KR102512506B1 - Method, System, and Computer-Readable Medium for Automatic Air-Conditioner Control - Google Patents

Abstract

The present invention relates to an air conditioner automatic comfort control method, system, and computer-readable medium using a machine learning model. According to the present invention, by using multiple machine learning models learned based on the state value function and state-action value function, learning the user's operation of the air conditioner during automatic control using the multiple machine learning models, considering environmental information at the time of the operation, considering the individuality of the user and the spatial environment by the machine learning model that controls the air conditioner, automatic control of the air conditioner improves comfort while simultaneously saving energy.

Description

Air conditioner automatic comfort control method, system, and computer-readable medium using machine learning model {Method, System, and Computer-Readable Medium for Automatic Air-Conditioner Control}

The present invention relates to a method, system, and computer-readable medium for automatically controlling comfort of an air conditioner using a machine learning model, using a plurality of machine learning models learned based on a state value function and a state-action value function, The user's operation of the air conditioner during automatic control using the machine learning model is learned in consideration of the environmental information at the time of operation, and the machine learning model that controls the air conditioner takes into account the individuality of the user and the space environment. A method, system, and computer-readable medium for automatically controlling comfort of an air conditioner using a machine learning model, enabling automatic control to improve comfort and save energy at the same time.

Due to the rapid change in temperature due to global warming and the demand for more comfortable living, the number of air conditioners, especially air conditioners, in each home is gradually increasing. there is.

In addition, if you want to install an air conditioner in a large space such as an office or store, you request the installation from a professional company, but the professional company does not install an appropriate number of air conditioners in the space and installs more air conditioners than the appropriate number. There are many cases of installing air conditioners, and as a result, power consumption due to the operation of air conditioners increases, resulting in excessive maintenance costs.

In addition, in the case of commercial facilities such as offices and stores, there are cases in which employees, such as part-timers, operate air conditioners indiscriminately without considering power consumption, increasing the cost burden of the owner of the facility.

Therefore, it is necessary to develop a control method capable of rapidly adjusting the temperature of the space by controlling the operation of a plurality of air conditioners installed in the space and at the same time reducing power consumption of the air conditioner and heater.

Prior Patent 1 (Korean Patent Publication No. 10-2287293) of the applicant of the present invention is a system for controlling a plurality of air conditioners, the system comprising: one or more temperature sensors installed in each of one or more zones; a plurality of controllers controlling each of a plurality of air conditioners installed in each of the one or more zones; one or more repeaters receiving temperature information from the temperature sensor and transmitting air conditioner control information to the controller; Discloses a system including; and a service server that communicates with the repeater.

The service server of Prior Patent 1 includes a first air conditioner control information transmission step of transmitting first air conditioner control information allowing each of the plurality of air conditioners to operate with a preset output for a preset time to the controller through the repeater; a zone temperature information derivation step of deriving zone temperature information for each of the one or more zones based on temperature information received from one or more temperature sensors installed in each of the one or more zones during the predetermined time period; an operation priority determination step of determining operation priorities of a plurality of air conditioners installed for each of the one or more zones based on the zone temperature information; And based on set temperature information for each of the one or more zones, current zone temperature information for each of the one or more zones, and operation priorities of a plurality of air conditioners for each of the one or more zones, second air conditioner control information is derived, A second air conditioner control information transmission step of transmitting the second air conditioner control information to the controller through the repeater.

The present invention uses a plurality of machine learning models learned based on a state value function and a state-action value function, and considers environmental information at the time of operation of a user's operation of an air conditioner during automatic control using a plurality of machine learning models. A machine learning model that allows the automatic control of the air conditioner to improve comfort and save energy at the same time by considering the individuality of the user and the space environment. Its purpose is to provide a method, system, and computer-readable medium for automatically controlling the comfort of an air conditioner.

In order to solve the above problems, an embodiment of the present invention is an air conditioner automatic comfort control method using a machine learning model performed in a computing system having one or more processors and one or more memories, at predetermined time intervals, room temperature; And environmental information, including additional information including at least one of carbon dioxide concentration, office space size, fine dust concentration, and outdoor temperature, is input into the LSTM-based first machine learning model to maintain the air conditioner temperature at the current time and increase the air conditioner temperature , a state value information derivation step of deriving state value information including elements of user's comfort level and energy saving for each air conditioner when the temperature of the air conditioner decreases; At predetermined time intervals, room temperature; And additional information including at least one of carbon dioxide concentration, space size, fine dust concentration, and outdoor temperature; inputting environmental information including the LSTM-based second machine learning model to consider user comfort and energy saving factors. When optimally controlled, action value information including probability information corresponding to each of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease, and any one of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease based on the probability information Control information derivation step of deriving the air conditioner automatic control information belonging to; At a predetermined time interval, the state value information; At least one of probability information and the air conditioner automatic control information; a sample queue update step of updating a sample queue by accumulating each sample information including the environment information; And in the continuous sample information of the sample queue, considering whether the user's air conditioner operation was during a preset time period after each sample information, a compensation value is given to each sample information to derive learning data. Learning data derivation step; A model learning step of learning the first machine learning model and the second machine learning model using the learning data; provides a method for automatically controlling comfort of an air conditioner using a machine learning model.

In some embodiments of the present invention, the air conditioner automatic control information is input to an air conditioner operating in a space related to the environment information and used for automatic control of the air conditioner at predetermined time intervals, and the air conditioner automatic control information includes the air conditioner temperature It may belong to any one of maintaining, raising the temperature of the air conditioner to a preset temperature, and lowering the temperature of the air conditioner to a preset temperature.

In some embodiments of the present invention, in the step of deriving the learning data, after the air conditioner automatic control information becomes any one of maintenance, air conditioner temperature rise, and air conditioner temperature fall in the n th sample information, in the n th sample information If there is no user operation of the air conditioner during the period up to the n+pth sample information, a positive compensation value is given to the nth sample information, and in the model learning step, the environment information and the state for each sample information value information; The first machine learning model and the second machine learning model may be trained using at least one of the probability information and the air conditioner automatic control information and a compensation value.

In some embodiments of the present invention, in the step of deriving the learning data, after the air conditioner automatic control information becomes any one of maintenance, air conditioner temperature rise, and air conditioner temperature fall in the n th sample information, in the n th sample information If there is an air conditioner operation by the user during the section up to the n+pth sample information, a negative compensation value or a compensation value of 0 is given to the nth sample information, and in the model learning step, the above for each sample information environmental information, the state value information; The first machine learning model and the second machine learning model may be trained using at least one of the probability information and the air conditioner automatic control information and a compensation value.

In some embodiments of the present invention, the step of deriving the learning data may be performed during a period from the n-th sample information to the n+p-th sample information after the air conditioner automatic control information in the n-th sample information has increased the temperature of the air conditioner. When the user does not operate the air conditioner, a positive first compensation value is given to the nth sample information, and after the air conditioner automatic control information in the nth sample information maintains or drops the air conditioner temperature, the nth sample information If there is no user operation of the air conditioner during the interval from to the n+q-th sample information, a positive second compensation value is given to the n-th sample information, and the positive first compensation value is the positive second compensation value. As the compensation value of the sample information is larger than the value and the compensation value of the sample information is higher, the learning influence of the sample information on the first machine learning model and the second machine learning model can be positively increased in the model learning step.

In some embodiments of the present invention, in the step of deriving the learning data, after the air conditioner automatic control information becomes any one of maintenance, air conditioner temperature rise, and air conditioner temperature fall in the n th sample information, in the n th sample information If there is no user operation of the air conditioner during the period up to the n+pth sample information, a positive compensation value is given to the nth sample information, the air conditioner automatic control information is maintained in the nth sample information, the air conditioner temperature rises, and when the user operates the air conditioner during the interval from the nth sample information to the n+pth sample information after one of the air conditioner temperature drops, a negative compensation value or a compensation value of 0 for the nth sample information In the model learning step, when each sample information of the training data has a positive compensation value, the probability information and state value information of the corresponding sample information are fed back in a positive direction, so that the first machine learning model and the first machine learning model 2 When the machine learning model is trained and each sample information of the learning data has a reward value of 0 or a negative reward value, the sample information is not used for learning the first machine learning model and the second machine learning model, or The first machine learning model and the second machine learning model may be trained by feeding back the probability information and the state value information of the sample information in a negative direction.

In some embodiments of the present invention, the first machine learning model includes a first LSTM model, the second machine learning model includes a second LSTM model, and the first machine learning model includes a second machine learning model. can be used for learning.

In some embodiments of the present invention, the model learning step may include an output value obtained by inputting environmental information of each sample information included in the learning data to the first machine learning model, the state value information of the corresponding sample information, and the corresponding The first machine learning model is trained to reduce the first error derived based on the compensation value of the sample information, and the environmental information of each sample information included in the learning data is input to the second machine learning model. The second machine learning model can be trained by reducing a second error derived based on the output value, the state value information of the sample information, the probability information of the sample information, and the compensation value of the sample information. .

In order to solve the above problems, in one embodiment of the present invention, an automatic comfort control device for an air conditioner using a machine learning model implemented as a computing system having one or more processors and one or more memories, the computing system comprising: At set time intervals, room temperature; And environmental information, including additional information including at least one of carbon dioxide concentration, office space size, fine dust concentration, and outdoor temperature, is input into the LSTM-based first machine learning model to maintain the air conditioner temperature at the current time and increase the air conditioner temperature , a state value information derivation step of deriving state value information including elements of user's comfort level and energy saving for each air conditioner when the temperature of the air conditioner decreases; At predetermined time intervals, room temperature; And additional information including at least one of carbon dioxide concentration, space size, fine dust concentration, and outdoor temperature; inputting environmental information including the LSTM-based second machine learning model to consider user comfort and energy saving factors. When optimally controlled, action value information including probability information corresponding to each of air conditioner temperature maintenance, air conditioner temperature increase, air conditioner temperature decrease, and air conditioner temperature maintenance based on the probability information, air conditioner temperature increase, air conditioner temperature decrease Control information derivation step of deriving the air conditioner automatic control information belonging to; At a predetermined time interval, the state value information; At least one of probability information and the air conditioner automatic control information; a sample queue update step of updating a sample queue by accumulating each sample information including the environment information; And in the continuous sample information of the sample queue, considering whether the user's air conditioner operation was during a preset time period after each sample information, a compensation value is given to each sample information to derive learning data. Learning data derivation step; An air conditioner automatic comfort control device using a machine learning model is provided, which performs a model learning step of learning the first machine learning model and the second machine learning model using the learning data.

According to an embodiment of the present invention, by using a plurality of machine learning models learned based on the state value function and the state-action value function, the user's operation history, environment history, and the operation in the previous machine learning model By learning the machine learning model in consideration of compensation, it is possible to exert an effect of providing an automatic air conditioner control function considering the individual characteristics of the user and the environment.

According to an embodiment of the present invention, a user's operation of an air conditioner during automatic control using a plurality of machine learning models is learned in consideration of environmental information at the corresponding operation time, and a machine learning model that controls the air conditioner is configured to be used by the user and the user. Considering the individuality of the spatial environment, it is possible to exert an effect of deriving automatic control information.

According to an embodiment of the present invention, the automatic control of the air conditioner can achieve an effect of improving comfort and at the same time saving energy.

1 schematically illustrates an environment in which an automatic comfort control method for an air conditioner according to an embodiment of the present invention is performed.
FIG. 2 schematically illustrates all steps of an air conditioner automatic comfort control method according to an embodiment of the present invention and an internal configuration of a computing system performing the method.
3 schematically shows the overall operation process of the method for automatically controlling comfort of an air conditioner according to an embodiment of the present invention based on internal components and data.
4 schematically illustrates sample queue data according to an embodiment of the present invention.
5 illustratively illustrates a process of deriving learning data according to an embodiment of the present invention.
6 illustratively illustrates a process of deriving learning data according to an embodiment of the present invention.
7 schematically illustrates a learning process of a machine learning model based on learning data according to an embodiment of the present invention.
8 schematically illustrates structures of a first machine learning model and a second machine learning model according to an embodiment of the present invention.
9 illustratively illustrates the internal neural network structure of the first machine learning model according to an embodiment of the present invention.
10 illustratively shows an energy saving effect according to an automatic comfort control method for an air conditioner according to an embodiment of the present invention.
11 schematically illustrates the internal configuration of a computing device according to an embodiment of the present invention.

In the following, various embodiments and/or aspects are disclosed with reference now to the drawings. In the following description, for purposes of explanation, numerous specific details are set forth in order to facilitate a general understanding of one or more aspects. However, it will also be appreciated by those skilled in the art that such aspect(s) may be practiced without these specific details. The following description and accompanying drawings describe in detail certain illustrative aspects of one or more aspects. However, these aspects are exemplary and some of the various methods in principle of the various aspects may be used, and the described descriptions are intended to include all such aspects and their equivalents.

Moreover, various aspects and features will be presented by a system that may include a number of devices, components and/or modules, and the like. It should also be noted that various systems may include additional devices, components and/or modules, and/or may not include all of the devices, components, modules, etc. discussed in connection with the figures. It must be understood and recognized.

"Example", "example", "aspect", "exemplary", etc., used herein should not be construed as preferring or advantageous to any aspect or design being described over other aspects or designs. . The terms '~unit', 'component', 'module', 'system', 'interface', etc. used below generally mean a computer-related entity, and for example, hardware, hardware It may mean a combination of and software, software.

Also, the terms "comprises" and/or "comprising" mean that the feature and/or element is present, but excludes the presence or addition of one or more other features, elements and/or groups thereof. It should be understood that it does not.

In addition, terms including ordinal numbers, such as first and second, may be used to describe various components, but the components are not limited by the terms. These terms are only used for the purpose of distinguishing one component from another. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element, without departing from the scope of the present invention. The terms and/or include any combination of a plurality of related recited items or any of a plurality of related recited items.

In addition, in the embodiments of the present invention, unless otherwise defined, all terms used herein, including technical or scientific terms, are generally understood by those of ordinary skill in the art to which the present invention belongs. has the same meaning as Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the embodiments of the present invention, an ideal or excessively formal meaning not be interpreted as

1 schematically illustrates an environment in which an automatic comfort control method for an air conditioner according to an embodiment of the present invention is performed.

A method for automatically controlling comfort of an air conditioner according to embodiments of the present invention is performed in a computing system having one or more processors and one or more memories.

As shown in FIG. 1, an air conditioner and an environment information sensor are provided in a space such as a room. In this computing system, an operation control signal of the air conditioner is transmitted to the air conditioner, and the air conditioner operates according to the operation control signal received by the computing system.

An operating state of such an air conditioner may be changed even by a user's manipulation.

On the other hand, the environmental information sensor collects indoor temperature, carbon dioxide concentration, fine dust concentration, external temperature, etc. and transmits them to the computing system. Some functions of the environment information sensor can be performed by the air conditioner. For example, indoor temperature information may be sensed by an air conditioner and transmitted to a computing system. In addition, some functions of the environment information sensor correspond to the form of receiving data from an external system connected through a network. For example, the outdoor temperature may be received from an external server through the Internet.

Such an environment information sensor is a conceptual configuration, and can be implemented in various forms regardless of location as long as it provides space information to a computing system, and some functions may be physically integrated into an air conditioner.

As shown in FIG. 1, the computing system may create air conditioner control information for a plurality of spaces to control the air conditioner, but in one embodiment of the present invention, it may also perform control for a single space.

FIG. 2 schematically illustrates all steps of an air conditioner automatic comfort control method according to an embodiment of the present invention and an internal configuration of a computing system performing the method.

An air conditioner automatic comfort control method according to embodiments of the present invention is performed in a computing system having one or more processors and one or more memories.

In step S100, at a predetermined time interval, room temperature; And environmental information, including additional information including at least one of carbon dioxide concentration, office space size, fine dust concentration, and outdoor temperature, is input into the LSTM-based first machine learning model to maintain the air conditioner temperature at the current time and increase the air conditioner temperature , a state value information derivation step of deriving state value information including user comfort and energy saving factors for each air conditioner temperature is lowered.

The environmental information is continuously received at predetermined time intervals, which is sequentially input to the 1st machine learning model based on LSTM, and the LSTM module of the 1st machine learning model maintains the temperature of the air conditioner at the current time and increases the temperature of the air conditioner. , when the air conditioner temperature is lowered, the state value information is derived considering the user's comfort and energy cost. The first machine learning model considers the energy cost and user comfort from the environmental information input in sequence including the past, and based on the environmental information input in time series, the value at each of the air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease derive a value The state value information may include vector values such as [X1, X2, X3], where X1 is a value when the temperature of the air conditioner is maintained and can be expressed as a real number, and X2 is a value when the temperature of the air conditioner rises. It can be expressed as a real number, and X3 can be expressed as a real number as a value when the air conditioner temperature is lowered or lowered.

For example, if the comfort control method of the present invention operates at 5-minute time intervals from 8:00 and the current time is 9:00, 8:00, 8:05, 8:10, 8:15, 8:20, The environmental information at 8:25, 8:30, 8:35, 8:40, 8:45, 8:50, 8:55, and 9:00 is input to the first machine learning model, and at 9:00 A vector of a set of real numbers representing each state value information including elements of user's comfort level and energy saving for each of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease can be derived.

In step S200, at a predetermined time interval, room temperature; And additional information including at least one of carbon dioxide concentration, space size, fine dust concentration, and outdoor temperature; inputting environmental information including the LSTM-based second machine learning model to consider user comfort and energy saving factors. When optimally controlled, action value information including probability information corresponding to each of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease, and any one of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease based on the probability information A control information derivation step of deriving the air conditioner automatic control information belonging to is performed.

In the control information derivation step, as in step S100, environment information according to time series is received, and air conditioner automatic control information is derived accordingly.

An example of the air conditioner automatic control information is to maintain the air conditioner temperature, increase the temperature of the air conditioner by 1 degree Celsius, decrease the temperature of the air conditioner by 1 degree Celsius, or maintain the temperature of the air conditioner, increase the temperature of the air conditioner by 2 degrees Celsius, decrease the temperature of the air conditioner by 2 degrees Celsius, or maintain the temperature of the air conditioner, Celsius X may correspond to an air conditioner temperature increase, and Celsius X may correspond to an air conditioner temperature decrease. That is, in the case of an increase or decrease in the air conditioner temperature, an increase or decrease by a predetermined temperature may be derived, or even a temperature range may be variably derived according to circumstances.

The second machine learning model corresponds to a deep learning-based learned reasoning model that derives optimal air conditioner automatic control information in consideration of energy cost and user comfort from environmental information input in sequence including the past. Similarly, the second machine learning model may include an LSTM module.

The air conditioner automatic control information derived in step S200 is input to the air conditioner of the corresponding space, and the air conditioner drives the air conditioner according to the air conditioner automatic control information. The air conditioner control according to the air conditioner automatic control information can be controlled by directly inputting the air conditioner automatic control information into the air conditioner itself, or by inputting the air conditioner automatic control information into a separate controller in the middle and controlling the air conditioner by the controller. can be implemented

In step S300, at a preset time interval, the state value information; At least one of probability information and the air conditioner automatic control information; A sample queue update step of updating the sample queue by accumulating each sample information including the environment information; is performed.

The sample information includes output information of the first machine learning model and the second machine learning model, and environment information according to the operation time interval in S100 and S200. Each sample information may directly or indirectly include information about the corresponding time and whether or not the user operated the air conditioner.

In step S400, in the sample information of the sample queue, learning data is derived by assigning a compensation value to each sample information in consideration of whether or not the user's air conditioning operation was performed for a preset time period after each sample information. learning data derivation step; is performed.

Preferably, the learning data may include each sample information and a compensation value given to each sample information, and a detailed derivation process of the compensation value will be described later.

In one embodiment of the present invention, in step S400, a preset rule is applied to the stacked sample queue, a compensation value is assigned, and sample information to be used as training data is derived.

In step S500, a model learning step of learning the first machine learning model and the second machine learning model using the learning data is performed.

Preferably, the first machine learning model and the second machine learning model are machine learning models for a specific space, and perform optimal control of the air conditioner tailored to the user and space by learning the specificity of the space and the user of the space. You can exert the effect you can. A first machine learning model and a second machine learning model may be provided for each space.

Figure 2 (B) schematically shows the internal configuration of a computing system that performs the comfort control method of the present invention.

The computing system includes a state value information derivation unit, a control information derivation unit, a sample queue update unit, a learning data derivation unit, and a model learning unit, each of which includes the above-described target temperature derivation step, control information derivation step, and sample queue update unit. step, learning data derivation step, and model learning step are performed.

3 schematically shows the overall operation process of the method for automatically controlling comfort of an air conditioner according to an embodiment of the present invention based on internal components and data.

Environmental information collected at predetermined time intervals from air conditioners, environmental information sensors, external servers, or user input (eg, area information of the corresponding space, etc.) is input to the first machine learning model and the second machine learning model. .

Preferably, the first machine learning model may correspond to a value neural network, and the second machine learning model may correspond to a policy neural network.

The value output from the first machine learning model is not actually used to control the air conditioner, but is used in the sample queue and used to train the second machine learning model. Such a first machine learning model plays a kind of CRITIC role in reinforcement learning. Such a first machine learning model is learned in the model learning step by assigning a compensation value to the sample information of the sample queue as described above.

Meanwhile, the value output from the second machine learning model is primarily used for automatic control of the air conditioner, and at the same time, it is included in the sample information of the sample queue.

A compensation value is given to the sample information of the sample queue, and the second machine learning model is learned. In this case, the inference result or characteristic of the first machine learning model is reflected in the learning of the second machine learning model in the error function, so that the learning of the first machine learning model (some inference models) eventually learns the second machine learning model. results in

The air conditioner automatic control information is input to an air conditioner operating in a space related to the environment information and used for automatic control of the air conditioner at a preset time interval, and the air conditioner automatic control information maintains the air conditioner temperature and maintains the air conditioner temperature at a preset temperature. It belongs to either rising or lowering the temperature of the air conditioner to a preset temperature.

4 schematically illustrates sample queue data according to an embodiment of the present invention.

The sample queue data includes a plurality of sample information including environmental information collected according to a preset time period, inference results of the first machine learning model, and inference results of the second machine learning model. In the present invention, sample queue data is generated by accumulating such sample information in the computing system.

In embodiments of the present invention, the step of deriving the learning data may include determining whether the user operates the air conditioner for a predetermined time or more after the air conditioner automatic control information becomes one of maintenance, air conditioner temperature rise, and air conditioner temperature fall. A compensation value is given to each sample information by determining whether the

The first machine learning model and the second machine learning model are linked as a result, and the air conditioner is controlled according to the air conditioner automatic control information derived by the second machine learning model in which environmental information including time-series past history is input. In the air conditioner controlled as described above, when there is no user manipulation, the inference results of the first machine learning model and the second machine learning model are judged to be inference results that satisfy the user's comfort, and the inference at this time By learning the first machine learning model and the second machine learning model based on the result and environment information, the first machine learning model and the second machine learning model are separately labeled in consideration of the individual space, environment information, and individuality of the user. It can exert the effect of continuously learning without work or updating. The user's operation of the air conditioner may serve as a kind of labeling.

5 illustratively illustrates a process of deriving learning data according to an embodiment of the present invention. 6 illustratively illustrates a process of deriving learning data according to an embodiment of the present invention.

In an embodiment of the present invention, in the step of deriving the learning data, after the air conditioner automatic control information becomes any one of maintenance, air conditioner temperature rise, and air conditioner temperature fall in the n th sample information, in the n th sample information If there is no user operation of the air conditioner during the period up to the n+pth sample information, a positive compensation value is given to the nth sample information, and in the model learning step, the environment information and the state for each sample information value information; The first machine learning model and the second machine learning model are trained using at least one of the probability information and the air conditioner automatic control information and a compensation value.

Preferably, in the step of deriving the learning data, after the air conditioner automatic control information becomes any one of maintenance, air conditioner temperature rise, and air conditioner temperature fall in the nth sample information, n+pth from the nth sample information If there is an operation of the air conditioner by the user during the period up to the sample information, a negative compensation value or a compensation value of 0 is given to the nth sample information, and in the model learning step, the environmental information, the above state value information; The first machine learning model and the second machine learning model are trained using at least one of the probability information and the air conditioner automatic control information and a compensation value.

Preferably, in the step of deriving the learning data, after the air conditioner automatic control information in the nth sample information has increased the temperature of the air conditioner, the user operates the air conditioner during a period from the nth sample information to the n+p1th sample information. If there is no , a positive first compensation value is given to the n-th sample information.

Referring to FIG. 5, assuming that p1 is 7, the sample information of T1 has a compensation value of 0 or a negative compensation value because there is a user's manipulation in T2.

Similarly, the sample information of T2 has a compensation value of 0 or a negative compensation value because there was a user's manipulation in T2.

On the other hand, in the case of the T3th sample information, it is confirmed that there is no user manipulation from T3 to T10, and accordingly, in the case of the T3 sample information, a first compensation value is given. This is despite the increase in air conditioner temperature that reduces air conditioning, recognizing that there was no user operation during the preset period corresponds to very accurate operation of the air conditioner automatic control information belonging to the sample information at the time, and giving a large compensation value for this. will be.

Then, in the case of the T4th sample information, it is confirmed that there is no user manipulation from T4 to T11, and accordingly, in the case of the T4 sample information, a first compensation value is given.

Then, in the case of the T5 th sample information, it is confirmed that there is no user manipulation from T5 to T12, and accordingly, in the case of the T5 sample information, a first compensation value is given.

Then, in the case of the T6th sample information, it is confirmed that there is no user manipulation from T6 to T13, and accordingly, in the case of the T6 sample information, a first compensation value is given.

Then, in the case of the T7 th sample information, it is confirmed that there is no user manipulation from T7 to T14, and accordingly, in the case of the T7 sample information, a first compensation value is given.

Thereafter, in the case of the T8 th sample information, a compensation value of negative or 0 is given because there was a user's manipulation at T15.

In one embodiment of the present invention, the first compensation value may be fixed as a positive constant (for example, +2) without considering the discount rate. However, in a preferred embodiment of the present invention, T3 to T7 are 1A discount rate is applied to the compensation value.

For example, if the discount rate is 0.1 and the reward value of T3 is 2, the reward value of T4 is 2*0.9, the reward value of T5 is 2*0.9*0.9, and the reward value of T6 is 2*0.9 *0.9*0.9, and the compensation value of T7 becomes 2*0.9*0.9*0.9*0.9.

On the other hand, after the air conditioner automatic control information in the nth sample information maintains or decreases the air conditioner temperature, when there is no user operation of the air conditioner during the period from the nth sample information to the n+p2th sample information, the nth sample A positive second compensation value is assigned to the information.

Referring to FIG. 6, assuming that p1 is 12, the sample information of T1 has a compensation value of 0 or a negative compensation value because there is a user's manipulation in T2.

Similarly, the sample information of T2 has a compensation value of 0 or a negative compensation value because there was a user's manipulation in T2.

On the other hand, in the case of the T3th sample information, it is confirmed that there is no user manipulation from T3 to T15, and accordingly, in the case of the T3 sample information, a second compensation value is given. This is to recognize that no manipulation by the user during a predetermined period after the automatic control corresponds to very accurate manipulation of the air conditioner automatic control information belonging to the sample information at the time, and to give a large compensation value for this.

As described with reference to FIG. 5, such a second compensation value may be applied to a sequence of sample information thereafter. Also, the same discount rate as the first compensation value may be applied to consecutive second compensation values.

More preferably, the first compensation value of the quantity is greater than the second compensation value of the quantity, and the higher the compensation value of the sample information, the first machine learning model and the second machine learning model of the corresponding sample information in the model learning step. The learning effect on the learning model is positively increased.

That is, the first compensation value corresponds to an automatic control result of energy saving such as an air conditioner temperature rise. Accordingly, energy saving characteristics may be maintained for the first and second machine learning models by training the first and second machine learning models with a higher weight on the corresponding sample information.

More preferably, p1 is set smaller than p2. In the case of an increase in the air conditioner temperature for energy saving, there is a high possibility that the user will operate the air conditioner afterwards, and therefore, if it is equal to p2, there is a possibility that compensation will not be received even though the control is appropriate. Therefore, by setting p1 smaller than p2 as in the embodiments of FIGS. 5 and 6, it is possible to achieve an effect of more accurately securing energy-saving labeling data.

That is, in the model learning step, according to whether the initial air conditioner automatic control information corresponds to maintenance, air conditioner temperature rise, or air conditioner temperature decrease in the continuous sample information of the time section derived as learning data, the corresponding learning data The learning degree or learning method of the first machine learning model and the second machine learning model using

More preferably, when the initial air conditioner automatic control information is increased, a higher learning degree or learning weight is assigned compared to air conditioner temperature maintenance and air conditioner temperature decrease.

That is, for example, the learning data when the initial air conditioner automatic control information is maintained is maintained learning data, and when the air conditioner automatic control information is initially increased, the learning data when the air conditioner automatic control information is increased is set to increase learning data. In the case where the learning data when the initial air conditioner automatic control information is descending is set as descending learning data, the first machine learning model and the second machine learning are respectively one maintenance learning data, ascending learning data, and descending learning data. In the case of learning with the model, the first machine learning model and the second machine learning model have a greater learning effect by the ascending learning data than the learning effect by the maintenance learning data or the descending learning data.

That is, even if the same learning data, in the case of ascending learning data, it has a greater influence on the learning of the first machine learning model and the second machine learning model.

In the case of ascending learning data, this is data that satisfies the user's comfort despite performing energy saving, and by giving a larger learning weight than other maintenance learning data or descending learning data, the first machine learning model and the second machine learning data A machine learning model can be learned in the direction of saving energy while satisfying individual characteristics of the user.

Preferably, in the model learning step, when the initial air conditioner automatic control information corresponds to an air conditioner temperature increase in the continuous sample information of the time section derived as learning data, the initial air conditioner automatic control information is the air conditioner temperature The learning weights of the first machine learning model and the second machine learning model using the corresponding learning data are higher than when the temperature is maintained or the temperature of the air conditioner is lowered.

7 schematically illustrates a learning process of a machine learning model based on learning data according to an embodiment of the present invention.

Some of the sample information of the sample queue is derived as learning data through the process described above. Such learning data includes a plurality of environment information according to time series, a reasoning result of the first machine learning model, a reasoning result of the second machine learning model (air conditioner automatic control information), and a compensation value.

Preferably, the first machine learning model is trained using environment information, state value information, and compensation value of each sample information included in the plurality of learning data according to time series.

Preferably, the second machine learning model is trained using environment information, state value information, probability information, and/or air conditioner automatic control information and compensation values of each of the sample information included in the plurality of learning data according to the time series. .

As described above, in the step of deriving the learning data, after the air conditioner automatic control information becomes one of maintenance, air conditioner temperature rise, and air conditioner temperature fall in the nth sample information, n+p in the nth sample information. If there is no user operation of the air conditioner during the period up to the sample information, a positive compensation value is given to the nth sample information, the air conditioner automatic control information is maintained in the nth sample information, the air conditioner temperature rises, and the air conditioner temperature After one of the descents, if there is an air conditioner operation by the user during the interval from the nth sample information to the n+pth sample information, a negative compensation value or a compensation value of 0 is given to the nth sample information. .

In the model learning step, when each sample information of the learning data has a positive compensation value, the probability information and state value information of the corresponding sample information are fed back in a positive direction, so that the first machine learning model and the second machine learning model learn That is, when similar environmental information is input in time series, the first machine learning model and the second machine learning model can reproduce the inference results (sample information of the learning data) of the first machine learning model and the second machine learning model at the time. Train the learning model.

On the other hand, if each sample information of the learning data has a compensation value of 0 or a negative compensation value, the sample information is not used for learning the first machine learning model and the second machine learning model, or the probability information of the sample information, By feeding back the state value information in a negative direction, the first machine learning model and the second machine learning model are trained. That is, when similar environmental information is input in time series, the first machine learning model prevents the inference results (sample information of the learning data) of the first machine learning model and the second machine learning model at the time from being reproduced or vice versa. The model and the second machine learning model are trained.

8 schematically illustrates structures of a first machine learning model and a second machine learning model according to an embodiment of the present invention.

Preferably, the first machine learning model includes a first LSTM model, the second machine learning model includes a second LSTM model, and the first machine learning model is used for learning the second machine learning model. do.

The model learning step is based on the output value obtained by inputting the environmental information of each sample information included in the learning data to the first machine learning model, the state value information of the corresponding sample information, and the compensation value of the corresponding sample information. The first machine learning model is trained to reduce the derived first error.

For example, in one embodiment of the present invention, environment information 1 of FIG. 8 corresponds to environment information of sample information 1 of learning data. The 1st LSTM is learned to narrow the difference (error) between the inference value 1 and GT1 (ground truth), which is obtained by inputting the environment information of sample information 1 into the 1st LSTM.

In determining such an error or ground truth, the state value information of the corresponding sample information and the compensation value of the corresponding sample information are used.

In the simplest embodiment, when the compensation value of the sample information has a positive value, GT can be designated as the state value information. Preferably, as the compensation value increases from a positive value, the compensation value is reflected in the Air function to increase the difference, and accordingly, the learning effect of the LSTM model can be increased. That is, in this way, energy saving characteristics can be imparted.

Alternatively, in the simplest embodiment, when the compensation value of the sample information is negative or 0, the sample information may be excluded from learning or GT may be designated as the reverse of the state value information. That is, in this way, energy saving characteristics can be imparted.

The output value obtained by inputting the environmental information of each sample information included in the learning data to the second machine learning model, the state value information of the corresponding sample information, the probability information of the corresponding sample information, and the compensation value of the corresponding sample information The second machine learning model is learned by reducing the second error derived based on the second error.

For example, in one embodiment of the present invention, environment information 1 of FIG. 8 corresponds to environment information of sample information 1 of learning data. The second LSTM is learned to reduce the difference (error) between the inference value 2 and GT2 (ground truth), which is obtained by inputting the environment information of the sample information 1 into the 2 LSTM.

In determining such an error or ground truth, the state value information of the corresponding sample information, probability information (or air conditioner automatic control information), and the compensation value of the corresponding sample information are used.

In the simplest embodiment, when the compensation value of the sample information has a positive value, GT2 can be designated as the corresponding probability information. Preferably, as the compensation value increases from a positive value, the compensation value is reflected in the Air function to increase the difference, and accordingly, the learning effect of the LSTM model can be increased. That is, in this way, energy saving characteristics can be imparted.

Alternatively, in the simplest embodiment, when the compensation value of the sample information is negative or 0, the corresponding sample information may be excluded from learning or GT may be designated as the reverse of the corresponding probability information. That is, in this way, energy saving characteristics can be imparted.

In a preferred embodiment of the present invention, the output value obtained by inputting the environmental information of each sample information included in the learning data to the second machine learning model, the state value information of the corresponding sample information, the probability information of the corresponding sample information, and learning the second machine learning model by reducing a second error derived based on the compensation value of the corresponding sample information. In this case, the learned first machine learning model is used to learn the second machine learning model, so that the accuracy of reasoning of the second machine learning model can be further increased. This can be done by using the inference value of the first machine learning model in the corresponding sample information to calculate the second error.

In another embodiment of the present invention, operation may be performed only with the second machine learning model without the first machine learning model. In this case, the second machine learning model is derived based on the output value obtained by inputting the environment information of each sample information included in the learning data to the second machine learning model, the probability information of the corresponding sample information, and the compensation value of the corresponding sample information. By reducing the error, the second machine learning model is trained.

Reinforcement learning is a type of machine learning. When an action is taken in an environment, it later judges whether it is a good action or a wrong action, and then rewards (or punishes) to learn on its own through repetition.

Reinforcement learning assumes that there are two components, an environment and an agent, as follows.

An agent decides an action in a particular environment, and the environment rewards that decision. This reward is often determined all at once after multiple actions are taken, rather than immediately. This is because in many cases, when a particular action is taken, it is not possible to evaluate the action immediately.

In such reinforcement learning, artificial neural networks based on deep learning are mainly used when an agent determines an action and learns by itself with a reward given by the environment. The artificial neural network determines an action using the environment and the state of the agent as input values, and if there is a reward, it learns the previous input values and actions positively.

A Markov decision process (MDP) is a decision-making model based on a Markov process. The MDP is composed of a state set S, an action set A, a state transition probability matrix P, a reward function R, and a discount factor γ, as shown in the formula below.

The state set is the set of all possible states in MDP S={s_1,s_2,… ,s_|S|}. As shown in the equation below, the state S_t at any point in time becomes a specific state included in the state set S.

The action set is the set of all actions that the agent, the action subject, can perform A=a1,a2,... ,a|A|. The agent takes the action At=a,a∈A at some point. The state transition probability formula in MDP is shown below.

Pas,s′ is the probability that an agent will change to state s′ when it takes action a in state s.

A reward function is a function that gives an agent a reward for an action taken in a certain state. The formula is as follows.

Reward function Ras Returns the expected value of the reward when action a is taken in state s as a number.

The discount factor is a value between 0 and 1 that determines how much past behavior is reflected. If the discount factor γ is 1, then <1,1,1,1,1>, and if the discount factor γ is 0.9, <1,0.9,0.81,0.729,0.6561> do. In other words, compensation for the distant past is reduced and reflected.

The state-value function refers to the value that an agent in an air conditioner performs in a given environment (air conditioner target temperature). Such a value may include user comfort and energy saving effect. The agent receives a reward as the state changes through actions, and the final value obtained by adding the element of how long it took to receive the reward can be summarized as the following equation.

(Where s is the state, R is the reward, S _t is the state at time t, and r is the discount factor)

In the above state value function, it is assumed that the state changes over time while the agent performs certain actions, and at this time, a reward is received, and the value obtained by adding the discounted reward over time can be expressed as the above expression. .

Actions are determined by policy π, and states are determined by determined actions. That is, the state value function returns the value when policy π is followed in the current state s.

On the other hand, the state-action-value function, also called the Q function, returns the value when action a is performed in a certain state s, which is the expected value of the total reward. The formula of the Q function is as follows.

As in the above formula, we can see that the condition for action a is added in the state value function. The state-action value function is the value of performing action a in accordance with policy π in the current state s.

9 illustratively illustrates internal neural network structures of a first machine learning model and a second machine learning model according to an embodiment of the present invention.

10 illustratively shows an energy saving effect according to an automatic comfort control method for an air conditioner according to an embodiment of the present invention.

11 schematically illustrates the internal configuration of a computing device according to an embodiment of the present invention.

The computing device 1000 illustrated in FIG. 2 may include components of the computing device 11000 illustrated in FIG. 11 .

As shown in FIG. 11, a computing device 11000 includes at least one processor 11100, a memory 11200, a peripheral interface 11300, an input/output subsystem ( I/O subsystem 11400, a power circuit 11500, and a communication circuit 11600 may be included at least. At this time, the computing device 11000 may correspond to the service server 100 shown in FIG. 2 .

The memory 11200 may include, for example, high-speed random access memory, magnetic disk, SRAM, DRAM, ROM, flash memory, or non-volatile memory. . The memory 11200 may include a software module, a command set, or other various data necessary for the operation of the computing device 11000.

In this case, access to the memory 11200 from other components, such as the processor 11100 or the peripheral device interface 11300, may be controlled by the processor 11100.

Peripheral interface 11300 may couple input and/or output peripherals of computing device 11000 to processor 11100 and memory 11200 . The processor 11100 may execute various functions for the computing device 11000 and process data by executing software modules or command sets stored in the memory 11200 .

The input/output subsystem can couple various input/output peripherals to peripheral interface 11300. For example, the input/output subsystem may include a controller for coupling a peripheral device such as a monitor, keyboard, mouse, printer, or touch screen or sensor to the peripheral device interface 11300 as needed. According to another aspect, input/output peripherals may be coupled to the peripheral interface 11300 without going through the input/output subsystem.

The power circuit 11500 may supply power to all or some of the terminal's components. For example, power circuit 11500 may include a power management system, one or more power sources such as a battery or alternating current (AC), a charging system, a power failure detection circuit, a power converter or inverter, a power status indicator or power It may contain any other components for creation, management and distribution.

The communication circuit 11600 may enable communication with another computing device using at least one external port.

Alternatively, as described above, the communication circuit 11600 may include an RF circuit and transmit/receive an RF signal, also known as an electromagnetic signal, to enable communication with other computing devices.

The embodiment of FIG. 11 is just one example of the computing device 11000, and the computing device 11000 may omit some components shown in FIG. 11, further include additional components not shown in FIG. It may have a configuration or arrangement combining two or more components. For example, a computing device for a communication terminal in a mobile environment may further include a touch screen or a sensor in addition to the components shown in FIG. , Bluetooth, NFC, Zigbee, etc.) may include a circuit for RF communication. Components that may be included in the computing device 11000 may be implemented as hardware including one or more signal processing or application-specific integrated circuits, software, or a combination of both hardware and software.

Methods according to embodiments of the present invention may be implemented in the form of program instructions that can be executed through various computing devices and recorded in computer readable media. In particular, the program according to the present embodiment may be composed of a PC-based program or a mobile terminal-specific application. An application to which the present invention is applied may be installed in the service server 100 or the user terminal 500 through a file provided by the file distribution system. For example, the file distribution system may include a file transmission unit (not shown) for transmitting the file according to a request of the service server 100 or the user terminal 500 .

The device described above may be implemented as a hardware component, a software component, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA) , a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. The device can be commanded. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed on networked computing devices and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

According to an embodiment of the present invention, a system for controlling a plurality of air conditioners sets operation priorities for the plurality of air conditioners for each zone, and operates the plurality of air conditioners to adjust the temperature of the corresponding zone based on the operation priorities. Since the order is determined, it is possible to exert an effect of performing cooling by minimizing power consumption.

According to an embodiment of the present invention, since the system for controlling a plurality of air conditioners derives zone temperature information of a corresponding zone based on one or more temperature sensors installed in the zone, the temperature information of the corresponding zone rather than the temperature near the air conditioner is obtained. It is possible to exert an effect of controlling the operation of a plurality of air conditioners in accordance with the temperature of the corresponding zone by calculating the temperature.

According to an embodiment of the present invention, the operation priority determination step determines the operation priority for each air conditioner by operating only a specific air conditioner and calculating the amount of change in zone temperature information for a predetermined time, It is possible to exert an effect of determining operation priorities by considering all various factors related to cooling and heating performance, such as the period and the available range of cooling and heating.

According to an embodiment of the present invention, the second air conditioner control information controls the air conditioner having the highest first priority for a corresponding zone to operate at maximum output for a predetermined time, so that cooling is performed to a set temperature within a short time. At the same time, it can exert the effect of optimizing power consumption.

According to an embodiment of the present invention, the second air conditioner control information causes the air conditioner having the highest first priority to operate at maximum output for a predetermined period of time, after which the zone temperature information of the corresponding zone is higher than the set temperature information. In this case, since the lower priority air conditioners are additionally operated, the effect of reducing the power consumption of the plurality of air conditioners can be exerted.

According to an embodiment of the present invention, the operation priority determination step determines the operation priority by considering not only a plurality of air conditioners installed in the corresponding area, but also one or more air conditioners installed in other areas that affect the corresponding area. It can exert an effect that can easily perform cooling and heating for.

According to an embodiment of the present invention, in the operation prioritization step, since the operation priority is determined by grouping two or more air conditioners with respect to a plurality of air conditioners for the corresponding zone, cooling to the set temperature for the corresponding zone is more rapid. It can exert a cooling effect that can be achieved.

According to an embodiment of the present invention, since the service server provides current zone temperature information for each zone and state information of a plurality of air conditioners installed for each zone to the user terminal, the user can easily grasp the current heating and cooling state, etc. can be effective.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

Claims (9)

An automatic comfort control method for an air conditioner using a machine learning model performed in a computing system having one or more processors and one or more memories, the method comprising:
At predetermined time intervals, room temperature; And environmental information, including additional information including at least one of carbon dioxide concentration, office space size, fine dust concentration, and outdoor temperature, is input into the LSTM-based first machine learning model to maintain the air conditioner temperature at the current time and increase the air conditioner temperature , a state value information derivation step of deriving state value information including elements of user's comfort level and energy saving for each air conditioner when the temperature of the air conditioner decreases;
At predetermined time intervals, room temperature; And additional information including at least one of carbon dioxide concentration, space size, fine dust concentration, and outdoor temperature; inputting environmental information including the LSTM-based second machine learning model to consider user comfort and energy saving factors. When optimally controlled, action value information including probability information corresponding to each of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease, and any one of air conditioner temperature maintenance, air conditioner temperature increase, and air conditioner temperature decrease based on the probability information Control information derivation step of deriving the air conditioner automatic control information belonging to;
At a predetermined time interval, the state value information; At least one of probability information and the air conditioner automatic control information; a sample queue update step of updating a sample queue by accumulating each sample information including the environment information; and
In the sample information of the sample queue, learning data derivation for deriving learning data by assigning a compensation value to each sample information in consideration of whether or not the user's air conditioner operation was during a preset time period after each sample information step;
A method for automatically controlling comfort of an air conditioner using a machine learning model, including a model learning step of learning the first machine learning model and the second machine learning model using the learning data.
The method of claim 1,
The air conditioner automatic control information is input to an air conditioner operating in a space related to the environment information and is used for automatic control of the air conditioner at predetermined time intervals,
The air conditioner automatic control information belongs to any one of maintaining the temperature of the air conditioner, raising the temperature of the air conditioner to a preset temperature, and lowering the temperature of the air conditioner to a preset temperature.
The method of claim 1,
The step of deriving the learning data,
When the user does not operate the air conditioner during the period from the nth sample information to the n+pth sample information after the air conditioner automatic control information is maintained, the air conditioner temperature rises, or the air conditioner temperature falls in the nth sample information In, a positive compensation value is given to the nth sample information,
In the model learning step, the environment information and the state value information for each sample information; An air conditioner automatic comfort control method using a machine learning model, wherein the first machine learning model and the second machine learning model are learned using at least one of the probability information and the air conditioner automatic control information and a compensation value.
The method of claim 1,
The step of deriving the learning data,
When the user operates the air conditioner during the interval from the nth sample information to the n+pth sample information after the air conditioner automatic control information is maintained, the air conditioner temperature rises, or the air conditioner temperature falls in the nth sample information In, a negative compensation value or a compensation value of 0 is given to the nth sample information,
In the model learning step, the environment information and the state value information for each sample information; Learning the first machine learning model and the second machine learning model using at least one of the probability information and the air conditioner automatic control information and a compensation value. Air conditioner automatic comfort control method using machine learning model.
The method of claim 1,
The step of deriving the learning data,
After the air conditioner automatic control information in the nth sample information has risen in the air conditioner temperature, if there is no user operation of the air conditioner during the period from the nth sample information to the n+p1th sample information, the nth sample information is positive. A first compensation value is given,
After the air conditioner automatic control information in the nth sample information maintains or decreases the air conditioner temperature, if there is no user operation of the air conditioner during the period from the nth sample information to the n+p2th sample information, the nth sample information A positive second compensation value is given to
The first compensation value of the quantity is greater than the second compensation value of the quantity, and the higher the compensation value of the sample information, the learning of the first machine learning model and the second machine learning model of the corresponding sample information in the model learning step. A method for automatically controlling comfort of an air conditioner using a machine learning model with a positive effect.
The method of claim 1,
The step of deriving the learning data,
When the user does not operate the air conditioner during the period from the nth sample information to the n+pth sample information after the air conditioner automatic control information is maintained, the air conditioner temperature rises, or the air conditioner temperature falls in the nth sample information In, a positive compensation value is given to the nth sample information,
When the user operates the air conditioner during the interval from the nth sample information to the n+pth sample information after the air conditioner automatic control information is maintained, the air conditioner temperature rises, or the air conditioner temperature falls in the nth sample information In, a negative compensation value or a compensation value of 0 is given to the nth sample information,
In the model learning step,
When each sample information of the learning data has a positive compensation value, the probability information and state value information of the corresponding sample information are fed back in a positive direction to learn the first machine learning model and the second machine learning model,
If each sample information of the training data has a compensation value of 0 or a negative compensation value, the corresponding sample information is not used for learning the first and second machine learning models, or the probability information and state value of the corresponding sample information An automatic comfort control method for an air conditioner using a machine learning model to learn a first machine learning model and a second machine learning model by feeding back information in a negative direction.
The method of claim 1,
The first machine learning model includes a first LSTM model,
The second machine learning model includes a second LSTM model,
The automatic comfort control method for an air conditioner using a machine learning model, wherein the first machine learning model is used for learning the second machine learning model.
The method of claim 7,
In the model learning step,
A first error derived based on the output value obtained by inputting the environmental information of each sample information included in the learning data to the first machine learning model, the state value information of the corresponding sample information, and the compensation value of the corresponding sample information To reduce, training the first machine learning model,
The output value obtained by inputting the environmental information of each sample information included in the learning data to the second machine learning model, the state value information of the corresponding sample information, the probability information of the corresponding sample information, and the compensation value of the corresponding sample information A method for automatically controlling comfort of an air conditioner using a machine learning model, wherein the second machine learning model is learned by reducing a second error derived based thereon.
An automatic comfort control device for an air conditioner using a machine learning model implemented as a computing system having one or more processors and one or more memories,
The computing system,
At predetermined time intervals, room temperature; And environmental information, including additional information including at least one of carbon dioxide concentration, office space size, fine dust concentration, and outdoor temperature, is input into the LSTM-based first machine learning model to maintain the air conditioner temperature at the current time and increase the air conditioner temperature , a state value information derivation step of deriving state value information including elements of user's comfort level and energy saving for each air conditioner when the temperature of the air conditioner decreases;
At predetermined time intervals, room temperature; and additional information including at least one of carbon dioxide concentration, space size, fine dust concentration, and outdoor temperature; inputting the environmental information including the into the LSTM-based second machine learning model to consider the user's comfort and energy saving factors. When optimally controlled, action value information including probability information corresponding to each of air conditioner temperature maintenance, air conditioner temperature increase, air conditioner temperature decrease, and air conditioner temperature maintenance based on the probability information, air conditioner temperature increase, air conditioner temperature decrease Control information derivation step of deriving the air conditioner automatic control information belonging to;
At a predetermined time interval, the state value information; At least one of probability information and the air conditioner automatic control information; a sample queue update step of updating a sample queue by accumulating each sample information including the environment information; and
In the continuous sample information of the sample queue, learning to derive learning data by assigning a compensation value to each sample information in consideration of whether or not the user's air conditioner operation was during a preset time period after each sample information data extraction step;
An air conditioner automatic comfort control device using a machine learning model, performing a model learning step of learning the first machine learning model and the second machine learning model using the learning data.

Images