KR102279351B1 - A fault diagnosis method of based ai on optimization of air conditioner - Google Patents

Abstract

The present invention has the following method comprising: a step of normalizing by collecting sensing data for a normal type according to an operation of an operating air conditioner installed in the site from a database of a control server; a step of correcting the normal type data of the operating air conditioner and constructing a normal type regression model; a step of detecting a fault candidate of the operating air conditioner through the learning model determining part based on the normal type regression model of the operating air conditioner; and a step of diagnosing a specific fault type in the control server for the detected fault candidate of the operating air conditioner. Therefore, the present invention is capable of having an effect of allowing a standard air conditioner to virtually function as the operating air conditioner.

Description

A FAULT DIAGNOSIS METHOD OF BASED AI ON OPTIMIZATION OF AIR CONDITIONER method based on artificial intelligence through the adaptation of standard air conditioners and operating air conditioners

The present invention relates to an AI-based failure diagnosis method through the adaptation of a standard air conditioner and an operating air conditioner, and more particularly, even if a lot of data required for machine learning learning about the operating state of each air conditioner operating in a building site is not accumulated, the standard Set up a simulation model of an air conditioner that can play a critical role, and automatically generate a lot of data about normal and fault conditions, based on this, build a machine learning learning model in advance, and operate the operating air conditioner collected at the building site. By correcting the state data to be suitable for the operation state data of the standard air conditioner, the standard air conditioner can play the role of the operational air conditioner virtually, and machine learning learning of the operational air conditioner that is suitable for the operation state data of the operational air conditioner that requires fault diagnosis By applying to the model and detecting and diagnosing the failure of the operating air conditioner, it is possible to accurately detect and diagnose the failure from the early stage of operation of the air conditioner, and to reduce the discomfort and economic loss of heating and cooling that may occur due to improper failure of the air conditioner. It relates to an AI-based fault diagnosis method through the adaptation of an operating air conditioner.

In general, in the air conditioning system in a building, the scale of automation and facility systems is getting larger and more complex, and accordingly, the occurrence of a failure in the system and the prompt identification of the cause have not been made. The purpose of monitoring and diagnosing the failure of the air conditioning system using a computer is to accurately detect and diagnose the cause and location of the failure because the failure of one part of the air conditioning system has the characteristic that it can affect many other places. In particular, it is necessary to provide information to the operator by quickly and accurately diagnosing the failure of the Air Handing Unit (AHU), which has the most complex structure and function among the facilities of the air conditioning system. .

However, since there are many cases where the user or manager directly observes the air conditioner and diagnoses the failure, rather than the automated method to diagnose the failure of the air conditioner, the accuracy of failure diagnosis is lowered, and the failure detection time is delayed, so the optimization for failure There was a problem that could not be addressed.

While solving these problems, research on a method for diagnosing an air conditioner failure that can easily diagnose the operating state of the air conditioner and accurately detect a failure has been conducted.

Prior Patent 1 presents a new air conditioner failure diagnosis method in order to solve the problems of the existing air conditioner failure diagnosis method. This method constructs an air conditioner model in the form of a data-based learning model, and includes a plurality of sensor detection signals and air conditioners. After constructing a failure model corresponding to the error of each sensor based on the estimated control signal provided from the model, the failure is detected using the air conditioner model and the failure model.

However, in Prior Patent 1, it takes a lot of time and effort to construct an air conditioner model for each air conditioner installed in a building site, and it is applied from the early stage of operation of the air conditioner because failure data can be learned only when various failure types are accumulated for each air conditioner. it is impossible to do

Prior Patent 2 provides a method for defining a reference model equation between measured values of the normal state sensor of the air conditioner and performing fault detection and diagnosis using the reference model equation. This method is independent of the measured values of the normal state of the air conditioner. The residual is calculated by comparing the value of the dependent variable calculated by substituting the value classified as a variable into the independent variable of the standard model formula and the value classified as the dependent variable among the measured values. If the residual is greater than a certain size, a failure is diagnosed.

However, in Prior Patent 2, it is difficult to apply from the initial operation of the air conditioner because it takes a lot of time and effort because it takes a lot of time and effort to construct a reference model by collecting the measured values of the air conditioner steady state for each number of air conditioners installed at the building site. There is a problem in that it is difficult to accurately detect a failure because it diagnoses whether there is a failure.

Prior Patent 1: Korean Patent Publication No. 10-1127941 (2012.03.12.) Prior Patent 2: Korean Patent Publication No. 10-1151867 (2012.05.24.)

The present invention establishes a simulation model of an air conditioner that can play a standard role without accumulating a lot of data required for machine learning learning about the operating state for each air conditioner operating at a building site in order to detect and diagnose a failure of the air conditioner, and set a normal state It automatically generates a lot of data about the operating conditions and fault conditions, and based on this, builds a machine learning learning model in advance, and corrects the operating condition data of the operating air conditioner that requires fault diagnosis to be suitable for the operating condition data of the standard air conditioner. After building the machine learning model of the air conditioner, the corrected operating condition data is applied to the machine learning learning model of the operating air conditioner to detect and diagnose the failure of the operating air conditioner, thereby accurately detecting and diagnosing the failure from the beginning of the operation of the air conditioner. An object of the present invention is to provide a method that can reduce the discomfort and economic loss of heating and cooling that may occur due to improper failure of the air conditioner.

The artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention for achieving the above object is a simulation model calculated by the control server by inputting the equipment specifications and operating conditions for the standard air conditioner into the control server Building a machine learning learning model for the driving state of the standard air conditioner based on the data generated through the standard air conditioner learning model building unit; Normalizing by collecting the sensing data for the normal type according to the operation of the operation air conditioner installed in the field from the database of the control server; Correcting the normal type data of the operating air conditioner by applying the normalized sensing data from the control server to the normal type regression model of the standard air conditioner built in the standard air conditioner learning model building unit and constructing a normal type regression model; detecting a failure candidate of the operating air conditioner through a learning model determiner based on the normal type regression model of the operating air conditioner; and diagnosing a specific failure type in the control server for the detected failure candidates of the operating air conditioner.

The control server, the communication unit for receiving the driving state sensing information of the operating air conditioner; A standard air conditioner learning model building unit that learns and verifies machine learning on normal and failure types of standard air conditioners; an operation air conditioner correction unit for collecting sensing data for the normal type of the operation air conditioner and correcting the error; an operation air conditioner failure detection unit that finds a failure candidate for the operation state of the operation air conditioner; an operation air conditioner failure type diagnosis unit for diagnosing a specific failure type with respect to the detected failure candidates of the operation air conditioner; It is characterized in that it comprises a learning model determining unit that performs the machine learning learning and verification.

The step of constructing a machine learning learning model for the driving state of the standard air conditioner through the standard air conditioner learning model building unit is the step of setting a simulation model for the driving state of the standard air conditioner in the control server, the control server through the learning model determining unit In the step of automatically generating the driving condition data of the standard air conditioner, normalizing the generated operating condition data of the standard air conditioner in the control server, building a normal type machine learning regression model of the standard air conditioner through the learning model determiner, learning It characterized in that it comprises the step of building a machine learning classification model for each failure type of the standard air conditioner through the model determination unit.

Collecting and normalizing the sensing data for the normal type according to the operation of the operating air conditioner from the database of the control server may include: collecting normal type data according to the operating state of the operating air conditioner; It characterized in that it comprises the step of normalizing the normal type data of the operating air conditioner in the control server.

The step of correcting the normal type data of the operating air conditioner and building a normal type regression model includes applying the normal type data of the operating air conditioner to the normal type regression model of the standard air conditioner in the standard air conditioner learning model building unit to determine the amount of supply fan power. predicting; determining an optimal polynomial regression model between a normal type measured air supply fan power amount and a predicted air supply fan power amount of the operating air conditioner through the learning model determiner; calculating a normal type corrected air supply fan power amount by applying the actual air supply fan power amount of the operating air conditioner to the optimal polynomial regression model; constructing a normal type regression model of the operating air conditioner by generating normal type correction data of the operating air conditioner; and measuring the similarity of distribution between the normal type of the operating air conditioner corrected supply air power amount and the estimated air supply fan power amount by applying the operating air conditioner normal type correction data to the built-up normal type regression model of the operating air conditioner.

Detecting a failure candidate of the operating air conditioner through the learning model determiner based on the normal type regression model of the operating air conditioner may include: collecting and normalizing sensing data for which failure diagnosis of the operating air conditioner is required; calculating the corrected power supply fan power amount for fault diagnosis by applying the actual air supply fan power amount of the operating air conditioner to the optimal polynomial regression model; generating correction data for fault diagnosis of operating air conditioners by adding an amount of power supply fan for fault diagnosis; calculating a similarity index by applying correction data for diagnosing a failure of the operating air conditioner to the normal type regression model of the operating air conditioner; and detecting a failure candidate based on the calculated similarity index.

The step of diagnosing a specific failure type in the control server with respect to the detected failure candidate of the operating air conditioner may include, when the failure candidate of the operation air conditioner is detected, machine learning classification for each failure type of the standard air conditioner in the standard air conditioner learning model building unit calculating classification accuracy by applying correction data for diagnosing operating air conditioner failures to the model; first diagnosing a failure type based on the calculated classification accuracy; Secondary diagnosis of the failure type by additionally executing machine learning classification learning if the primary failure type diagnosis is deferred; and taking temporary measures so that the operating performance is not the lowest according to the diagnosed failure type.

The learning model determining unit of the control server further comprises a machine learning engine for performing the machine learning training training and prediction, wherein the machine learning engine is a random forest regression model that is an ensemble model of machine learning. , random forest classification model, polynomial regression model, and a learning model built according to the driving condition of standard air conditioners by executing the CTGAN algorithm in the artificial intelligence deep learning field that automatically generates tabular data. It features machine learning learning to accurately diagnose failures by applying the operating state data of air conditioners operated at the building site to the building site.

The present invention establishes a simulation model of an air conditioner that can play a standard role without accumulating operating state data for each air conditioner operating at a building site in order to detect and diagnose a failure of the air conditioner, and learn machine learning for the normal state and the failure state There is an effect that can be provided at the same time when a building automation system (BAS) is installed by building a model in advance.

In addition, by correcting the operation state data of the operating air conditioner collected at the building site to be suitable for the operation state data of the standard air conditioner, there is an effect that the standard air conditioner can play the role of the operating air conditioner virtually.

In addition, by applying the operating condition data of the operating air conditioner that requires fault diagnosis to the machine learning learning model of the operating air conditioner that is suitable for detecting and diagnosing the failure of the operating air conditioner, it is possible to accurately detect and diagnose the failure from the early stage of operation of the air conditioner. , there is an effect that can reduce the discomfort of heating and cooling and economic loss that may occur due to improper failure of the air conditioner.

1 is a view showing an overall sequence of an AI-based fault diagnosis method through the adaptation of a standard air conditioner and an operating air conditioner according to an embodiment of the present invention.
2 is a view showing the configuration of a control server and peripherals of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.
3 is a view showing in detail the entire operation process of the standard air conditioner of the AI-based fault diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.
4 is a step of constructing a machine learning learning model for the operating state of a standard air conditioner of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention in detail through the standard air conditioner learning model building unit. It is a drawing.
5 is a diagram showing in detail the step of collecting and normalizing the sensing data for the normal type according to the operation of the operation air conditioner of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention in detail. .
6 is a view showing in detail the steps of correcting the normal type data of the operating air conditioner and constructing the normal type regression model of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.
7 is a detailed step of detecting a failure candidate of the operating air conditioner through the learning model determiner based on the normal type regression model of the operating air conditioner of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention. It is a drawing that is clearly shown.
8 is a view showing in detail the step of diagnosing a specific failure type for a detected failure candidate of the operation air conditioner of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention.
9 is a diagram illustrating the classification of failure types in the standard air conditioner of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.
10 is a diagram showing the result of detecting a failure candidate by applying the operating air conditioner operation state data to the standard air conditioner normal type machine learning regression model of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention to be.
11 is a diagram showing the results of first diagnosing the failure by applying the data for diagnosing the operation air conditioner failure of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention to the machine learning classification model of the standard air conditioner. to be.
12 is a second failure by applying the data for the operation air conditioner failure diagnosis reserved for the primary failure diagnosis of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention to the machine learning classification model of the standard air conditioner. A diagram showing the results of diagnosis.

Hereinafter, specific embodiments for carrying out the present invention will be described with reference to the drawings. The embodiment of the present invention is intended to explain one invention, and the scope of rights is not limited to the illustrated embodiment, and the illustrated drawings are limited to the drawings because only the essential content is enlarged and illustrated for the clarity of the invention and incidental elements are omitted. should not be interpreted as such.

The present invention establishes a machine learning learning model for the operating state of a standard air conditioner based on the data generated through the simulation model calculated by the control server by inputting the equipment specifications and operating conditions for the standard air conditioner into the control server. Building through the unit 40 (S10); Collecting sensing data for a normal type according to the operation of the operation air conditioner 110 installed in the field from the database 30 of the control server 10 and normalizing it in the control server 10 (S20); Compensating the normal type data of the operating air conditioner 110 for the normalized sensing data in the control server 10 and constructing a normal type regression model (S30); detecting a failure candidate of the operating air conditioner through the learning model determiner 80 based on the normal type regression model of the operating air conditioner 110 (S40); and diagnosing a specific failure type in the control server 10 with respect to the detected failure candidate of the operation air conditioner 110 (S50).

The control server 10, the communication unit 20 for receiving the driving state sensing information of the operating air conditioner (110); A standard air conditioner learning model building unit 40 for learning and verifying machine learning for the normal type and failure type of the standard air conditioner; an operation air conditioner correction unit 50 for collecting sensing data for a normal type of the operation air conditioner 110 and correcting the error; an operation air conditioner failure detection unit 60 for finding a failure candidate for the operation state of the operation air conditioner 110; an operation air conditioner failure type diagnosis unit 70 for diagnosing specific failure types with respect to the detected failure candidates of the operation air conditioner 110; It is characterized in that it comprises a learning model determining unit 80 that performs the machine learning learning and verification.

In the step (S10) of building a machine learning learning model for the driving state of the standard air conditioner through the standard air conditioner learning model building unit 40, a simulation model for the driving state of the standard air conditioner is set in the control server 10 step (S11), automatically generating the driving state data of the standard air conditioner in the control server 10 through the learning model determining unit 80 (S12), normalizing the generated driving state data of the standard air conditioner in the control server Step (S13), building a normal type machine learning regression model of the standard air conditioner through the learning model determining unit 80 (S14), classifying the machine learning by failure type of the standard air conditioner through the learning model determining unit 80 It characterized in that it comprises a step (S15) of building a model.

The step (S20) of collecting and normalizing sensing data for a normal type according to the operation of the operation air conditioner 110 from the database 50 of the control server 10 is normal according to the operation state of the operation air conditioner 110. collecting type data (S21); Normalizing the normal type data of the operation air conditioner 110 in the control server 10 (S22) characterized in that it comprises a.

The step (S30) of correcting the normal type data of the operating air conditioner and building a normal type regression model is the operation of the operating air conditioner 110 in the normal type regression model of the standard air conditioner in the standard air conditioner learning model building unit 40. Predicting the amount of power supply fan by applying the normal type data (S31); determining an optimal polynomial regression model between the normal type measured air supply fan power amount and the predicted air supply fan power amount of the operating air conditioner 110 through the learning model determining unit 80 (S32); calculating the normal type corrected air supply fan power amount by applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model (S33); generating a normal type correction data for the operating air conditioner 110 and constructing a normal type regression model of the operating air conditioner 110 (S34); By applying the normal type correction data of the operating air conditioner 110 to the regression model for the normal type of the constructed operating air conditioner 110, the distribution similarity of the operating air conditioner 110 normal type corrected supply fan power amount and the estimated supply air fan power amount is measured. It characterized in that it includes a step (S35).

The step of detecting a failure candidate of the operating air conditioner 110 through the learning model determining unit 80 (S40) is a step of collecting and normalizing the sensing data for which the failure diagnosis of the operating air conditioner 110 is required (S41) ); calculating the corrected power supply fan power amount for fault diagnosis by applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model (S42); generating correction data for fault diagnosis of the operating air conditioner 110 by adding a corrected supply fan power amount for fault diagnosis (S43); calculating a similarity index by applying correction data for fault diagnosis of the operating air conditioner 110 to the normal type regression model of the operating air conditioner 110 (S44); and detecting a failure candidate based on the calculated similarity index (S45).

The step of diagnosing a specific failure type with respect to the detected failure candidate of the operating air conditioner 110 (S50) is, when the operation air conditioner 110 is detected as a failure candidate, the standard air conditioner learning model building unit 40 Calculating classification accuracy by applying the correction data for the operation air conditioner 110 failure diagnosis to the machine learning classification model for each failure type (S51); first diagnosing a failure type based on the calculated classification accuracy (S52); Secondary diagnosis of the failure type by additionally executing machine learning classification learning if the primary failure type diagnosis is withheld (S53); It characterized in that it comprises a step (S54) of taking temporary measures so that the operating performance is not the lowest according to the diagnosed failure type.

The learning model determining unit 80 of the control server 10 further includes a machine learning engine 81 for performing the machine learning learning training and prediction, and the machine learning engine 81 is an ensemble of machine learning. (Ensemble) Random Forest Regression Model, Random Forest Classification Model, Polynomial Regression Model, and CTGAN Algorithm in Artificial Intelligence Deep Learning that automatically generates tabular data Thus, it is characterized in that machine learning is performed to accurately diagnose failure by applying the driving state data of the air conditioner operated at the building site to the learning model built according to the operating state of the standard air conditioner.

1 is a view showing the overall sequence of an AI-based fault diagnosis method through the adaptation of a standard air conditioner and an operating air conditioner according to an embodiment of the present invention, and the equipment specifications and operating conditions for the standard air conditioner are provided by the control server 10 A machine learning learning model consisting of a normal type machine learning regression model for the driving state of a standard air conditioner and a machine learning classification model for each failure type based on the data generated through the simulation model calculated by the control server 10 by input to the standard air conditioner Building through the learning model building unit 40 (S10); Collecting sensing data for a normal type according to the operation of the operation air conditioner 110 installed in the field from the database 30 of the control server 10 and normalizing it in the control server 10 (S20); A polynomial regression model between the normal type measured air supply fan power amount and the predicted air supply fan power amount of the operating air conditioner 110, which is the normal type data of the standard air conditioner, through the learning model determining unit 80, using the normalized sensing data in the control server 10 and , determining and building a normal type regression model of the operating air conditioner 110 by generating normal type correction data of the operating air conditioner 110 and correcting the normal type data of the operating air conditioner 110 based on this (S30); Detecting a failure candidate of the operating air conditioner 110 through the learning model determiner 80 based on the machine learning learning model built in the standard air conditioner learning model building unit 40 (S40); and diagnosing a specific failure type in the control server 10 with respect to the detected failure candidate of the operation air conditioner 110 (S50).

2 is a view showing the control server 10 and peripheral configuration of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.

3 is a view showing in detail the entire operation process of the standard air conditioner of the AI-based fault diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.

4 is a view showing in detail the step of constructing a machine learning learning model for the operating state of the standard air conditioner of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.

5 is a view showing in detail the step of collecting and normalizing the sensing data for the normal type of the operating air conditioner 110 of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.

6 is a diagram showing in detail the steps of correcting the normal type data of the operating air conditioner 110 and constructing a normal type regression model of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention. .

7 is a failure candidate of the operating air conditioner 110 through the learning model determiner based on the normal type regression model of the operating air conditioner 110 of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention. It is a diagram showing the step of detecting in detail.

8 is a diagram showing in detail the step of diagnosing a specific failure type with respect to the detected failure candidate of the operation air conditioner 110 of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention.

9 is a diagram illustrating the classification of failure types in the standard air conditioner of the AI-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention.

10 is a result of detecting failure candidates by applying the operation state data of the operating air conditioner 110 to the standard air conditioner steady state machine learning regression model of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner according to the present invention. is a diagram showing

11 is a result of first diagnosing the failure by applying the failure diagnosis data for the operation air conditioner 110 of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention to the machine learning classification model of the standard air conditioner. is a diagram showing

12 shows the failure diagnosis data of the operation air conditioner 110 for which the primary failure diagnosis of the artificial intelligence-based failure diagnosis method through the adaptation of the standard air conditioner and the operation air conditioner according to the present invention is reserved to the machine learning classification model of the standard air conditioner. It is a diagram showing the results of the secondary fault diagnosis.

1 to 2, the step (S10) of building a machine learning learning model for the operating state of the standard air conditioner through the standard air conditioner learning model building unit 40 (S10) is the equipment specifications and operating conditions for the standard air conditioner. Machine learning consisting of a normal type machine learning regression model and a machine learning classification model for each type of failure based on data generated through a simulation model that is input to the control server 10 and calculated from the control server 10 The learning model is built through the standard air conditioner learning model building unit 40 .

At this time, the control server 10, the communication unit 20 for receiving the driving state sensing information of the operating air conditioner 110, the standard air conditioner learning model for learning and verifying machine learning for the normal type and the failure type of the standard air conditioner The unit 40, the operation air conditioner correction unit 50 that collects sensing data for the normal type of the operation air conditioner 110 and corrects the error, and the operation that finds a failure candidate for the operating state of the operation air conditioner 110 Air conditioner failure detection unit 60, operation air conditioner failure type diagnosis unit 70 for diagnosing specific failure types for the detected failure candidates of the operation air conditioner 110, learning model determining unit 80 for machine learning learning and verification includes

The control server 10 operates the actual standard air conditioner through the sensing data transmitted from the building automatic control system (BAS), which is the control unit 100 for controlling the operation air conditioner 110 , and the machine learning model of the standard air conditioner 110 . The state value is calculated, and the control unit 100 of the operation air conditioner 110 is controlled through this.

The communication unit 20 is connected to the control server 10 and connected to the control unit 100 for controlling the operation air conditioner 110 to transmit a control signal of the operation air conditioner 110 to the control unit 100, and the control unit 100 The operating state of the operating air conditioner 110 transmitted through the and sensing data of sensors formed in the operating air conditioner 110 are transmitted to the control server 10 .

In this case, the operating air conditioner 110 refers to an air conditioner that is actually installed and operated in the field, and the standard air conditioner is a configuration that serves as a reference for the operation of the operating air conditioner 110, and serves as a standard for the operating conditions of conventional operating air conditioners. By setting a simulation model so as to be able to do so, by building a machine learning learning model through the standard air conditioner learning model building unit 40 for the normal state and the failure state, the role of the operating air conditioner 110 in the control server 10 can be virtual. It is an air conditioner set up to do this.

The standard air conditioner learning model building unit 40 builds a machine learning learning model for the normal type and failure type of the standard air conditioner based on the normal type and failure type data of the standard air conditioner.

The operating air conditioner correction unit 50 collects sensing data of the operating air conditioner 110 for the normal type of the operating air conditioner 110 and compares the error correction with the operating state value of the standard air conditioner.

The operation air conditioner failure detection unit 60 detects a failure candidate for the operation state of the operation air conditioner 110 by comparing the state of the standard air conditioner compared in the control server 10 with a set fixed type reference value.

The operation air conditioner failure type diagnosis unit 70 determines specific failure types for the detected failure candidates of the operation air conditioner 110 by failure type built through the standard air conditioner learning model building unit 40 and the learning model determining unit 80 . It is diagnosed through a machine learning classification model.

The database 30 stores the machine learning simulation model data of the standard air conditioner calculated through the control server 10 and the sensing data and the machine learning data according to the operation of the operating air conditioner 110 .

The learning model determining unit 80 performs machine learning learning performed in the control server 10 and verification thereof.

And, the learning model determining unit 80 further includes a machine learning engine 81 for performing machine learning training training and prediction, and the machine learning engine 81 is an ensemble of machine learning. Random forest regression model, random forest classification model, polynomial regression model, artificial intelligence that automatically generates tabular data by executing the CTGAN algorithm in the field of deep learning. Machine learning is applied to the learning model built according to the operating state of the air conditioner to accurately diagnose the failure by applying the operating state data of the air conditioner operated at the building site.

Referring to FIG. 4 , the step (S10) of building a machine learning learning model for the driving state of the standard air conditioner through the standard air conditioner learning model building unit (S10) sets a simulation model for the driving state of the standard air conditioner in the control server 10 (S11), automatically generating the driving state data of the standard air conditioner in the control server 10 through the learning model determining unit 80 (S12), normalizing the generated driving state data of the standard air conditioner Step (S13), building a normal type machine learning regression model of the standard air conditioner through the learning model determining unit 80 (S14), machine learning for each failure type of the standard air conditioner through the learning model determining unit 80 and constructing a classification model (S15).

The step (S11) of setting a simulation model for the operating state of the standard air conditioner is performed using a simulation tool (EnergyPlus, etc.) that supports modeling of air conditioners that can play a standard role for the purpose of diagnosing failures of air conditioners. It sets the operating state model of the air conditioner and automatically generates normal and failure type data for each heating and cooling season through simulation.

The 22 types of failure types for generating failure state data by classification in the standard air conditioner are shown in FIG. 9 .

The elements constituting the operating state data of the standard air conditioner are the operating date, operating time, outdoor air temperature, supply air temperature, mixing temperature, ventilation temperature, coil valve position, air conditioning air volume, coil flow rate, outdoor air intake amount, and outdoor air damper opening rate for each heating and cooling season. , exhaust damper opening rate, mixed air damper opening rate, economizer status, supply fan wattage, status type, etc. The status types are classified into normal type and failure type.

In the step (S12) of automatically generating the driving state data of the standard air conditioner in the control server 10 through the learning model determining unit 80, the driving state data of the standard air conditioner is individually generated by the learning model determining unit 80. Through the application of the CTGAN algorithm in the artificial intelligence deep learning field, tens of thousands of data for each normal and failure type are automatically generated and added as the driving state data of the standard air conditioner to increase the machine learning learning accuracy. .

CitiGan (Conditional Tabular Generative Adversarial Network; CTGAN) discriminates a model made by artificial intelligence deep learning with a generator to solve the problem of generating tabular data (numerical and classified text data) It is a generative adversarial network (GAN) algorithm that uses additional information to conditionally use additional information while adversarially training two different AI models, called discriminators, to automatically learn data similar to tabular training data. It can be mass-produced.

In the step (S13) of normalizing the generated driving state data of the standard air conditioner in the control server 10, the size is converted to normalize the size between different element values of the driving state data of the standard air conditioner.

)를 생성한다.At this time, outside air temperature, supply air temperature, mixing temperature, and ventilation temperature are normalized as a relative ratio to the set temperature of the supply air, and the air conditioning air volume, coil flow rate, outside introduction amount, and supply fan power amount are normalized as a relative ratio to each rated capacity. The final operating state data of the air conditioner (

) is created.

) 중에서 정상 유형 데이터를 대상으로 표준 공조기의 정상 유형 데이터(

)를 구성하며, 구성 요소 중에서 냉난방 계절별로 외기온도, 급기온도, 혼합온도, 환기온도, 코일밸브 위치, 공조풍량, 코일유량, 외기도입량, 외기댐퍼 개도율, 배기댐퍼 개도율, 혼합공기댐퍼 개도율, 이코너마이저 상태를 속성(독립변수)으로 하여 표준 공조기의 정상 유형 회귀학습용 속성 데이터(

)로 설정하고, 상기 속성 데이터(

)와 99% 이상의 매우 높은 종속관계를 가지고 있는 급기팬 전력량을 레이블(종속변수)로 하여 표준 공조기의 정상 유형 회귀학습용 레이블 데이터(

)로 설정한 후에, 머신러닝 랜덤포레스트 회귀학습을 하여 표준 공조기 정상 유형 회귀모델(

)을 구축한다. The step (S14) of building a normal type machine learning regression model of the standard air conditioner is the final driving state data (

) of the normal type data of the standard air conditioner (

), and among the components, outdoor air temperature, supply air temperature, mixing temperature, ventilation temperature, coil valve position, air conditioning air volume, coil flow rate, outdoor air intake rate, outdoor air damper opening rate, exhaust damper opening rate, mixed air damper opening rate for each heating and cooling season Attribute data for normal type regression learning of standard air conditioners with rate and economizer state as attributes (independent variables)

), and the attribute data (

) and the amount of supply fan power that has a very high dependence of more than 99% as the label (dependent variable), label data for normal type regression learning of standard air conditioners (

), then machine learning random forest regression learning is performed to model the standard air conditioner normal type regression model (

) to build

At this time, the learning model determining unit 80 of the control server 10 further includes a machine learning engine 81 for performing artificial intelligence training and prediction.

)의 요소 중에서 상태 유형을 제외한 나머지 요소들을 속성(종속변수)으로 표준 공조기의 고장 유형 분류학습용 속성 데이터(

)로 설정하고, 상태 유형을 레이블(독립변수)로 하여 표준 공조기의 고장 유형 분류학습용 레이블 데이터(

)로 설정하여, 고장유형별로 머신러닝 랜덤포레스트 분류학습을 하여 표준 공조기 고장 유형별 머신러닝 분류모델(

)을 구축한다. The step (S15) of building a machine learning classification model for each failure type of the standard air conditioner through the learning model determining unit 80 is the final driving state data (

) of the elements except for the state type as attributes (dependent variables), attribute data for classification learning of failure types of standard air conditioners (

) and set the state type as a label (independent variable) to label data (

) to perform machine learning random forest classification learning by failure type, and a machine learning classification model for each standard air conditioner failure type (

) to build

Referring to FIG. 5 , when the machine learning learning model is built through the step (S10) of building the machine learning learning model for the driving state of the standard air conditioner through the standard air conditioner learning model building unit 40, the operation air conditioner 110 ), the sensed data of the operating air conditioner 110 is normalized through the step (S20) of collecting and normalizing the sensing data for the normal type according to the operation from the database 30 of the control server 10.

The step (S20) of collecting and normalizing the sensing data for a normal type according to the operation of the operation air conditioner 110 from the database 30 of the control server 10 is performed according to the operation state of the operation air conditioner 110. It includes a step of collecting normal type data (S21), and normalizing the normal type data of the operation air conditioner 110 in the control server 10 (S22).

The step of collecting normal type data according to the operating state of the operating air conditioner 110 (S21) is to collect the normal type sensing information of each operating air conditioner 110 in the database 30 of the control server 10 is a step

At this time, from the database 30 of the control server 10 at the site where the automatic building control system (BAS), which is the control unit 100 of the operation air conditioner 110, is operated, the operation air conditioner 110 is in a normal state without failure. Only the sensing information is retrieved for about 10 days in units of 1 minute or 5 minutes to configure the normal type data of the operating air conditioner 110 .

Elements of the sensing information include operation date, operating time, outdoor temperature, supply air temperature, mixing temperature, ventilation temperature, coil valve position, air conditioning air volume, coil flow rate, outdoor air intake amount, outdoor air damper opening rate, exhaust damper opening rate, mixed air These are the damper opening rate, economizer status, and supply fan power.

)와의 분포 유사도를 비교분석 하기 위해서 수집된 운용 공조기(110)의 정상 유형 데이터를 정규화를 한다. In the step (S22) of normalizing the normal type data of the operating air conditioner 110 in the control server 10, the normal type data (

) and normalization of the collected normal type data of the air conditioner 110 in order to compare and analyze the distribution similarity.

)를 생성한다.At this time, the outdoor air temperature, the supply air temperature, the mixing temperature, and the ventilation temperature among the normal type data of the operating air conditioner 110 are normalized in a relative ratio to the supply air set temperature, and the air conditioning air volume, coil flow rate, outdoor air introduction amount and supply fan power amount are respectively Normalized by the relative ratio to the rated capacity of the final normal type data (

) is created.

Referring to FIG. 6 , through the step (S20) of collecting and normalizing sensing data for a normal type according to the operation of the operation air conditioner 110 from the database 30 of the control server 10, the operation air conditioner 110 of When the normalization of the normal type data is completed, the sensing data normalized in the control server 10 through the step (S30) of correcting the normal type data of the operating air conditioner 110 and building a normal type regression model is determined by the learning model determining unit ( 80), the optimal polynomial regression model between the measured normal type air supply fan power amount and the predicted air supply fan power amount of the operating air conditioner 110, which is the normal type data of the standard air conditioner, and the operating air conditioner normal type correction data are generated by generating the operating air conditioner 110. A normal type regression model is built and distribution similarity to the normal type data of the operating air conditioner 110 is measured based on this.

)에 운용 공조기(110)의 정상 유형 데이터를 적용하여 급기팬 전력량을 예측하는 단계(S31), 상기 학습모델 결정부(80)를 통해 상기 운용 공조기(110)의 정상 유형 실측 급기팬 전력량과 예측 급기팬 전력량 간의 최적 다항 회귀모델을 결정하는 단계(S32), 상기 최적 다항 회귀모델에 운용 공조기(110)의 실측 급기팬 전력량을 적용하여 정상 유형 급기팬 보정 급기팬 전력량을 산출하는 단계(S33), 운용 공조기(110)의 정상 유형 보정 데이터를 생성하여 운용 공조기(110)의 정상 유형 머신러닝 회귀모델을 구축하는 단계(S34), 상기 표준 공조기 학습모델 구축부(40)에 있는 표준 공조기의 정상 유형에 대한 머신러닝 회귀모델에 운용 공조기(110) 정상 유형 보정 데이터를 적용하여 운용 공조기(110) 정상 유형 보정 급기팬 전력량과 추정 급기팬 전력량의 분포 유사도를 측정하는 단계(S35)를 포함한다.The step (S30) of determining and building a normal type regression model of the operating air conditioner 110 and correcting the normal type data of the operating air conditioner 110 based on this (S30) is the standard in the standard air conditioner learning model building unit 40 Normal type regression model of air conditioner (

) by applying the normal type data of the operating air conditioner 110 to predicting the supply fan power amount (S31), and predicting the normal type actual air supply fan power amount of the operating air conditioner 110 through the learning model determiner 80 Determining an optimal polynomial regression model between the supply air fan wattage (S32), applying the actual air supply fan wattage of the operating air conditioner 110 to the optimal polynomial regression model to calculate the normal type supply air fan corrected supply air wattage (S33) , generating the normal type correction data of the operating air conditioner 110 to build a normal type machine learning regression model of the operating air conditioner 110 (S34), the normality of the standard air conditioner in the standard air conditioner learning model building unit 40 Measuring the similarity of distribution between the operating air conditioner 110 normal type corrected supply fan power amount and the estimated supply air fan power amount by applying the operating air conditioner 110 normal type correction data to the machine learning regression model for the type (S35).

)에 운용 공조기(110)의 정상 유형 데이터를 적용하여 급기팬 전력량을 예측하는 단계(S31)에서는 운용 공조기(110)의 최종 정상 유형 데이터(

) 중에서 정상 유형 실측 급기팬 전력량(

)을 제외한 나머지 요소들을 속성(독립변수)으로 하여 운용 공조기(110)의 정상 유형 회귀학습용 속성 데이터(

)로 설정하고, 정상 유형 실측 급기팬 전력량(

)을 레이블(종속변수)로 설정한다.The normal type regression model of the standard air conditioner (

) in the step (S31) of applying the normal type data of the operating air conditioner 110 to predict the amount of power supply fan, the final normal type data (

) of the normal type actual air supply fan wattage (

), with the remaining elements as attributes (independent variables), attribute data for normal type regression learning of the operation air conditioner 110 (

), and the normal type measured air supply fan wattage (

) as the label (dependent variable).

)에 상기 운용 공조기(110)의 정상 유형 회귀학습용 속성 데이터(

)를 수학식 1에 적용하여 운용 공조기(110)의 정상 유형 예측 급기팬 전력량(

)을 예측한다.A machine learning regression model for the normal type of the standard air conditioner stored in the standard air conditioner learning model building unit 40 (

) to the normal type regression learning attribute data of the operation air conditioner 110 (

) by applying Equation 1 to the normal type predicted supply fan power amount of the operating air conditioner 110 (

) is predicted.

Equation 1

은 표준 공조기 정상 유형 회귀모델,

은 운용 공조기의 정상 유형 예측 급기팬 전력량,

는 운용 공조기의 정상 유형 회귀학습용 속성 데이터이다.here,

is a standard air conditioner normal type regression model,

is the normal type predicted supply fan wattage of the operating air conditioner,

is attribute data for normal type regression learning of operating air conditioners.

)과 정상 유형 예측 급기팬 전력량(

) 간의 오차를 보정하기 위하여, 상기 2개의 급기팬 전력량 간의 함수 관계를 가장 잘 나타낼 수 있는 다항 회귀모델을 결정한다.In the step (S32) of determining the optimal polynomial regression model between the normal type actually measured air supply fan power amount and the predicted air supply fan power amount of the operating air conditioner 110 through the learning model determining unit 80, the normal type measurement of the operating air conditioner 110 is performed (S32). Supply fan power (

) and normal type predicted supply fan wattage (

), a polynomial regression model that can best represent the functional relationship between the two supply fan power amounts is determined.

)과 정상 유형 예측 급기팬 전력량(

) 간의 오차를 보정하면 상기 운용 공조기(110)의 정상 상태 데이터를 표준 공조기의 정상 상태 데이터에 적합화 함으로써 표준 공조기가 상기 운용 공조기(110) 역할을 가상적으로 할 수 있도록 한다. Normal type actual air supply fan power amount of the operating air conditioner 110 (

) and normal type predicted supply fan wattage (

), the normal state data of the operating air conditioner 110 is adapted to the normal state data of the standard air conditioner so that the standard air conditioner can virtually function as the operating air conditioner 110 .

)을 속성(독립변수)으로 설정하고, 정상 유형 예측 급기팬 전력량(

)을 레이블(종속변수)로 설정하여 1차 항부터 5차 항까지 머신러닝 다항 회귀분석(Polynomial Regression Analysis)을 하여 수학식 2의 평균제곱오차(Mean Square Error)가 가장 작은 최적 다항 회귀모델(

)을 결정한다.Normal type actual air supply fan power amount of the operating air conditioner 110 (

) as an attribute (independent variable), and the normal type predicted supply fan wattage (

), A label (set as a dependent variable), the first polynomial regression machine learning from the anti to the fifth section (Polynomial Regression Analysis) to the mean square error (Mean Square Error) is the least optimal polynomial regression model of equation (2) (

) to determine

Equation 2

=

=

는 머신러닝 다항 회귀분석을 차수별(1, 2, 3, 4, 5)로 학습한 후의 평균제곱오차,

은 k차 다항 회귀모델의 운용 공조기의 정상 유형 실측 급기팬 전력량,

는 k차 다항 회귀모델의 운용 공조기의 정상 유형 예측 급기팬 전력량, k는 다항 회귀 차수, i는 운용 공조기의 정상 유형 급기팬 전력량 데이터의 개수이다.here,

is the mean square error after learning machine learning polynomial regression by order (1, 2, 3, 4, 5),

is the power amount of the air supply fan measured in the normal type of the operating air conditioner of the k-order polynomial regression model,

is the normal type predicted supply fan wattage of the air conditioner operated by the k-order polynomial regression model, k is the polynomial regression order, i is the number of wattage data of the normal type air supply fan of the operating air conditioner.

)과 정상 유형 예측 급기팬 전력량(

) 간에 보정해야 할 실제 오차를 산출하기 위해서 최적 다항 회귀모델(

)에 실측 급기팬 전력량(

)을 수학식 3에 적용하여 다시 운용 공조기(110)의 정상 유형 보정 급기팬 전력량(

)을 산출한다.The step (S33) of calculating the normal type corrected air supply fan power amount by applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model (S33) is the normal type actual air supply fan power amount of the operating air conditioner 110 (

) and normal type predicted supply fan wattage (

) to calculate the actual error to be corrected between the optimal polynomial regression models (

) in the actual air supply fan wattage (

) by applying Equation 3 to the normal type corrected supply air power amount of the air conditioner 110 again (

) is calculated.

Equation 3

는 최적 다항 회귀모델,

는 운용 공조기의 정상 유형 보정 급기팬 전력량,

는 운용 공조기의 정상 유형 실측 급기팬 전력량이다.here,

is the optimal polynomial regression model,

is the normal type corrected supply fan wattage of the operating air conditioner,

is the wattage of the normal type actual air supply fan of the operating air conditioner.

)에서 실측 급기팬 전력량(

)을 제외한 나머지 요소들을 속성(독립변수)으로 하여 운용 공조기(110)의 정상 유형 보정 데이터(

)를 생성하고, 운용 공조기(110)의 정상 유형 보정 급기팬 전력량(

)을 추가하여 레이블(종속변수)로 설정하여 머신러닝 랜덤포레스트 회귀학습을 하여 운용 공조기(110)의 정상 유형 회귀모델(

)을 구축한다.In the step (S34) of generating a normal type correction data of the operating air conditioner 110 to build a normal type regression model of the operating air conditioner 110, the normal type final data of the operating air conditioner 110 (

) in the actual air supply fan wattage (

) with the remaining elements as attributes (independent variables), the normal type correction data (

), and the normal type corrected supply fan wattage (

) and set it as a label (dependent variable) to perform machine learning random forest regression learning to perform a normal type regression model (

) to build

)에 운용 공조기(110)의 정상 유형 보정 데이터(

)를 수학식 4에 적용하여 정상 유형 추정 급기팬 전력량(

)을 산출한다.The step (S35) of measuring the distribution similarity between the normal type corrected supply air power amount and the estimated air supply fan power amount of the operating air conditioner 110 is a normal type regression model of the constructed operating air conditioner 110 (

) to the normal type correction data of the operating air conditioner 110 (

) to Equation 4 to estimate the normal type of supply fan power (

) is calculated.

Equation 4

은 운용 공조기 정상 유형 회귀모델의 정상 유형 추정 급기팬 전력량(

)은 운용 공조기의 정상 유형 추정 급기팬 전력량,

는 운용 공조기의 정상 유형 보정 데이터이다.here,

is the normal type estimated supply fan wattage (

) is the estimated power supply fan power for the normal type of the operating air conditioner,

is the normal type correction data of the operating air conditioner.

)과 정상 유형 추정 급기팬 전력량(

) 간 급기팬 전력량의 평균제곱근오차 변동계수(Coefficient of Variance of the Root Mean Square Error;

)와 분포 유사도를 측정할 수 있는 지수인 와서스타인거리 변동계수(Coefficient of Variance of the Wasserstein distance;

)와 에너지거리 변동계수(Coefficient of Variance of the Energy distance;

)를 수학식 5에 의해 산출한다.Normal type corrected supply fan power amount of the operating air conditioner 110 (

) and the estimated normal type supply fan wattage (

) Coefficient of Variance of the Root Mean Square Error;

) and the coefficient of variation of the Wasserstein distance, which is an index that can measure the distribution similarity (Coefficient of Variance of the Wasserstein distance;

) and Coefficient of Variance of the Energy distance;

) is calculated by Equation 5.

Equation 5

은 평균제곱근오차,

은 평균제곱근오차 변동계수,

은 와서스타인거리,

은 와서스타인거리 변동계수,

은 에너지거리,

은 에너지거리 변동계수,

는 운용 공조기의 정상 유형 보정 급기팬 전력량,

는

의 평균,

은 운용 공조기의 정상 유형 추정 급기팬 전력량,

는 운용 공조기의 정상 유형 보정 급기팬 전력량 분포와 운용 공조기의 정상 유형 추정 급기팬 전력량 분포의 결합확률분포의 집합, i는 운용 공조기의 정상 유형 급기팬 전력량 데이터의 개수이다. here,

is the root mean square error,

is the root mean square error coefficient of variation,

Silver Wesstein Street,

is the Wersstein distance coefficient of variation,

silver energy distance,

is the coefficient of variation of the energy distance,

is the normal type corrected supply fan wattage of the operating air conditioner,

is

mean of,

is the estimated power supply fan power for the normal type of the operating air conditioner,

is the set of combined probability distributions of the normal type corrected supply fan wattage distribution of the operating air conditioner and the normal type estimated supply fan wattage distribution of the operating air conditioner, and i is the number of data on the normal type supply fan wattage data of the operating air conditioner.

The Wasserstein distance is an index for measuring the distance between two probability distributions. The correlation between the two probability distributions is measured, and the expected value of the distance between the two distributions is calculated as an estimated value that minimizes, and the similarity between the two probability distributions is measured. is used as an exponent, and if the two distributions match perfectly, the Werstein distance becomes 0.

The energy distance is an index for measuring the statistical distance between probability distributions, and is used as an index for verifying the similarity between two probability distributions. If the two distributions completely match, the energy distance becomes 0.

Referring to FIG. 7 , the normal type data of the operating air conditioner 110 is corrected by correcting the normal type data of the operating air conditioner 110 and the normal type data of the operating air conditioner 110 is corrected through the step S30 of building a normal type regression model. When the type learning model is determined, the machine learning learning model built in the standard air conditioner learning model building unit 40 through the step (S40) of detecting a failure candidate of the operating air conditioner 110 through the learning model determining unit 80 Based on the learning model determination unit 80, a failure candidate of the operating air conditioner 110 is detected.

The step of detecting a failure candidate of the operating air conditioner 110 through the learning model determiner 80 based on the regression model of the operating air conditioner 110 (S40) is a failure diagnosis of the operating air conditioner 110 is required. Collecting and normalizing the sensed data (S41), applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model and calculating the corrected air supply fan power amount for fault diagnosis (S42), Corrected air supply fan power amount for fault diagnosis generating correction data for fault diagnosis of the operating air conditioner 110 by adding a step (S43), calculating the similarity index by applying the correction data for fault diagnosis of the operating air conditioner 110 to the normal type regression model of the operating air conditioner 110 ( S44), and detecting a failure candidate based on the calculated similarity index (S45).

The step (S41) of collecting and normalizing the sensing data for which the failure diagnosis of the operation air conditioner 110 is required is the sensing for which the failure diagnosis of each operation air conditioner 110 in the database 30 of the control server 10 is required. This is the information gathering step.

At this time, the failure diagnosis of the operation air conditioner 110 is required from the database 30 of the control server 10 at the site where the automatic building control system (BAS), which is the control unit 100 of the operation air conditioner 110, is operated. The corresponding sensing information is retrieved to compose the data required for fault diagnosis of the operating air conditioner 110, and the components of the sensing information include the operation date, operating time, outside air temperature, supply air temperature, mixing temperature, ventilation temperature, coil valve position, These are air conditioning air volume, coil flow rate, outside intake amount, outside air damper opening rate, exhaust damper opening rate, mixed air damper opening rate, economizer status, and supply fan power.

)와의 분포 유사도를 비교분석 하기 위해서 정규화를 하여 운용 공조기(110)의 고장진단용 실측 데이터(

)를 생성하면서 급기팬 전력량을 고장진단용 실측 급기팬 전력량(

)으로 설정한다. The collected fault diagnosis of the operating air conditioner 110 is converted to the normal type data of the standard air conditioner (

) and normalized in order to compare and analyze the distribution similarity with the actual measurement data for fault diagnosis of the operating air conditioner 110 (

) and measure the supply fan wattage for fault diagnosis (

) is set.

At this time, the outdoor air temperature, supply air temperature, mixing temperature and ventilation temperature are normalized in relative ratios to the supply air set temperature, and the air conditioning air volume, coil flow rate, outdoor air introduction amount, and supply fan power amount are normalized in relative ratios to each rated capacity.

)에 상기 운용 공조기(110)의 고장진단용 실측 급기팬 전력량(

)을 수학식 6에 적용하여 운용 공조기(110)의 고장진단용 보정 급기팬 전력량(

)을 산출한다.The step (S42) of calculating the corrected supply air power amount for fault diagnosis by applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model (S42) is the optimal polynomial regression model (

) in the actual air supply fan power amount for fault diagnosis of the operating air conditioner 110 (

) by applying Equation 6 to the corrected supply fan wattage for fault diagnosis of the operating air conditioner 110 (

) is calculated.

Equation 6

는 최적 다항 회귀모델,

는 운용 공조기의 고장진단용 보정 급기팬 전력량,

는 운용 공조기의 고장진단용 실측 급기팬 전력량이다.here,

is the optimal polynomial regression model,

is the calibrated supply fan wattage for fault diagnosis of the operating air conditioner,

is the actual power amount of the air supply fan for fault diagnosis of the operating air conditioner.

)에서 고장진단용 실측 급기팬 전력량(

)을 제외한 나머지 요소들을 속성(독립변수)으로 하여 운용 공조기(110)의 고장진단용 회귀모델 속성 데이터(

)로 설정하고, 운용 공조기(110)의 고장진단용 회귀모델 속성 데이터(

)에 운용 공조기(110)의 고장진단용 보정 급기팬 전력량(

)을 추가하여 운용 공조기(110)의 고장진단용 분류모델 데이터(

)로 구성한다.In the step (S43) of generating correction data for fault diagnosis of the operating air conditioner 110 by adding the corrected supply fan power for fault diagnosis of the operating air conditioner 110, the actual measurement data for fault diagnosis of the operating air conditioner 110 (

), the actual amount of power supply fan for fault diagnosis (

) with the remaining elements as attributes (independent variables), regression model attribute data for fault diagnosis of the operating air conditioner 110 (

), and the regression model attribute data for fault diagnosis of the operating air conditioner 110 (

) to the corrected supply fan wattage for fault diagnosis of the operating air conditioner 110 (

) by adding the classification model data for fault diagnosis of the operating air conditioner 110 (

) is composed of

)에 운용 공조기(110)의 고장진단용 회귀모델 속성 데이터(

)를 수학식 7에 적용하여 운용 공조기(110)의 고장진단용 추정 급기팬 전력량(

)을 산출한다.The step of calculating the similarity index by applying the correction data for fault diagnosis of the operating air conditioner 110 to the normal type regression model of the operating air conditioner 110 (S44) is the normal type regression model of the operating air conditioner 110 (

) in the regression model attribute data for fault diagnosis of the operating air conditioner 110 (

) by applying Equation 7 to the estimated power supply fan power for fault diagnosis of the operating air conditioner 110 (

) is calculated.

Equation 7

은 운용 공조기의 정상 유형 다항 회귀모델,

는 운용 공조기의 고장진단용 추정 급기팬 전력량,

는 운용 공조기의 고장진단용 회귀모델 속성 데이터이다.here,

is the stationary type polynomial regression model of the operating air conditioner,

is the estimated supply fan wattage for fault diagnosis of the operating air conditioner,

is the regression model attribute data for fault diagnosis of operating air conditioners.

)과 운용 공조기(110)의 보정 급기팬 전력량(

)을 수학식 8에 적용하여 평균제곱근오차 변동계수(Coefficient of Variance of the Root Mean Square Error;

), 와서스타인거리 변동계수(Coefficient of Variance of the Wasserstein distance;

) 및 에너지거리 변동계수(Coefficient of Variance of the Energy distance;

)의 유사도 지수를 산출한다. Estimated air supply fan power amount for fault diagnosis of the calculated operation air conditioner 110 (

) and the corrected supply fan power amount of the operating air conditioner 110 (

) by applying Equation 8 to the coefficient of variation of the root mean square error (Coefficient of Variance of the Root Mean Square Error;

), Coefficient of Variance of the Wasserstein distance;

) and Coefficient of Variance of the Energy distance;

) to calculate the similarity index.

Equation 8

는 평균제곱근오차,

는 평균제곱근오차 변동계수,

는 와서스타인거리,

는 와서스타인거리 변동계수,

는 에너지 거리,

는 에너지거리 변동계수,

는 운용 공조기의 고장진단용 보정 급기팬 전력량,

는

의 평균,

는 운용 공조기의 고장진단용 추정 급기팬 전력량,

는 운용 공조기의 고장진단용 보정 급기팬 전력량 분포와 운용 공조기의 고장진단용 예측 급기팬 전력량 분포의 결합확률분포의 집합이며, i는 운용 공조기 고장진단용 데이터의 개수이다. here,

is the root mean square error,

is the root mean square error coefficient of variation,

is Wersstein Street,

is the Wersstein distance coefficient of variation,

is the energy distance,

is the coefficient of variation of the energy distance,

is the calibrated supply fan wattage for fault diagnosis of the operating air conditioner,

is

mean of,

is the estimated supply fan wattage for fault diagnosis of the operating air conditioner,

is the set of the combined probability distribution of the distribution of the corrected supply fan wattage for fault diagnosis of the operating air conditioner and the predicted supply fan wattage distribution for the fault diagnosis of the operating air conditioner, and i is the number of data for fault diagnosis of the operating air conditioner.

In the step of detecting a failure candidate based on the calculated similarity index (S45), the calculated root mean square error coefficient of variation, the Weinstein distance coefficient of variation, and the energy distance variation coefficient each have an error range of 2 from the set reference value. If more than one is out of range, it is detected as a failure candidate, and if less than one is out of the error range from the reference value, it is detected as normal.

), 와서스타인거리 변동계수는 기준값보다 높을 경우(

), 에너지거리 변동계수는 기준값보다 높을 경우(

)에 오차 범위를 벗어났다고 진단한다.At this time, the reference value of each similarity index can be set arbitrarily, but the root mean square error coefficient of variation, the Weinstein distance coefficient of variation, and the energy distance coefficient of variation, respectively. Coefficient, set to 1.5 times the coefficient of variation of energy distance, and when the coefficient of variation of root mean square error is higher than the reference value (

), the Werstein distance coefficient of variation is higher than the reference value (

), when the energy distance variation coefficient is higher than the reference value (

) is out of the error range.

Referring to FIG. 8 , the operation air conditioner 110 through the step S40 of detecting a failure candidate of the operation air conditioner 110 through the learning model determiner 80 based on the normal type regression model of the operation air conditioner 110 . ) is detected, the specific failure type of the operating air conditioner 110 is diagnosed through the learning model determiner 80 based on the machine learning learning model built in the standard air conditioner learning model building unit 40 .

)에 운용 공조기(110) 고장진단용 보정 데이터를 적용하여 분류 정확도를 산출하는 단계(S51), 산출된 분류 정확도를 기반으로 고장 유형을 1차 진단하는 단계(S52), 1차 고장 유형 진단을 유보한 경우 머신러닝 분류 학습을 추가 실행하여 고장 유형을 2차 진단하는 단계(S53), 상기 진단된 고장 유형에 따라 작동 성능이 최저가 되지 않도록 임시 조치하는 단계(S54)를 포함한다.The step of diagnosing a specific failure type in the control server 10 for the failure candidate of the operating air conditioner 110 (S50) is, when the operation air conditioner 110 is detected as a failure candidate, machine learning for each failure type of the standard air conditioner classification model (

), calculating the classification accuracy by applying the correction data for fault diagnosis to the operating air conditioner 110 (S51), first diagnosing the failure type based on the calculated classification accuracy (S52), deferring the primary failure type diagnosis In one case, it includes a step of secondary diagnosis of the failure type by additionally executing machine learning classification learning (S53), and a step of taking temporary measures so that the operating performance is not the lowest according to the diagnosed failure type (S54).

)를 상기 표준 공조기 학습모델 구축부(40)에 있는 모든 표준 공조기 고장유형별 머신러닝 분류모델(

)에 개별적으로 수학식 9를 적용하여 분류모델의 고장 유형으로 분류되는 정확도(

)를 각각 산출한다.When the operation air conditioner 110 is detected as a failure candidate, calculating the classification accuracy by applying the correction data for failure diagnosis of the operation air conditioner 110 to a machine learning classification model for each failure type of the standard air conditioner (S51) is detected as a failure candidate Classification model attribute data (

) is a machine learning classification model for all standard air conditioner failure types in the standard air conditioner learning model construction unit 40 (

) by applying Equation 9 individually to the accuracy (

) are calculated respectively.

Equation 9

는 표준 공조기의 고장 유형별 머신러닝 분류모델,

은 운용 공조기의 예측 고장 유형,

는 운용 공조기의 고장진단용 분류모델 속성 데이터,

은 분류 정확도, i는 표준 공조기의 특정 고장 유형이다.here,

is a machine learning classification model by failure type of standard air conditioners,

is the predicted failure type of the operating air conditioner,

is the classification model attribute data for fault diagnosis of operating air conditioners,

is the classification accuracy, i is the specific failure type of the standard air conditioner.

The step (S52) of first diagnosing the failure type based on the calculated classification accuracies includes the highest among those in which the calculated classification accuracy is 75% or more and the classification accuracy in which the error with the highest classification accuracy is 5% or less If there is only one, it is diagnosed as a failure type of a specific classification model, and if there are two or more items with high classification accuracy, the primary diagnosis of the failure type is reserved.

)와 레이블 데이터(

) 중에서 상태 유형이 정상 유형과 분류 정확도가 높게 산출된 고장 유형들을 속성 데이터(

)로 설정하고, 상태 유형을 레이블 데이터(

)로 설정하여, 상기 학습모델 결정부(80)를 통하여 머신러닝 랜덤포레스트(Random Forest) 분류 학습을 추가로 실행하여 표준 공조기의 2차 고장유형 머신러닝 분류모델(

)을 구축하고, 고장 진단이 유보된 운용 공조기의 고장진단용 분류모델 속성 데이터(

)를 수학식 10에 적용하여 분류 정확도 지수를 고장유형별로 산출한 후에 분류 정확도(

)가 가장 높은 것을 고장 유형으로 최종 진단한다.If the primary failure type diagnosis is withheld, the step of diagnosing the failure type by additionally executing machine learning classification learning (S53) is the attribute data for failure type classification learning of the standard air conditioner (

) and label data (

), the state type is the normal type and the failure types calculated with high classification accuracy are classified as attribute data (

), and set the state type to label data (

), by additionally executing machine learning random forest classification learning through the learning model determining unit 80, the secondary failure type machine learning classification model of the standard air conditioner (

), and classification model attribute data (

) is applied to Equation 10 to calculate the classification accuracy index for each failure type, and then the classification accuracy (

) is the highest, and finally diagnosed as the failure type.

Equation 10

는 표준 공조기의 2차 고장유형 머신러닝 분류모델,

은 운용 공조기의 예측 고장 유형,

는 고장 진단이 유보된 운용 공조기의 고장진단용 분류모델 속성 데이터,

는 분류 정확도 지수, j는 표준 공조기의 특정 고장 유형이다.here,

is the secondary failure mode machine learning classification model of standard air conditioners,

is the predicted failure type of the operating air conditioner,

is the classification model attribute data for fault diagnosis of operating air conditioners for which fault diagnosis is deferred,

is the classification accuracy index, and j is the specific failure type of the standard air conditioner.

The step (S54) of temporarily taking measures so that the operating performance of the operating air conditioner 110 is not the lowest according to the diagnosed failure type (S54) is the operating performance of the operating air conditioner 110 is malfunctioned and the operating performance is lowered to the lowest before being completely repaired. This is a step in which the operator takes measures according to the temporarily prepared countermeasures according to the type of failure, as this may increase.

If the failure type is a stuck cooling coil valve or a heating coil valve, temporary measures can be taken to operate the bypass valve with manual operation experience until the stuck valve is repaired.

If the failure type is supply air fan motor degradation, take temporary measures such as a frequency converter if the motor speed can be changed, or adjust the blade angle or belt tension if the motor speed cannot be changed. can

If the failure type is filter contamination, check and replace the filter, and if the failure type is damper stuck, temporary measures may be taken to manually operate the damper.

In addition, depending on the failure type, the air conditioner operator may take various temporary measures to prevent the operating performance of the operating air conditioner 110 from being the lowest.

)에 운용 공조기(110)의 고장진단용 데이터 7종류를 적용하여 운용 공조기(110)의 고장 후보를 검출한 결과 데이터 1과 2는 정상 상태, 데이터 3부터 데이터 8까지는 평균제곱근오차 변동계수, 와인스타인거리 변동계수, 에너지거리 변동계수 유사도 지수가 설정된 기준값과의 오차 범위가 2개 이상 벗어나기 때문에 고장 후보로 검출되었음을 알 수 있다.Referring to FIG. 10 , a normal type machine learning regression model of the operation air conditioner 110 according to an embodiment disclosed in the present invention (

As a result of detecting failure candidates of the operating air conditioner 110 by applying 7 types of fault diagnosis data of the operating air conditioner 110 to ), data 1 and 2 are in a normal state, data 3 to 8 are the root mean square error coefficient of variation, Weinstar It can be seen that the distance variation coefficient and energy distance variation coefficient similarity index are detected as failure candidates because the error range from the set reference value is out of two or more.

)에 개별적으로 적용하여 분류 정확도를 산출하여 1차로 운용 공조기(110)의 고장 유형을 진단한 결과이다.Referring to FIG. 11 , a machine learning classification model for each failure type of a standard air conditioner from data 3 to data 8 detected as failure candidates among 7 types of operation state data of the operating air conditioner 110 according to an embodiment disclosed in the present invention (

) is the result of diagnosing the type of failure of the operating air conditioner 110 primarily by calculating the classification accuracy by applying it individually.

As a result of the first failure diagnosis, in the case of data 7, the diagnosis was completed as failure type 10, and in the case of data 4, data 5, data 6, and data 8, the failure diagnosis was postponed because there were two or more items with high classification accuracy. , in the case of data 3, the highest classification accuracy is 75% or less, so it is a result of diagnosis as a normal type.

)에 데이터 4, 데이터 5, 데이터 6, 데이터 8을 적용하여 분류 정확도를 산출한 결과이다. 상기 산출 결과를 보면 데이터 4는 고장 유형 09로, 데이터 5는 고장 유형 12로, 데이터 6은 고장 유형 11로, 데이터 8은 유형 18로 각각 100%로 정확하게 진단되었음을 알 수 있다.12, according to the embodiment disclosed in the present invention, the secondary failure type machine learning classification model (

) is the result of calculating the classification accuracy by applying data 4, data 5, data 6, and data 8. From the above calculation results, it can be seen that data 4 was correctly diagnosed as failure type 09, data 5 as failure type 12, data 6 as failure type 11, and data 8 as failure type 18, respectively, with 100% accuracy.

In order to detect and diagnose the failure of the air conditioner, the present invention set up a simulation model of the air conditioner that can play a standard role even without accumulating the operating state data for each air conditioner operating in a building site, and provides information about the normal state and the failure state. There is an effect that can be provided at the same time when a building automation system (BAS) is installed by building a machine learning learning model in advance.

In addition, by correcting the operation state data of the operating air conditioner collected at the building site to be suitable for the operation state data of the standard air conditioner, there is an effect that the standard air conditioner can play the role of the operating air conditioner virtually.

In addition, by applying the operating condition data of the operating air conditioner that requires fault diagnosis to the machine learning learning model of the operating air conditioner that is suitable for detecting and diagnosing the failure of the operating air conditioner, it is possible to accurately detect and diagnose the failure from the early stage of operation of the air conditioner. , there is an effect that can reduce the discomfort of heating and cooling and economic loss that may occur due to improper failure of the air conditioner.

The AI-based fault diagnosis method through the adaptation of the standard air conditioner and the operating air conditioner as described above is not limited to the configuration and operation method of the above-described embodiments. The above embodiments may be configured so that various modifications can be made by selectively combining all or part of each of the embodiments.

10: control server 20: communication department
30: Database 40: Standard air conditioner learning model construction unit
50: operation air conditioner correction unit 60: operation air conditioner failure detection unit
70: operation air conditioner failure type diagnosis unit
80: learning model determining unit 81: machine learning engine
100: control unit
110: operation air conditioner

Claims (8)

Based on the data generated through the simulation model calculated by the control server 10 by inputting the equipment specifications and operating conditions for the standard air conditioner into the control server 10, a machine learning learning model for the operating state of the standard air conditioner is applied to the standard air conditioner. Building through the learning model building unit 40 (S10);
Normalizing by collecting the sensing data for the normal type according to the operation of the operation air conditioner 110 installed in the field from the database 30 of the control server 10 (S20);
By applying the normalized sensing data in the control server 10 to the normal type regression model of the standard air conditioner built in the standard air conditioner learning model building unit 40, the normal type data of the operating air conditioner 110 is corrected and the normal type regression building a model (S30);
Detecting a failure candidate of the operating air conditioner 110 through the learning model determiner 80 based on the normal type regression model of the operating air conditioner 110 (S40);
Including a step (S50) of diagnosing a specific failure type in the control server 10 for the detected failure candidate of the operation air conditioner 110,
Correcting the normal type data of the operating air conditioner 110 and building a normal type regression model (S30),
predicting the amount of power supplied by the air conditioner by applying the normal type data of the operating air conditioner 110 to the normal type regression model of the standard air conditioner in the standard air conditioner learning model building unit 40 (S31);
determining an optimal polynomial regression model between the normal type measured air supply fan power amount and the predicted air supply fan power amount of the operating air conditioner 110 through the learning model determining unit 80 (S32);
calculating the normal type corrected air supply fan power amount by applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model (S33);
generating a normal type correction data of the operating air conditioner 110 to build a normal type regression model of the operating air conditioner 110 (S34);
Measuring the similarity of distribution between the operating air conditioner 110 normal type corrected supply fan power amount and the estimated supply air fan power amount by applying the operating air conditioner 110 normal type correction data to the built-up normal type regression model of the operating air conditioner 110 ( S35),
Detecting a failure candidate of the operating air conditioner 110 through the learning model determiner 80 based on the normal type regression model of the operating air conditioner 110 (S40),
Normalizing by collecting the sensing data required for fault diagnosis of the operating air conditioner 110 (S41);
calculating the corrected power supply fan power amount for fault diagnosis by applying the actual air supply fan power amount of the operating air conditioner 110 to the optimal polynomial regression model (S42);
generating correction data for diagnosing the failure of the operating air conditioner 110 by adding the amount of power supplied to the corrected air supply fan for diagnosis of the failure (S43);
calculating a similarity index by applying correction data for fault diagnosis of the operating air conditioner 110 to the normal type regression model of the operating air conditioner 110 (S44);
AI-based failure diagnosis method through the adaptation of standard air conditioners and operating air conditioners, characterized in that it comprises the step (S45) of detecting a failure candidate based on the calculated similarity index.

delete delete delete delete delete According to claim 1,
Diagnosing a specific failure type with respect to the detected failure candidate of the operation air conditioner 110 (S50),
When it is detected as a failure candidate of the operating air conditioner 110, the operation air conditioner 110 failure diagnosis correction data is applied to the machine learning classification model for each failure type of the standard air conditioner in the standard air conditioner learning model building unit 40. Classification accuracy calculating (S51);
first diagnosing a failure type based on the calculated classification accuracy (S52);
In the step of diagnosing the first failure type (S52), if the first failure type diagnosis is reserved, further executing machine learning classification learning to perform secondary diagnosis of the failure type (S53);
Artificial intelligence-based failure diagnosis method through the adaptation of standard air conditioners and operating air conditioners, characterized in that it comprises the step (S54) of taking temporary measures so that the operating performance is not the lowest according to the diagnosed failure type.
delete

Images