KR20180057780A - Energy management system and method using machine learning - Google Patents

Abstract

The present invention provides an energy management system for energy management. The energy management system for energy management comprises: a meter measuring an electricity consumption amount of a specific area; an area agent receiving control information on operation of multiple electronic devices from a user terminal capable of controlling the operation of the multiple electronic devices installed in the area and generating operation state current measurement data of the multiple electronic devices in accordance with the control information, wherein the area agent receives electricity consumption amount data of the area from the meter; and a machine learning device receiving the electricity consumption amount data and the operation state current measurement data from the area agent and generating operation state estimation data on the multiple electronic devices by a machine learning model including the electricity consumption amount data as input data, wherein the machine learning device learns the machine learning model using operation state error data in accordance with differences between the operation state current measurement data and the operation state estimation data.

Description

TECHNICAL FIELD [0001] The present invention relates to an energy management system and an energy management method using machine learning,

The present invention relates to an energy management system (EMS) technology.

Recently, the rapid increase in energy demand has worsened the electricity shortage. In order to overcome this power shortage, social costs have increased rapidly due to the addition of power generation and transmission and distribution facilities, and the supply of electricity has been delayed. As a result, the government is shifting from past supply-oriented to demand management-oriented energy policies.

Power demand management is a way to meet stable power demand at the lowest cost through changing consumer usage patterns. Demand management of electricity can be divided into demand reaction and energy efficiency improvement. The application of this power demand management to buildings, homes, and factories is expected to have a significant effect on energy demand management.

Recently, various smart grid technologies such as renewable energy such as solar power, LED lighting, ESS (Enery Storage System), electric vehicle and smit meter have been introduced into buildings, Market demand for BEMS (Building Energy Management System), HEMS (Home Energy Management System) and FEMS (Factory Management System) technologies is increasing.

On the other hand, the energy management system (EMS) is greatly influenced by the characteristics of residents (residents) in the space to be managed. Since the characteristics of these residents are difficult to model mathematically, Development of management system technology is needed.

In order to apply the machine learning method to the energy management system, the actual data for learning the machine learning model is required. In most of the management spaces, the sensor for generating the actual data is not installed. It is difficult to apply it as it is. In addition, when real-time data that is difficult to generate through sensors, such as user convenience, is required, it is difficult to implement an energy management system with a more general machine learning technique.

In this context, it is an object of the present invention, in one aspect, to provide a technique for applying a machine learning technique to an energy management system. In another aspect, an object of the present invention is to provide a technique for generating experimental data to learn a machine learning model in an energy management system applying a machine learning technique.

In order to achieve the above-mentioned object, in one aspect, the present invention provides a sensor network installed in a specific area; A zone agent for managing environmental data obtained from the sensor network; A machine for receiving the environmental data from the zone agent, generating prediction data for energy management in the zone through a machine learning model including the environmental data as input data, and transmitting the prediction data to the zone agent Wherein the zone agent performs an energy management process for the zone using the prediction data, receives user feedback from a registered user terminal corresponding to the zone, And transmits the data to the machine learning apparatus. The machine learning apparatus provides an energy management system that learns the machine learning model using error data generated according to a difference between the actual data and the predicted data.

In another aspect, the invention provides a method comprising: obtaining environmental data for a zone from a sensor network installed in a particular zone;

Generating prediction data for energy management in the zone through a machine learning model including the environmental data as input data; Transmitting a feedback request signal to a registered user terminal corresponding to the zone; Generating actual data corresponding to the predicted data using user feedback received from the user terminal; And learning the machine learning model using the difference between the actual data and the predictive data.

In another aspect, the invention provides a meter comprising: a meter for measuring a power usage of a particular area; Receiving control information on operation of the plurality of electronic devices from a user terminal capable of controlling operation of a plurality of electronic devices installed in the zone, and outputting operating state actual data of the plurality of electronic devices in accordance with the control information A zone agent that generates and receives power usage data of the zone from the meter; And receiving operating state actual data and power consumption data from the zone agent and generating operating state prediction data for the plurality of electronic devices through a machine learning model including the power consumption data as input data, And a machine learning device for learning the machine learning model using operation state error data corresponding to the difference between the actual state data of the operating state and the operation state prediction data.

In another aspect, the present invention provides a method comprising: obtaining environmental data for a zone from a sensor network installed in a particular zone; Generating user data by using a user operation signal or a radio signal received from a registered user terminal corresponding to the zone; Generating prediction data for energy management in the zone through a machine learning model including the environment data and the user data as input data; Generating actual data corresponding to the predicted data; And learning the machine learning model using the difference between the actual data and the predictive data.

In this energy management method, the user operation signal is a set value operation signal for an energy device located in the zone, the user data includes a user convenience index or a user satisfaction index, and in the step of generating the user data, The user convenience index or the user satisfaction index may be generated according to a set value operation signal for the energy device.

In the energy management method, the user data may include user position information or user attitude information, and in the step of generating the user data, the user position information or the user attitude information may be generated using the radio signal of the user terminal Can be generated.

In this energy management method, in the step of generating the user data, the user convenience index or the user satisfaction index may be generated by further using the user location information or the user attitude information.

As described above, according to the present invention, an energy management system can be configured using a machine learning technique. In addition, according to the present invention, it is possible to implement an energy management system by efficiently reflecting characteristics of residents, which are difficult to model mathematically through a machine learning technique. In addition, according to the present invention, it is possible to generate actual measurement data in an energy management system using a machine learning technique to learn a machine learning model. Further, according to the present invention, it is possible to generate actual measurement data even in a space free of a suitable sensor, to learn a machine learning model, to generate actual measurement data which is difficult to be generated even through a sensor such as user convenience, do.

1 is a conceptual diagram of an energy management system according to an embodiment.
2 is a diagram showing a hierarchical structure of a community.
3 is a system configuration diagram of a community level.
4 is a system configuration diagram of a building level.
5 is a system block diagram of a zone level.
6 is a flowchart of an energy demand forecasting method according to an embodiment.
7 is a block diagram of a machine learning model for estimating the number of occupants.
Figure 8 is a block diagram of a generalized machine learning model that can be applied to embodiments of the present invention.
9 is a diagram showing a first example of actual data generation according to an embodiment.
FIG. 10 is a flowchart schematically showing a process of machine learning for estimating the number of occupants.
11 is a flowchart of an energy management method according to an embodiment.
12 is a view for explaining an energy management system according to another embodiment.

Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. It should be noted that, in adding reference numerals to the constituent elements of the drawings, the same constituent elements are denoted by the same reference symbols as possible even if they are shown in different drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

In describing the components of the present invention, terms such as first, second, A, B, (a), and (b) may be used. These terms are intended to distinguish the constituent elements from other constituent elements, and the terms do not limit the nature, order or order of the constituent elements. When a component is described as being "connected", "coupled", or "connected" to another component, the component may be directly connected to or connected to the other component, It should be understood that an element may be "connected," "coupled," or "connected."

1 is a conceptual diagram of an energy management system according to an embodiment.

Referring to FIG. 1, the energy management system 100 includes a community system and a cloud system.

The community system includes at least one agent device 110, and the cloud system includes at least one or more machine learning devices 120.

The agent apparatus 110 generates measurement data from various sensors and meters located in the space of the community, and transmits the measurement data to the machine learning apparatus 120 through the network. The agent device 110 includes a UI (User Interface) device, and transmits the configuration information acquired from the user to the machine learning device 120 through the UI device.

The machine learning apparatus 120 generates predictive data and optimization data through a machine learning model using measurement data and setting information received from the agent apparatus 110 as input data, and transmits the predictive data and the optimization data to the agent apparatus 110.

The agent apparatus 110 manages the energy of the community by using the predictive data and the optimization data received from the machine learning apparatus 120. [

In the community, the agent devices 110 can be hierarchically arranged, and the agent devices 110 disposed at each layer can independently transmit and receive information while communicating with the machine learning device 120.

The hierarchical structure of the community will be described with reference to Figs. 2 to 5. Fig.

2 is a diagram showing a hierarchical structure of a community.

2, a community energy management agent (CEMA) is located at the highest community level and at least one building energy management agent (BEMA) is connected to the community agent (CEMA) at the following building level Located. At least one zone agent (ZEMA) is connected to each building agent (BEMA) at the next zone level, and each zone agent (ZEMA) is located at the lowest sensor network level Energy devices - such as HAVC devices, lighting devices, and personalization devices - such as a personal computer (PC) - meters, sensors, and so on.

3 is a system configuration diagram of a community level.

Referring to FIG. 3, at least one building 310 is located in the community. The community may include community devices not belonging to the building 310, for example, a community may include a community load 320 - a streetlight, a traffic signal system, and so on. The community may include a community distributed power source 330, a community ESS (Energy Storage System) 340, a community EV (electric vehicle) charging station 350, etc. as a community device.

The community may include all of the above-described community devices-community load 320, community distributed power 330, community ESS 340, community EV charging station 350, or may include only at least one community device have.

The community has a community agent (CEMA) that manages the energy of the community as a whole. The community agent (CEMA) is associated with the building agent (BEMA) located in each building (310), and supervises each building agent (BEMA). The community agent CEMA is connected to the community devices 320, 330, 340, and 350 that do not belong to the respective buildings and acquires the state information of the community devices 320, 330, 340, and 350, And controls each of the community devices 320, 330, 340, and 350.

4 is a system configuration diagram of a building level.

Referring to FIG. 4, each building is subdivided into at least one zone 410.

The person building the energy management system according to one embodiment can subdivide each building into a plurality of thermal zones. Here, the thermal zone may mean a zone in which a thermal energy device - for example, a cooling / heating device - is independently controlled. Different thermal zones can be controlled in different thermal states. For example, the first thermal zone may be controlled at a room temperature of 23 degrees, and the second thermal zone may be controlled at a room temperature of 28 degrees.

The building may include building devices that do not belong to each zone 410. For example, the building may include a building load 420 - an elevator load, etc., a building distributed power source 430, a building ESS 440, a building EV charging station 450, and the like.

The building has a building agent (BEMA) that manages the energy of the entire building. The building agent (BEMA) is associated with a zone agent (ZEMA) located in each zone 410, and administers each zone agent (ZEMA). In addition, the building agent BEMA is connected to the building devices 420, 430, 440, and 450 that are not included in each zone, and acquires status information of the respective building devices 420, 430, 440, and 450, And controls each of the building devices 420, 430, 440, and 450.

5 is a system block diagram of a zone level.

Referring to FIG. 5, energy zones 312 and 314 and personalization equipment 316 may be located in each zone.

The personalization device 316 is an electric device that reflects the characteristics of the individual. For example, a personal computer (PC), a desk stand illumination, and the like belong to the personalization device 316. The energy devices 312 and 314 are electric devices not corresponding to the personalization device 316 and are mainly electric devices for controlling the environment of each zone. For example, a heating, ventilation, and Ari condition (HVAC) device 312 and a lighting device 314 belong to an energy device. Devices 312, 314, and 316 may be attached with device sensors 560. The device sensor 560 may be, for example, a sensor for the on-off state of the devices 312, 314, and 316 and may be a meter that monitors the energy usage of the devices 312, 314,

A plurality of sensors may be located in each zone. For example, a temperature sensor 520, a CO2 sensor 530, a humidity sensor 540, an illuminance sensor 550, and the like may be located in each zone.

The sensors 520, 530, 540, 550 and the instrument sensor 560 may constitute the sensor network 510. The configurations of the sensor network 510 may then communicate measurement values to the zone agent (ZEMA) while communicating with the zone agent (ZEMA).

The zone is located with the Zone Agent (ZEMA), which manages the energy of the entire area. The zone agent ZEMA obtains the environmental data from the sensors 520, 530, 540 and 550 and the appliance energy usage data (appliance energy consumption data from the energy devices 312 and 314 and the personalization appliance 316) Which can be regarded as a kind of environmental data).

Energy management systems can use the hierarchical structure of the community to predict energy demand.

6 is a flowchart of an energy demand forecasting method according to an embodiment.

Referring to FIG. 6, energy demand is predicted for each zone (S602).

The zone agent (ZEMA) located in each zone transmits the environmental data acquired from the sensor network and the energy consumption data of the equipment obtained from the electric equipment to the machine learning device and receives the zone energy demand forecast data.

When energy demand is predicted for each district, energy demand is predicted for each building (S604).

The building agent (BEMA) located in each building transmits the zone energy demand forecast data for each zone and the status information of the building devices not belonging to each zone to the machine learning device and receives the building energy demand forecast data.

Once energy demand is predicted for each building, the energy demand for the entire community is then predicted (S606).

The Community Agent (CEMA) transmits the building energy demand forecast data for each building and the status information of the community devices not belonging to each building to the machine learning device and receives the community energy demand forecast data.

This hierarchical structure of the energy management system enables forecasting of energy demand for each district.

The energy management system can estimate the number of occupants using a machine learning device.

7 is a block diagram of a machine learning model for estimating the number of occupants.

The energy management system can predict the number of occupants in each zone before anticipating the energy demand for the zone. The energy management system can predict the number of occupants first, thereby improving user convenience and energy efficiency for residents in each area.

Referring to FIG. 7, the machine learning model 710 for predicting occupancy number may include temperature data, humidity data, CO2 data, and personalization device energy usage data as input data. Temperature data, humidity data, CO2 data, and personalization appliance energy usage data are all variables related to attendance. For example, an increase in the number of occupants in each zone leads to a higher CO2 concentration. The machine learning model (710) learns the relationship between the concentration of CO2 and the number of occupants, and predicts the number of occupants when CO2 data is input at the prediction stage. Personalization energy consumption data can also be closely correlated with the number of occupants. For example, if your PC is in a high energy state, it may mean that the person who runs the PC is in that area. The machine learning model 710 learns the relationship between the energy usage data of the personalization machine and the number of occupants, and when the personalization machine energy consumption data is input at the prediction stage, the number of occupants can be predicted.

The machine learning model 710 may further use roughness data or near field communication (Bluetooth) data with the user terminal as input data. In order to generate the local area communication data, a local area communication device (for example, a Bluetooth device) may be arranged in each area. Such a short-range communication device can determine whether the user is located indoors through short-range communication with a user terminal (e.g., mobile phone) and generate it as short-range communication data.

The machine learning model 710 may use both temperature data, humidity data, CO2 data, and personalization device energy usage data as input data, but depending on the embodiment, only a part of the data may be used as input data. For example, the machine learning model 710 can calculate the number of lost workers using CO2 data as input data. According to the embodiment, the temperature data, the humidity data, and the device energy usage data for the personalization device may be further included as input data.

The machine learning model 710 may compare the occupancy number prediction data with the actual data to generate error data and change the parameters and structure in the machine learning model 710 so that the error value in the error data becomes smaller. At this time, each zone may not include a sensor for measuring the number of occupants. In such a situation, the machine learning model 710 may utilize other information in order to acquire actual measurement data. For example, a zone agent (ZEMA) may include a UI device, and the user input information may be utilized as actual data of the machine learning model 710 at irregular intervals.

The machine learning model 710 can generate the occupancy pattern as well as the occupancy number. The occupancy pattern may be a value representing the occupancy over time, and the machine learning model 710 may also generate such occupancy patterns. The machine learning apparatus can generate occupant estimation information including the occupant number and occupant pattern through the machine learning model 710, and generate the occupant estimating data using the occupant estimating information.

The energy management system can generate control scenarios for equipment located in the zone in addition to the number of occupants by means of a machine learning device. At this time, the machine learning apparatus can predict the user convenience and create the control scenario so that the user convenience is maintained at a predetermined value or more.

Figure 8 is a block diagram of a generalized machine learning model that can be applied to embodiments of the present invention.

Referring to FIG. 8, the environment data obtained from the sensor network may be passed to the machine learning model (MLM) input data of the machine learning device via the zone agent.

The machine learning apparatus can then generate predictive data for energy management in each zone through this machine learning model. The generated prediction data can be transmitted to the zone agent and used as data of the energy management process for each zone.

On the other hand, in the machine learning model, the learning can be enhanced according to the error data generated according to the difference between the actual data and the predictive data. As a specific example, the machine learning model can reinforce learning so that the size of error data (or a numerical value corresponding to this size) generated according to the difference between the actual data and the predictive data becomes a certain size or less. In the machine learning model, if the error data is large, the error data can be made small by changing the internal parameters or the internal structure.

Since the machine learning model is difficult to completely simulate the actual environment of each zone, the machine learning model can keep the error data below a certain value.

On the other hand, in order to learn such a machine learning model, actual measurement data is required. However, when there is no sensor for generating actual measurement data in each zone, it becomes a problem. One embodiment of the present invention provides a technique for generating user feedback received from a user terminal as actual data.

9 is a diagram showing a first example of actual data generation according to an embodiment.

Referring to FIG. 9, a sensor network 510 installed in a specific zone may transmit environment data to a zone agent (ZEMA) (S900). Environmental data may include, for example, temperature sensor readings, humidity sensor readings or CO2 sensor readings.

The zone agent (ZEMA) may transmit the environmental data to the machine learning apparatus 120 while managing the environmental data (S902).

The machine learning apparatus 120 may generate prediction data for energy management in the zone through a machine learning model including environmental data as input data, and may transmit the generated prediction data to the zone agent ZEMA (S904) .

The prediction data may include, by way of example, the number of occupants of the zone. At this time, the machine learning apparatus can generate the prediction data on the number of occupants for the zone by using the temperature sensor measurement value, the humidity sensor measurement value, and the CO2 sensor measurement value.

The Zone Agent (ZEMA) can use the prediction data to perform the energy management process for the zone. For example, it is possible to operate an energy device - for example, an air conditioner - to a suitable range for the number of occupants, and to control the operation of the devices to meet energy demand forecasting data.

Then, the zone agent ZEMA may transmit a feedback request signal to the registered user terminal 920 in response to the zone (S906). The feedback request signal may be, for example, a message asking whether to rewire.

The zone agent ZEMA receives user feedback (e. G., Response to reacquisition) from the user terminal 920 (S908), generates actual data using the user feedback and transmits it to the machine learning device 120 (S910).

Then, the machine learning apparatus 120 can learn the machine learning model using error data generated according to the difference between the actual data and the predicted data.

On the other hand, the prediction data generated by the machine learning apparatus may include a user convenience index. For example, a machine learning device may create control scenarios for devices located in a zone. When creating control scenarios, a control scenario may be created such that the user-friendliness index is maintained above a predetermined value. In the energy management system, minimizing the energy cost can also be an objective function, but user convenience can be another objective function. Accordingly, when the machine learning apparatus generates the control scenario, the control scenario can be created such that the usability index is maintained at a predetermined value or more. Since the user convenience index is subject to the subjective judgment of the resident, the machine learning apparatus can estimate the user's convenience when performing the created control scenario as predicted data.

In this embodiment, the Zone Agent (ZEMA) may receive user feedback on user convenience from the user terminal 920. [ As an example, the zone agent (ZEMA) may receive the comfort index of the room temperature from the user terminal 920. As a specific example, the user terminal 920 may display various levels of comfort and the value may be sent to the ZONE agent (ZEMA) if the user selects one of these steps. The zone agent (ZEMA) can synthesize such user feedback to generate actual data for the user convenience index. If there are more than two user terminals corresponding to the zone, the zone agent (ZEMA) can statistically process the user feedback - for example, averaging - to generate the actual data for the user convenience index.

On the other hand, if the feedback request signal is continuously transmitted to the user terminal 920 (e.g., a mobile communication device, a smart phone, etc.), the user may feel uncomfortable. To reduce this inconvenience, the zone agent ZEMA may send a feedback request signal to the user terminal 920, but may not send a feedback request signal when the learning of the machine learning model is terminated.

The machine learning apparatus 120 receives the measured data from the zone agent (ZEMA) until the machine learning model is learned at a certain level or more, and terminates the learning of the machine learning model when the predetermined level or more is reached. Here, the predetermined level may be determined by a value corresponding to the size of the error data.

The learning may not be terminated permanently and may be resumed. The zone agent ZEMA periodically generates actual data and transmits it to the machine learning device 120. When the size of the error data becomes larger than a predetermined value, the machine learning device 120 can restart the learning of the machine learning model have.

User feedback may be generated by an app (application) installed in the user terminal 920.

FIG. 10 is a flowchart schematically showing a process of machine learning for estimating the number of occupants.

Referring to FIG. 10, the zone agent may generate environmental data (temperature, humidity, CO2 concentration, etc.) for the subdivided zone and transmit it as input data to the machine learning device (S1000).

Then, the machine learning apparatus can generate the prediction data on the number of occupants through the current state machine learning model using the input data (S1002).

The generated prediction data is transmitted to the zone agent again, and the zone agent can transmit the predicted data to the user terminal. Then, the user can transmit the measured data in a manner of feeding back the predicted data shown in the user terminal (S1004). For example, the user can feedback whether the number of lost persons is correct through the user terminal.

The actual data (feedback data) is transmitted to the machine learning device again, and the machine learning device is used to enhance the machine learning (S1006).

11 is a flowchart of an energy management method according to an embodiment.

Referring to FIG. 11, the energy management system may acquire environmental data for a zone from a sensor network installed in a specific zone (S1100).

Then, the energy management system can generate the prediction data for energy management in the zone through a machine learning model including environmental data as input data (S1102).

The energy management system may transmit a feedback request signal to the registered user terminal corresponding to the zone (S1104).

The energy management system may generate actual data corresponding to the predicted data using user feedback received from the user terminal (S1106).

Then, the energy management system can learn the machine learning model using the difference between the actual data and the predicted data (S1108).

12 is a view for explaining an energy management system according to another embodiment.

A plurality of electronic devices 1232, 1234, 1236, 1238 may be installed in a specific area. The power consumption (energy usage) of the plurality of electronic devices 1232, 1234, 1236, and 1238 can be measured by the meter 1210.

Since the amount of power consumed by each of the plurality of electronic devices 1232, 1234, 1236 and 1238 is difficult to grasp through one meter 1210, the system 1200 can use the machine learning device 120 The amount of power used can be predicted.

The user terminal 1220 may control the operation of the plurality of electronic devices 1232, 1234, 1236, and 1238. For example, the user terminal 1220 may include a communication device for generating a remote radio control signal (e.g., a remote control signal) for some of the electronic devices, have.

As another example, the user terminal 1220 may control the smart plug attached to some electronic devices to control the on / off of the electronic devices.

When the user terminal 1220 controls a plurality of electronic devices 1232, 1234, 1236, and 1238, it may transmit control information about operation of the electronic device to the zone agent (ZEMA). For example, when a user terminal 1220 generates a remote radio control signal, the user terminal 1220 may transmit control information corresponding to the remote radio control information to the zone agent (ZEMA).

The zone agent ZEMA receives control information about the operation of the plurality of electronic devices 1232, 1234, 1236 and 1238 from the user terminal 1220 and controls the plurality of electronic devices 1232, 1234, 1236, 1238 can be generated. For example, the zone agent (ZEMA) can generate whether or not each electronic device 1232, 1234, 1236, and 1238 is operating with the operating status measurement data. As another example, the Zone Agent (ZEMA) may generate with the operating state actual data whether the power usage of each of the electronic devices 1232, 1234, 1236, 1238 is in an increasing state or a decreasing state. For example, the zone agent ZEMA can check whether the power consumption of the air conditioner 1236 is in an increasing state or a decreasing state, according to an increase or decrease in the air conditioner set value.

Zone agent (ZEMA) may receive power usage data of the zone from meter (1210), and may transmit power usage data and the above-mentioned operational state measured data to machine learning device (120).

The machine learning apparatus 120 receives operating state actual data and power consumption data from the zone agent ZEMA and supplies the plurality of electronic apparatuses 1232, 1234, 1236, and 1234 via a machine learning model that includes power usage data as input data. 1238, and can learn the machine learning model using the operating state error data according to the difference between the operating state actual data and the operating state prediction data.

When the machine learning reaches a certain level or more, the machine learning apparatus 120 predicts the operating state, operation pattern, power consumption, etc. of each electronic device 1232, 1234, 1236, and 1238 with high accuracy .

On the other hand, data input to the machine learning model may be data measured by a sensor or the like (for example, temperature data, meter data, etc.), or may be data generated by processing basic data generated by another apparatus.

13 is a diagram for explaining generation of input data of a machine learning model based on basic data.

Referring to FIG. 13, a plurality of input data may be input to the machine learning model (MLM). Since the performance of the machine learning model (MLM) may vary depending on the fitness of the input data, it is necessary to collect the input data having a high correlation with the predicted data as much as possible and input it into the machine learning model (MLM). For example, when the machine learning model (MLM) calculates the room temperature set value as the predicted data, the accuracy of the predicted data can be increased when the user location information is inputted into the machine learning model (MLM). It is necessary to set the operation degree of the cooling and heating machine to be different according to the case where the user is located close to the air conditioner or the air conditioner. When the user location information is inputted as input data, the machine learning model (MLM) The indoor temperature set value can be calculated in consideration of the position.

However, these input data, which can enhance the accuracy of the machine learning model (MLM), have only been generated through expensive sensors and equipment. For example, conventionally, the user location information is generated only through the user location tracking sensor and its control equipment. However, these sensors and equipment are expensive and therefore difficult to apply to popular systems.

The embodiment of the present invention may further include an input data estimation model (DEM: input data estimation model) for generating input data suitable for the machine learning model (MLM) using the basic data.

Referring to FIG. 13, the first input data, the second input data, and the third input data are input to the machine learning model (MLM). At this time, the third input data is data generated by the input data estimation model (DEM) using the basic data.

The input data estimation model (DEM) may be a program in which a pre-written algorithm is implemented or a computer device that drives the program. Alternatively, the input data estimation model (DEM) may be another machine learning model described with reference to FIG. 7 - for example, a recumbent numerical machine learning model (see 710 in FIG. 7).

14 shows examples of input / output data of the input data estimation model.

Referring to FIG. 14A, the first input data estimation model DEM1 receives user's terminal's wireless signal and can generate user location information or user's posture information.

The wireless signal of the user terminal may be, for example, a remote wireless control signal (infrared communication signal, WiFi signal, Bluetooth signal, etc.) The position can be calculated, and user position information can be generated according to these calculated values.

User posture information can also be generated by analyzing the pattern of the wireless signal of the user terminal. For example, by analyzing the type of distance change and the type of intensity conversion of the wireless signal, it is possible to determine whether the user is sitting or lying, or the user is further combined with the position information to estimate the sitting position when the user is positioned adjacent to the sofa, If it is located adjacent to the bed, it can be estimated to be lying down.

Referring to FIG. 14 (b), the second input data estimation model DEM2 may receive the setting value adjustment information and generate the user convenience index or the user satisfaction index.

The set value adjustment information may be, for example, temperature set value adjustment information for the cooler, and may be temperature set value adjustment information for the heater. Alternatively, the setting value adjustment information may be on / off control information for the lighting apparatus, or may be wind direction adjustment information of an air conditioner or the like.

The setting value adjustment information may further include details such as the number of adjustments, the frequency of adjustment, the adjustment tendency - for example, a tendency to raise the temperature, a tendency to decrease the temperature, and the like.

The setting value adjustment information may be generated at the user terminal. For example, control information for an in-zone electronic device received from a user terminal-for example, a communication terminal such as a smart phone-may be utilized as set value adjustment information.

The second input data estimation model DEM2 can generate the user convenience index or the user satisfaction index using the setting value adjustment information. For example, the second input data estimation model (DEM2) can generate the user convenience index or the user satisfaction index by lowering the user convenience index or the user satisfaction index if the adjustment frequency or the adjustment frequency is high.

Referring to (c) of FIG. 14, a plurality of input data estimation models may be combined in a complex manner.

For example, the generated value-user position information or user attitude information of the first input data estimation model DEM1 is input to the second input data estimation model DEM2, Index or user satisfaction index.

FIG. 15 is a diagram showing an example that can be applied to FIG. 13. FIG.

Referring to FIG. 15, humidity data, temperature data, CO2 data, and energy usage data are input as basic data to an input data estimation model (DEM) to generate the number of occupants.

Then, the occupant number data generated by the input data estimation model (DEM) can be used again as the input data of the machine learning model (MLM). The machine learning model (MLM) can input other occupant environment information and energy information as other input data.

The occupational environment information is, for example, occupational state information - personal load pattern of occupant, personal health information, personal position information, personal behavior information, and the like - and environmental information - temperature information, humidity information, CO2 information, - < / RTI > Here, the occupant state information and the like can be generated by another input data estimation model as described with reference to Fig.

The energy information may include, for example, common load information, personal usage information, and renewable energy information.

The machine learning model (MLM) can calculate the power consumption and user convenience index as prediction data by receiving the number of occupants, occupant environment information, and energy information.

Then, the error data is calculated while the predicted data and the actual data are compared, and the machine learning model (MLM) can be modified while the error data is fed back.

The embodiments of the present invention have been described above. According to this embodiment, the energy management system can be configured using the machine learning technique. In addition, according to this embodiment, the energy management system can be implemented by effectively reflecting the characteristics of residents, which are difficult to model mathematically through the machine learning technique. In addition, according to this embodiment, actual data can be generated in an energy management system using a machine learning technique to learn a machine learning model. Further, according to this embodiment, the machine learning model can be generated by generating the actual data even in a space without a suitable sensor, and the machine learning model can be learned by generating the actual data, which is difficult to generate through the sensor, .

It is to be understood that the terms "comprises", "comprising", or "having" as used in the foregoing description mean that the constituent element can be implanted unless specifically stated to the contrary, But should be construed as further including other elements. All terms, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined. Commonly used terms, such as predefined terms, should be interpreted to be consistent with the contextual meanings of the related art, and are not to be construed as ideal or overly formal, unless expressly defined to the contrary.

The foregoing description is merely illustrative of the technical idea of the present invention, and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the scope of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

Claims (15)

A sensor network installed in a specific area;
A zone agent for managing environmental data obtained from the sensor network;
A machine for receiving the environmental data from the zone agent, generating prediction data for energy management in the zone through a machine learning model including the environmental data as input data, and transmitting the prediction data to the zone agent A learning device,
Wherein the zone agent comprises:
Performing an energy management process for the zone using the prediction data, receiving user feedback from a registered user terminal corresponding to the zone, generating actual data using the user feedback, and transmitting the generated data to the machine learning device ,
The machine learning apparatus includes:
And the machine learning model is learned using error data generated according to a difference between the actual data and the predictive data. The method according to claim 1,
Wherein the prediction data includes a number of occupants,
Wherein the zone agent generates the actual data on the number of lost persons using the user feedback on the redundancy received from the user terminal. 3. The method of claim 2,
Wherein the sensor network comprises a temperature sensor and a CO2 sensor,
Wherein the machine learning device uses the measured values of the temperature sensor and the CO2 sensor to generate the prediction data for the number of occupants for the zone. The method according to claim 1,
Wherein the prediction data includes a user-friendliness index for the zone,
Wherein the zone agent generates the actual data for the user convenience index using the user feedback on user convenience received from the user terminal. The method according to claim 1,
Wherein the machine learning device receives the actual data from the zone agent until the machine learning model is learned over a certain level. 6. The method of claim 5,
Wherein the predetermined level is determined by a value corresponding to the size of the error data. 6. The method of claim 5,
Wherein the zone agent transmits a feedback request signal to the user terminal but does not transmit the feedback request signal when learning of the machine learning model ends. The method comprising: obtaining environmental data for the zone from a sensor network installed in a specific zone; generating prediction data for energy management in the zone through a machine learning model including the environmental data as input data;
Transmitting a feedback request signal to a user terminal registered corresponding to the zone, generating actual data corresponding to the predicted data using user feedback received from the user terminal, And
Learning the machine learning model using the difference between the actual data and the predictive data
Lt; / RTI > A meter for measuring the power usage of a particular area;
Receiving control information on operation of the plurality of electronic devices from a user terminal capable of controlling operation of a plurality of electronic devices installed in the zone, and outputting operating state actual data of the plurality of electronic devices in accordance with the control information A zone agent that generates and receives power usage data of the zone from the meter; And
Receiving the operating state actual data and the power consumption data from the zone agent and generating operating state prediction data for the plurality of electronic devices through a machine learning model including the power consumption data as input data, A machine learning device that learns the machine learning model using operation state error data corresponding to a difference between the actual condition data and the operation condition prediction data,
And an energy management system. 10. The method of claim 9,
Wherein the user terminal controls smart plugs attached to some of the plurality of electronic devices to control on / off of the some electronic devices. 10. The method of claim 9,
Wherein the user terminal comprises a communication device for generating a remote radio control signal for a part of the plurality of electronic devices, and when the remote radio control signal is generated, control information corresponding to the remote radio control signal To the zone agent. Obtaining environmental data for the zone from a sensor network installed in a specific zone;
Generating user data by using a user operation signal or a radio signal received from a registered user terminal corresponding to the zone;
Generating prediction data for energy management in the zone through a machine learning model including the environment data and the user data as input data;
Generating actual data corresponding to the predicted data; And
Learning the machine learning model using the difference between the actual data and the predictive data
Lt; / RTI > 13. The method of claim 12,
The user operation signal is a set value operation signal for the energy device located in the zone,
The user data includes a user convenience index or a user satisfaction index,
In the step of generating the user data,
And generating the user convenience index or the user satisfaction index according to a set value operation signal for the energy device. 13. The method of claim 12,
Wherein the user data includes user location information or user attitude information,
In the step of generating the user data,
And generating the user location information or the user attitude information using the radio signal of the user terminal. 15. The method of claim 14,
The user operation signal is a set value operation signal for the energy device located in the zone,
The user data further includes a user convenience index or a user satisfaction index,
In the step of generating the user data,
And generating the user convenience index or the user satisfaction index by further using the user location information or the user attitude information.

Images