metal medium is applied to a dielectric medium (which can be a gas or a material in a gas) in contact with an interface. It is based on the principle that the refractive index varies according to the local refractive index change of (there is), and it can be formed to maintain the resonance condition (condition where energy resonance absorption occurs) of the surface plasmon according to the changed refractive index by adjusting the light source, and the selection of metal type and light source It can be configured in various ways through adjustments, etc. Also, any one of these may be, for example, a light scattering particle measurement sensor. The light scattering particle measurement sensor uses the principle of generating scattered light when a light source is irradiated on a granular material, and can measure the diameter, number, and concentration of particles by detecting the scattered light. can be configured. With this light-scattering particle measurement sensor, it is possible to measure granular contaminants, such as fine dust particles. Also, any one of these may be, for example, a micro-cantilever sensor. A microcantilever sensor uses a principle in which a substance in a gas is adsorbed to a microcantilever and changes a vibration resonance frequency according to a change in load, and may have various forms depending on the formation method of the microcantilever. As described above, each of these sensors is a type of sensor having different measuring methods of pollutants, and standard data consisting of measured concentration values of pollutants in a standard gas can be generated by combining sensors having different types of measuring methods. can

Standard data for the standard gas can be generated by a combination including two or more selected from the exemplified metal oxide sensor, polymer sensor, photoacoustic spectroscopy sensor, surface plasmon resonance sensor, light scattering particle measurement sensor, and micro cantilever sensor, but this Since this is not a limiting meaning, other types of sensors can be added as needed to generate standard data with a combination of more diversified sensors. Each of these sensors has a different measurement method and different operation characteristics, for example, selectivity for pollutants, measurement accuracy, and measurement range are different, and the degree of influence by operating conditions such as temperature and humidity When present, the extent to which they are affected by mutual interference may all appear differently. Therefore, from a large number of standard data generated by exposing them together to the standard gas, the effects of such differences or mutual interference can be identified as a pattern, and the concentration measurement representing the actual concentration of pollutants contained in the standard gas when the standard data was generated The distribution pattern of values can be derived.

The pattern forming module 100 measures the concentration representing the actual pollutant concentration of the standard gas by learning the correlation between the distribution of the measured concentration values shown in the standard data and the actual pollutant concentration value of the standard gas by an artificial intelligence learning algorithm. Derive and collect the distribution pattern of values. In the present invention, the artificial intelligence learning algorithm can be used to learn data and identify correlations and similarities between data. Specifically, the process of the pattern formation module 100 deriving a distribution pattern (see FIG. 3), A pattern-comparison calculation module (see 300 in FIG. 1 ) may be used in the process of comparing the measurement data of the measurement module (see 200 in FIG. 1 ) with the collected distribution patterns (see FIG. 5 ). Therefore, referring to FIG. 6 for a moment, the artificial intelligence learning algorithm of the present invention will be described in more detail. The content of the artificial intelligence algorithm described below can be applied to both the pattern forming module 100 and the pattern-comparison calculation module 300 described later.

6 is a diagram for explaining the artificial intelligence learning algorithm of the present invention.

Referring to FIG. 6, the artificial intelligence learning algorithm of the present invention (the artificial intelligence learning algorithm of the pattern formation module 100 and the pattern-comparison calculation module 300, which is the artificial intelligence learning algorithm shown in the pollutant measurement method of the present invention described later) Algorithm] includes, for example, an artificial neural network (ANN) algorithm in which at least one hidden layer consisting of a plurality of calculation nodes (H _n ) is formed between an input layer and an output layer, and the artificial neural network The algorithm may include a convolutional neural network (CNN) algorithm. By such an artificial intelligence learning algorithm, for example, data can be repeatedly learned using a machine-learning method. For example, in an artificial neural network algorithm, nodes of each layer may be weighted functions, and as shown, artificial neurons may be formed in a network form, and calculations may be performed according to the algorithm. Accordingly, data may be input to the nodes (In ₎ of the input layer and propagate to the nodes (O _n ) of the output layer through a plurality of nodes formed in the hidden layer. In this process, the weight of the node may be updated by back propagation or the like to optimize the result. From the results of these operations, for example, modeling in which various correlations between input data, such as nonlinear correlations, are possible, and from this, correlations and similarities between data that are difficult to grasp conventionally easily in a plurality of input data can be identified. . For example, an artificial neural network that performs a desired type of operation may be configured by appropriately adjusting or selecting each node or a function that synthesizes nodes or processes a result of a node.

Therefore, for example, in the case of the pattern forming module (see 100 in FIG. 1), the concentration measurement value shown in the standard data is obtained from a plurality of standard data and data of the actual pollutant concentration value of the standard gas used in generating the standard data. It can be configured to learn the data with an artificial neural network algorithm that can learn the correlation between the distribution and the actual pollutant concentration value of the standard gas, and from this it is possible to derive a distribution pattern of concentration measurement values representing the actual pollutant concentration of the standard gas. possible (see Figure 3). In addition, for example, in the case of the pattern-comparison calculation module (see 300 in FIG. 1), the pattern forming module 100 provides the concentration measurement value distribution of the measurement data transmitted from the measurement module (see 200 in FIG. 1) Compared with the distribution patterns, it can be configured to learn the data with an artificial neural network algorithm that can learn the correlation between the distribution of the distribution patterns and the distribution of the distribution patterns of the measured data, and from this, the distribution with the highest similarity to the distribution of the measured values of the measured data It is possible to determine the pattern (see Fig. 5). In particular, the artificial neural network algorithm described above may include a convolutional neural network algorithm that can be effective in image or pattern processing, so that a distribution pattern can be derived through learning through a convolutional neural network algorithm, or a distribution pattern and concentration measurement value distribution of measurement data. It is possible to determine the degree of similarity by comparing . The convolutional neural network algorithm can process images or patterned data by, for example, extracting a feature map through a kind of operation matrix called a kernel or filter. It is possible to perform an operation that preserves the spatial information of . The convolutional neural network algorithm may include a convolution layer that performs these operations and may also include a pooling layer that converts the size of a feature map. The convolutional layer and the pooling layer may be formed on, for example, a hidden layer. Although not shown, one or a plurality of hidden layers may be formed, and the number of neurons may be expanded by increasing the hidden layer. In the case of having a plurality of hidden layers, for example, it can be composed of a deep neural network (DNN), from which machine-learning by deep learning proceeds, so that nonlinear correlations can be more appropriately identified. In addition, the configuration of synthetic neural networks and the like can be made more appropriate.

In this way, the learning algorithm of the pattern formation module 100 and the pattern-comparison calculation module 300 can be formed so that the operation by the artificial neural network algorithm is possible. However, it is not necessary to be limited in this way, and it is possible to selectively/additionally apply other artificial intelligence learning algorithms if necessary. For example, several other learning algorithms, such as a concept supervised learning algorithm that trains with data with answers, and an unsupervised learning algorithm with a concept training through classification through clustering of unanswered data, are selectively/additionally diverse. And these algorithms can be applied as artificial intelligence learning algorithms through deep learning, for example. The pattern formation module 100 and the pattern-comparison calculation module 300 may each include one or a plurality of calculation programs implemented by such an artificial intelligence learning algorithm, and in the case of the pattern-comparison calculation module 300, such calculations Learning from the program, determination of similarity, and calculation of actual values can be performed together. Or, if necessary, a calculation program for deriving correlation between data and determining similarity with an artificial intelligence learning algorithm, and another calculation program for calculating actual values by applying the results derived from the artificial intelligence learning algorithm may be included together. . The pattern formation module 100 and the pattern-comparison calculation module 300 can be configured by applying the artificial intelligence-based learning method in these various ways.

FIG. 3 is a view conceptually showing the process of deriving the distribution pattern shown in FIG. 1 and its contents.

Referring to FIG. 3, the process of deriving the distribution pattern by the pattern forming module (see 100 in FIG. 1) based on the artificial intelligence learning algorithm will be described in detail as follows. As shown in FIG. 3 , the above standard data and data of actual pollutant concentration values of the standard gas used in generating the standard data are learned in the pattern forming module 100, and learning is performed by an artificial intelligence learning algorithm accordingly. Learning may be by a machine-learning method. Accordingly, in the learning process based on artificial intelligence, the correlation between the distribution of the concentration measurement values of a plurality of different sensors shown in the standard data and the actual pollutant concentration values in the corresponding standard gas is comprehensively identified, and accordingly, the standard gas A distribution pattern of concentration measurements representing the actual contaminant concentration of That is, through artificial intelligence-based data learning, it is possible to derive the distribution of concentration measurement values of a plurality of sensors that match the actual pollutant concentration with a significantly high reliability, and the concentration measurement value distribution of such a plurality of sensors is one as shown in the figure. It can be assumed to represent the actual pollutant concentration by assuming a pattern of The distribution pattern of the concentration measurement values is, for example, as shown in FIG. 3, the concentration measurement values (concentration measurement values) that can be calculated by different sensors (sensor types 1 to 5) for different pollutants (pollutants a to e). A1 to E5), and such information is, for example, concentration measurements that can be calculated by a plurality of different sensors for a plurality of different pollutants in the form of a matrix (ie, spatial information may be listed). For example, one distribution pattern can be expressed as a matrix, and concentration measurement values (A1 to E5) that can be calculated by measuring with different sensors for different pollutants can be assigned to each component of the matrix. . Each concentration measurement value (A1 to E5) corresponding to a component of the matrix may correspond to a single number, and these numbers may have different values. Therefore, differences in the measurement results of different sensors for different pollutants can be identified within one distribution pattern, and the entire array of numbers showing these differences can form a characteristic pattern. Therefore, with such a distribution pattern, it is possible to represent the actual pollutant concentration value (and even the pollutant type if there are a plurality of pollutants) in the standard gas of the standard data learned when the distribution pattern is derived. That is, the distribution pattern is not a simple combination of measured data, and when standard data and standard data are created, information on actual pollutants in the standard gas is learned, and each sensor calculates a response pattern of multiple sensors when there are actual pollutants. It may be derived in the form of a distribution of possible concentration measurement values (concentration measurement values A1 to E5).

Illustratively, when the actual concentrations A to E, which are the actual concentration values of different pollutants in the standard gas in FIG. When viewing, the array of actual concentration values of these contaminants can be viewed as, for example, one row vector or column vector), for example, the concentration measurement values A1 to E5, each component of the derived distribution pattern, A5 goes to 4, 21, 21, 3, 7, second row B1 to B5 goes to 4, 20, 21, 11, 5, third row C1 to C5 goes to 3, 18, 31, 9, 2, fourth row D1 ~ D5 is 6, 7, 11, 51, 7, and the fifth row, E1 ~ E5, is combined with the value of 6, 6, 28, 50, 17, so that the whole can form a pattern by combination or arrangement of numbers. there is. The combination or arrangement of numbers in this distribution pattern is independent because it can be distinguished from other combinations or arrangements of numbers by itself, and therefore, the number of various actual pollutants is independent of another distribution pattern consisting of other combinations or arrangements of other numbers. Types and concentration values can be represented. Since these values are illustrative, it is not necessary to understand the contents of the present invention by limiting them to examples, but the distribution pattern can be more clearly understood through these examples.

In particular, as described above, standard data can be generated by changing the concentration while the type of pollutant in the standard gas is fixed, and multiple standard data can be accumulated for each concentration by repeating measurement for each fluctuating concentration. A situation in which these concentrations are present can be learned from a plurality of standard data, and one distribution pattern derived from such standard data can simultaneously represent the type and concentration of the corresponding specific pollutant. As described above, it is possible to accumulate data while changing the types and concentrations of pollutants in various ways when generating standard gas. By diversifying the generation method of standard data and increasing the amount of accumulated data, the types and concentrations of different pollutants can be calculated. A number of representative distribution patterns can be derived. Practically, the concentration of the standard gas can be subdivided and changed as desired, and it is possible to generate a plurality of distribution patterns representing various contaminant concentrations having correspondingly subdivided (concentration) intervals. In addition, since this process can be performed while changing the type of pollutant, a plurality of distribution patterns can be collected accordingly to represent various types of pollutants in the gas at the same time. As described above, the concentration and type of pollutants represented by the distribution patterns are the actual amounts directly recorded or calculated for the type and concentration of pollutants used in generating the standard gas. If the data are provided, it may be possible to determine the concentration and type of pollutants in the measured gas that generated the measured data virtually without error. However, even if they do not match completely, it is possible to find out information on pollutants in the measured gas with very high accuracy by applying a distribution pattern according to the degree of similarity. In this way, it is possible to derive and collect a distribution pattern of measured concentration values representing the actual concentration of pollutants in the standard gas (and even types of pollutants when there are a plurality of pollutants). Hereinafter, the measurement of the measured gas and the subsequent comparison process with the distribution pattern will be described in detail.

Referring back to FIG. 1 , the measurement module 200 generates measurement data and transmits it to the pattern-comparison calculation module 300 . The above-described distribution patterns derived and collected by the pattern forming module 100 may also be transmitted to the pattern-comparison calculation module 300 . The measurement module 200, the pattern formation module 100, and the pattern-comparison calculation module 300 may be disposed at the same location, but may be disposed at spatially different locations as needed. In particular, the measurement module 200 is disposed in the measurement area, measures the concentration of pollutants contained in the measurement gas with a plurality of different measurement sensors identical to those used for standard gas measurement, and generates and transmits measurement data. The actual measurement module 200 may be configured, for example, in the form of a combination of a plurality of measurement sensors, and, if necessary, a storage device for storing measurement data or a processing module capable of data processing (an operation program capable of data processing, etc.) It may be included), etc. It is also possible to include. In this case, the measurement area may be an area where a measurement gas requiring measurement of pollutants can be detected, and this may vary depending on the type of measurement gas. For example, when the flue gas of a power plant, etc. is a measurement gas, the measurement area may be around a corresponding facility, and the measurement module 200 may be disposed around the corresponding facility. However, since this is only one example, there is no need to be so limited, and if the measured gas is changed to another type, the measured area may also be changed. As described above, the standard gas may be artificially created using experimental equipment such as a standard gas generator (see 10 in FIG. 2), but the measurement gas may be directly emitted from pollutants that emit pollutants in the form of gas. Therefore, measurements can be made at points in close proximity to such contaminants. Measurement data for the measurement gas for which pollutant measurement is required may be generated by arranging the measurement module 200 at various points where measurement of the measurement gas is possible. The pattern forming module 100 and the pattern-comparison calculation module 300 may be disposed adjacent to the measurement module 200 or may be disposed at different locations apart from the measurement module 200, and the pattern-comparison with the pattern forming module 100 The calculation modules 300 may also be disposed adjacent to each other or disposed at different locations apart from each other. The pattern formation module 100, the measurement module 200, and the pattern-comparison calculation module 300 may be wired or wirelessly connected to each other to enable data transmission, and within such limits, mutual arrangements may be freely changed as needed. there is.

4 is a conceptual diagram illustrating a method of generating measurement data shown in FIG. 1 .

Referring to FIG. 4, the measurement data will be described in more detail. Measurement data is obtained from a plurality of measurement sensors (measurement sensor type1 to type n, where n is a natural number) as described above, and at this time, the measurement sensors are substantially the same as the sensors used for the above-mentioned standard gas measurement. It does not necessarily mean that the measurement target, measurement method, and precision match (e.g., the type or model of the device is identical), but it does not necessarily mean that the sensor used for standard gas measurement is reused as a measurement sensor, but the sensor used for standard gas measurement is necessary if necessary. It is not impossible to reuse as a measuring sensor, so such a possibility need not be ruled out}. That is, the concentration of pollutants contained in the measured gas is measured by a plurality of different measuring sensors identical to those used for measuring the standard gas, and the concentration measurement values of the pollutants in the measuring gas are measured by the plurality of different measuring sensors. Measurement data can be generated and transmitted. For example, each of the plurality of measurement sensors shown (measurement sensors type1 to type n, where n is a natural number) may be the same as each of the plurality of sensors described above (see sensor type1 to type n in FIG. 2), and thus each measurement It is also possible for the sensor to produce substantially the same result as each sensor described above, provided that the conditions are perfectly matched. Using this point, it is possible to directly compare the set of measurement data obtained from the measurement sensor with the distribution pattern obtained by learning the data measured by the above-described sensor. The plurality of measurement sensors of the measurement module 200 may also be made of a combination of sensors of different types that measure pollutants in the same way as the plurality of sensors described above, and the different types of sensors are also the metal oxide sensors described above. , A polymer sensor, an photoacoustic spectroscopy sensor, a surface plasmon resonance sensor, a light scattering particle measurement sensor, and a micro cantilever sensor. Since the description of the above-described sensors is applied exactly the same to the measurement sensors as it is, the detailed description thereof is replaced by the description of the above-described sensor. The measurement data has a substantial difference from the aforementioned standard data in that the measurement target is a measurement gas detected in a measurement area rather than a standard gas. Measurement data generated in this way can be transmitted to various points where actual measured values are calculated in the measurement area [eg, points where the pattern-comparison calculation module (see 300 in FIG. 1) is disposed], and the results calculated later are measured inversely. It can propagate to various points including regions.

Referring back to FIG. 1, the pattern-comparison calculation module 300 compares the concentration measurement value distribution of the measurement data transmitted from the measurement module 200 in this way with the distribution patterns provided by the pattern formation module 100. Calculate actual value by method. In particular, the pattern-comparison calculation module 300 compares the concentration measurement value distribution of the measurement data with distribution patterns, but learns and measures the correlation between the concentration measurement value distribution of the measurement data and the distribution pattern by the above-described artificial intelligence learning algorithm. The distribution pattern having the highest similarity with the distribution of the measured concentration values of the data is determined, and the pollutant concentration represented by the determined distribution pattern and the corresponding pollutant are calculated as actual values for the pollutant in the measured gas. That is, the distribution pattern derived from the analysis of the above-described standard gas and the data (measurement data) on the measurement gas that actually requires measurement are transmitted to the pattern-comparison calculation module 300 and compared with each other, and are compared with each other to the artificial intelligence learning algorithm. The distribution pattern with the highest similarity to the measured data is determined through data learning by At this time, by utilizing the convolutional neural network algorithm described above, the classification and comparison process for patterns can be performed more comprehensively and efficiently, and from this, a distribution having the highest degree of similarity with the concentration measurement value distribution of the measurement data transmitted from the measurement module 200 Patterns can be determined more effectively. Based on the above-described artificial intelligence learning algorithm, the actual value calculation process of the pattern-comparison calculation module 300 will be described in more detail as follows.

FIG. 5 is a diagram conceptually illustrating a process of comparing measurement data and distribution patterns of FIG. 4 and contents of the measurement data.

Referring to FIG. 5 , the measurement data may be a distribution of concentration measurement values (concentration measurement values a1 to e5) of a plurality of measurement sensors for contaminants in the measurement gas. Such measurement data can be arranged in a matrix-like form to enable easy comparison with, for example, the aforementioned distribution pattern. The concentration measurement values recorded in the measurement data are measured by different measurement sensors (measurement sensor types 1 to 5) for different contaminants (pollutants a to e) in the measurement gas, for example, as shown in FIG. 5. It can be composed of a combination of the calculated concentration measurement values (concentration measurement values a1 to e5), and such information can be arranged in the form of a matrix as shown and compared with the distribution pattern. In such a case, for example, concentration measurement values a1 to e5 assigned to each component of the measurement data matrix may correspond to a single number, and these numbers may have different values. Accordingly, concentration measurement values obtained by measuring the measurement gas of the measurement data may also form a measurement pattern consisting of an array or distribution of numbers. Therefore, similarity can be determined by comparing it with the collected distribution patterns. That is, the measurement data is a combination of concentration measurement values obtained by directly measuring the measurement gas by a plurality of measurement sensors, and can be compared by manipulation such as changing the arrangement in a form corresponding to the above-described distribution pattern. Such manipulation can be performed, for example, in a learning process of determining similarity by learning the correlation between measurement data and distribution patterns, and in such a way, similarity determination between measurement data and collected distribution patterns can be more easily accomplished. there is. For example, in the learning process, a group of distribution patterns having a relatively high similarity to the measured data is extracted through learning, and a distribution pattern having the highest similarity to the measured data is determined therefrom. Similarity can be defined in various ways, so it does not need to be particularly limited. As an example, when comparing information in a matrix form as shown, the similarity of the arrangement of components (and therefore each corresponding number), each It is possible to determine the overall degree of similarity by applying the similarity of values between components and the similarity of another value extracted from a matrix (for example, a value by a determinant) in a complex manner. However, since this is only an example, it is not necessary to be limited in this way, and the distribution pattern having the highest similarity to the measured data can be determined by setting the degree of similarity in a variety of possible ways.

Accordingly, for example, among a plurality of collected distribution patterns (distribution patterns 1 to n, where n is a natural number), the distribution pattern having the highest similarity to the measurement data transmitted by the measurement module (see 200 in FIG. 1) (e.g., distribution pattern 3) can be determined. In this way, when the distribution pattern having the highest degree of similarity is determined, the pattern-comparison calculation module (see 300 in FIG. 1) calculates the pollutant concentration represented by the determined distribution pattern and the corresponding pollutant as actual values for the pollutant in the measured gas do. That is, as shown in FIG. 1, the measured value calculated by the pattern-comparison calculation module 300 includes both the pollutant concentration and the corresponding type, which is the pollutant concentration represented by the distribution pattern determined to be most similar to the measured data and the pollutant concentration corresponding thereto. corresponding to the type of contaminant. In this way, it is possible to find out at once the types of pollutants contained in the measured gas and the concentration values of the pollutants through comparison between the measured data and the collected distribution patterns. The higher the degree of similarity, the more accurate the result can be derived. For example, if there is a distribution pattern that completely matches the measurement data, the reliability of the measurement result can increase dramatically. Since the degree of similarity does not need to be specified in one method as described above, the reliability of the measured value may be increased by changing the method of determining the degree of similarity in various ways. In particular, even if the determined distribution pattern and the measured data do not completely match each other, it can be seen that the distribution pattern reflects the characteristics of multiple sensors responding to one or more pollutants through learning, and the resulting error is eliminated in advance. An error of the measured value may be smaller than an error that appears by directly measuring the measured gas with each separate sensor. In this way, it is possible to obtain measurement results for pollutants in the measured gas, and in particular, it is possible to simultaneously find out the concentration value of each pollutant and the corresponding pollutant type for a plurality of pollutants in the measured gas. That is, as described above, the number of pollutants contained in the standard gas is plural, and therefore, the distribution pattern is composed of a distribution of measured concentration values representing the concentrations of a plurality of pollutants, and accordingly, the pattern-comparison calculation module 300 calculates The measured values may include both concentrations of a plurality of pollutants represented by the distribution pattern and a plurality of types of pollutants corresponding thereto. In this way, it is possible to measure pollutants more accurately and effectively.

Meanwhile, the pattern forming module 100 may additionally perform the following tasks together in the above-described learning process. For example, the pattern forming module 100 learns information about the error between the distribution of the concentration measurement value shown in the standard data and the actual pollutant concentration value of the standard gas, and a plurality of sensors used to measure the standard gas (FIG. 2 Among sensor types 1 to n in , where n is a natural number), a sensor with a measurement error exceeding the allowable range can be identified. This process can also proceed with learning by the above-described artificial intelligence learning algorithm, and thus, while deriving a distribution pattern in the learning process, the performance of a plurality of sensors that generate standard data can also be grasped. Thereafter, the pattern forming module 100 transmits, for example, an alarm signal to a sensor having a measurement error exceeding the allowable range among a plurality of sensors used for standard gas measurement and a plurality of measurement sensors of the measurement module 200 (FIG. 5 Among the measurement sensor types 1 to n of (where n is a natural number), replacement of a measurement sensor having a measurement error exceeding the allowable range may be indicated. That is, in the process of learning data, a sensor with an excessive measurement error is identified, and the corresponding sensor and the same measurement sensor are instructed to be replaced to obtain data (ie, standard data measured by a plurality of sensors for standard gas, and It is possible to improve the reliability of the measurement data measured for the measurement gas by the same plurality of measurement sensors. Therefore, if sensors and measurement sensors are changed through this process, the reliability of the entire data learning process, the derivation of a distribution pattern based on it, and the comparison of measurement data and distribution patterns, etc., can also be increased. It is also possible to increase the reliability of measurement results in this way.

7 is a flowchart illustrating a pollutant measurement method using artificial intelligence according to an embodiment of the present invention.

Hereinafter, based on the description of the pollutant measuring device using artificial intelligence as described above, the pollutant measuring method using artificial intelligence of the present invention (hereinafter, referred to as 'pollutant measuring method' has the same meaning) Explain in detail. The pollutant measuring method described below can be performed using the pollutant measuring device described above, and unless otherwise specified, the description of the configuration described in the pollutant measuring device is replaced with the above description and repeated description thereof. omit

Referring to FIG. 7 , the contaminant measuring method of the present invention may be composed of the following steps. First, (a) standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in a standard gas containing at least one type of pollutant, and actual pollutant concentrations in the standard gas used in generating the standard data The computer processes the data of the value, and the computer learns the correlation between the distribution of the concentration measurement value shown in the standard data and the actual pollutant concentration value of the standard gas by an artificial intelligence learning algorithm to determine the actual pollutant concentration of the standard gas. A distribution pattern of representative concentration measurement values is derived and collected (S100). Substantially, step (a) may be performed by the above-described pattern forming module (see 100 in FIG. 1), and therefore substantially the contents of step (a) are also standard data and standards by the above-described pattern forming module 100. It can be understood as equivalent to the process of deriving and collecting a distribution pattern through a learning process by an artificial intelligence learning algorithm in which data of actual pollutant concentration values of the standard gas used in data generation are provided. As described above, since the pattern forming module 100 can be implemented on a computer, the computer can smoothly perform these steps by mounting the above-described pattern forming module 100 or a program module capable of substantially equivalent actions thereto. . Content related to standard gas, content related to sensors, content related to standard data and actual pollutant concentration data of standard gas used when creating standard data, content related to artificial intelligence learning algorithm, and content related to distribution patterns are also provided. Since it is the same as the above, the detailed description thereof is replaced by the above description.

Thereafter, (b) in the measurement area, the concentration of pollutants contained in the measurement gas is measured with a plurality of different measurement sensors identical to those used in the measurement of the standard gas in step (a), and the pollutants in the measurement gas are measured in multiple ways. Measurement data consisting of concentration measurement values measured by different measurement sensors is generated and transmitted to a computer (S200). Substantially, step (b) may be performed by the above-described measurement module (see 200 in FIG. 1 ), and therefore substantially the contents of step (b) also measure the measurement gas with the above-described measurement module 200, It can be understood as equivalent to the process of generating and transmitting measurement data. As described above, since the measurement module 200 may include a plurality of different measurement sensors identical to those used for measuring the standard gas, the above-described measurement module 200 or sensor equipment substantially equivalent thereto, etc. Step (b) can be performed using The computer to which measurement data is transmitted in step (b) may be a computer equipped with a pattern-forming module (see 300 in FIG. 1) or a program module capable of substantially equivalent operations thereto. Since the contents related to the measurement sensor and the contents related to the measurement data are substantially the same as those described above, the detailed description thereof is also replaced by the above description.

Then, (c) the computer compares the concentration measurement value distribution of the measurement data transmitted in step (b) with the collected distribution patterns, and the artificial intelligence learning algorithm compares the correlation between the concentration measurement value distribution of the measurement data and the distribution pattern. Determines the distribution pattern that has the highest similarity with the concentration measurement value distribution of the measured data by learning by S300). Step (c) may also be substantially performed by the above-described pattern-comparison calculation module (see 300 in FIG. 1), and therefore substantially the contents of step (c) also include the pattern-comparison calculation module 300 described above. The learning process by the learning algorithm compares the concentration measurement value distribution of the measurement data with the collected distribution patterns, determines the distribution pattern with the highest similarity to the concentration measurement value distribution of the measurement data, and determines the pollutant represented by the determined distribution pattern. It can be understood as equivalent to the process of calculating the concentration and the pollutant corresponding thereto as an actual value for the pollutant in the measured gas. As described above, since the pattern-comparison calculation module 300 can also be implemented on a computer, the computer can smoothly perform these steps by loading the pattern-comparison calculation module 300 or a program module capable of substantially equivalent operations thereto. can Since the contents of the artificial intelligence learning algorithm, the process of comparing measurement data and distribution patterns, and the process of calculating actual values with the distribution patterns determined accordingly are all substantially the same as the above, detailed descriptions thereof are also replaced by the above description. .

Accordingly, the artificial intelligence learning algorithm of steps (a) and (c) in the pollutant measurement method of the present invention is an artificial neural network in which at least one hidden layer consisting of a plurality of calculation nodes is formed between the above-described input layer and output layer. network) algorithm, and the artificial neural network algorithm may include a convolutional neural network algorithm. In addition, the number of pollutants contained in the standard gas in step (a) is plural, and the distribution pattern consists of a distribution of measured concentration values representing the concentrations of a plurality of pollutants, and the measured values calculated in step (c) are the distribution pattern A plurality of pollutant concentrations represented by , and a plurality of pollutant types corresponding thereto may be included. Since these contents are also equivalent to the contents described above, the detailed description is replaced with the above description.

The computer of the pollutant measurement method may be as follows. The implementation method or form of the computer does not need to be specifically limited. For example, steps (a) and (c) of the method for measuring pollutants may be performed on a single physically integrated computer or on separate and independent computers. [Even in step (b), if data processing using a computer is required in addition to measurement of the measuring sensor, a computer can be involved] Computers can be various types of devices including a central processing unit (CPU) capable of processing and having a memory capable of embedding or storing them. Therefore, it is not necessary to be limited by the name of a computer, and for example, a control device such as a PLC (Programmed Logic Controller) may actually fall within the scope of the computer of the present invention. The pollutant measuring method of the present invention can be performed using such a computer device, and the above-described pollutant measuring device can be constructed using such a computer device equivalently.

In this method of measuring pollutants, in step (a), the computer learns information about the error between the distribution of concentration values shown in the standard data and the actual pollutant concentration value of the standard gas, It may include a process of discriminating a sensor having a measurement error exceeding an allowable range among sensors. That is, as described above, in the learning process of deriving the distribution pattern, information on data errors can be learned together, and through this, sensors with excessive measurement errors can be identified. As described above, the reliability of the overall pollutant measurement process can be improved by removing the sensors having excessive measurement errors and equivalent measurement sensors as necessary. As such, pollutants in the measurement gas can be more accurately and effectively measured using the method for measuring pollutants of the present invention.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art can realize that the present invention can be implemented in other specific forms without changing the technical spirit or essential features. you will be able to understand Therefore, the embodiments described above should be understood as illustrative in all respects and not limiting.

1: Pollutant measuring device using artificial intelligence
10: standard gas generator
20: data storage unit
100: pattern forming module
200: measurement module
300: pattern-comparison calculation module

Claims (11)

Standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in a standard gas containing at least one type of pollutant, and actual pollutant concentration values of the standard gas used in generating the standard data Data is received, and the correlation between the distribution of concentration measurements shown in the standard data and the actual pollutant concentration value of the standard gas is learned by an artificial intelligence learning algorithm to measure the concentration representing the actual pollutant concentration of the standard gas a pattern formation module for deriving and collecting a distribution pattern of values;
Arranged in the measurement area, a plurality of different measurement sensors identical to the sensors used for measuring the standard gas measure the concentration of pollutants contained in the measurement gas, and a plurality of different measurement sensors for pollutants in the measurement gas. a measurement module for generating and transmitting measurement data consisting of measured concentration values; and
The concentration measurement value distribution of the measurement data transmitted from the measurement module is compared with the distribution patterns provided by the pattern forming module, and the correlation between the concentration measurement value distribution of the measurement data and the distribution pattern is compared with the artificial intelligence learning algorithm. to determine the distribution pattern having the highest degree of similarity with the concentration measurement value distribution of the measurement data by learning by means of, and determining the pollutant concentration represented by the determined distribution pattern and the pollutant corresponding thereto as an actual value of the pollutant in the measured gas Including a pattern-comparison calculation module calculated by
One distribution pattern consists of an array of numbers corresponding to the concentration measurement values measured by the plurality of different sensors and is distinguished from other distribution patterns, and consists of an array of numbers corresponding to the concentration measurement values of the measurement data. A pollutant measuring device using artificial intelligence, characterized in that the measurement pattern and information are compared in the form of a matrix. According to claim 1,
The plurality of sensors used for measuring the standard gas and the plurality of measurement sensors of the measurement module are a pollutant measuring device using artificial intelligence consisting of a combination of sensors of different types with different measuring methods of pollutants. According to claim 2,
The different types of sensors include at least two selected from metal oxide sensors, polymer sensors, photoacoustic spectroscopy sensors, surface plasmon resonance sensors, light scattering particle measurement sensors, and microcantilever sensors. Contaminant measuring device using artificial intelligence. According to claim 1,
The artificial intelligence learning algorithm of the pattern forming module and the pattern-comparison calculation module includes an artificial neural network algorithm in which at least one hidden layer consisting of a plurality of calculation nodes is formed between an input layer and an output layer, and the artificial The neural network algorithm is a pollutant measuring device using artificial intelligence including a convolutional neural network algorithm. According to claim 1,
The number of pollutants contained in the standard gas is plural, the distribution pattern is composed of a distribution of measured concentration values representing concentrations of a plurality of pollutants, and the measured value calculated by the pattern-comparison calculation module is represented by the distribution pattern. A pollutant measuring device using artificial intelligence that includes a plurality of pollutant concentrations and a plurality of corresponding pollutant types. According to claim 1,
The pattern forming module learns information about the error between the distribution of the measured concentration value shown in the standard data and the actual pollutant concentration value of the standard gas, and sets an allowable range among the plurality of sensors used for measuring the standard gas. A pollutant measuring device using artificial intelligence to determine a sensor with a measurement error that exceeds. According to claim 6,
The pattern forming module transmits an alarm signal to detect a sensor having a measurement error exceeding the allowable range among the plurality of sensors used for measuring the standard gas and the same tolerance range among the plurality of measurement sensors of the measurement module. A pollutant measuring device using artificial intelligence that instructs replacement of a measuring sensor with a measurement error that exceeds. (a) Standard data consisting of concentration measurement values measured by a plurality of different sensors for pollutants in standard gas containing at least one type of pollutant, and actual pollutants in the standard gas used in generating the standard data The computer processes the data of the concentration value, and the computer learns the correlation between the distribution of the concentration measurement value shown in the standard data and the actual pollutant concentration value of the standard gas by an artificial intelligence learning algorithm, deriving and collecting a distribution pattern of concentration measurement values representing pollutant concentrations;
(b) In the measurement area, the concentration of pollutants contained in the measurement gas is measured with a plurality of different measurement sensors identical to those used for the measurement of the standard gas in step (a), and the pollutants in the measurement gas are measured. Generating measurement data consisting of concentration measurement values measured by a plurality of different measurement sensors and transmitting the same to a computer; and
(c) the computer compares the concentration measurement value distribution of the measurement data transmitted in step (b) with the collected distribution patterns, and artificially detects the correlation between the concentration measurement value distribution of the measurement data and the distribution pattern. It is learned by an intelligent learning algorithm to determine the distribution pattern having the highest similarity with the concentration measurement value distribution of the measurement data, and pollutant concentrations represented by the determined distribution pattern and pollutants corresponding thereto are determined as pollutants in the measured gas. Including the step of calculating the actual value for
One of the distribution patterns collected in the step (a) is distinguished from other distribution patterns by an array of numbers corresponding to the concentration measurement values measured by the plurality of different sensors, and the measurement in the step (c) A pollutant measurement method using artificial intelligence, characterized in that the measurement pattern consisting of an array of numbers corresponding to the concentration measurement values of the data and the information in the form of a matrix are compared. According to claim 8,
The artificial intelligence learning algorithm of steps (a) and (c) includes an artificial neural network algorithm in which at least one hidden layer consisting of a plurality of calculation nodes is formed between the input layer and the output layer, and the artificial neural network The algorithm is a pollutant measurement method using artificial intelligence including a convolutional neural network algorithm. According to claim 8,
The number of contaminants contained in the standard gas in step (a) is plural, and the distribution pattern is composed of a distribution of measured concentration values representing concentrations of a plurality of contaminants, and the measured values calculated in step (c) are: A pollutant measurement method using artificial intelligence including a plurality of pollutant concentrations represented by the distribution pattern and a plurality of pollutant types corresponding thereto. According to claim 8,
In the step (a), the computer learns information about the error between the distribution of the concentration measurement value shown in the standard data and the actual pollutant concentration value of the standard gas, and among the plurality of sensors used to measure the standard gas A method for measuring pollutants using artificial intelligence, including the process of determining a sensor with a measurement error exceeding the allowable range.

Images