KR20230174207A - Device for analyzing power demands and system comprising the same - Google Patents

Abstract

A power demand analysis device and a power management system including the same are provided. The power demand analysis device includes a processor and a storage unit storing a regression coefficient estimator and a classification variable determiner performed using the processor, wherein the regression coefficient estimator and the classification variable determiner are external and contain power usage data of the building and weather information. Data is provided and a regression tree model is created with at least one of the external data as a classification variable.

Description

Power demand analysis device and power management system including the same {Device for analyzing power demands and system comprising the same}

The present invention relates to a power demand analysis device and a power management system including the same. Specifically, it relates to a power demand analysis device that analyzes power demand using a quantile regression tree model, and a power management system including the same.

In order to efficiently operate power and reduce operating costs in buildings, factories, etc., accurate prediction of power consumption for buildings is necessary. Groups related to power consumption within a building (eg, heating appliances, lighting appliances, heating and cooling appliances, etc.) may have different power consumption patterns for each group. Accordingly, research is being conducted on technologies that can accurately predict power usage for buildings while considering these different power consumption patterns.

Republic of Korea Patent Publication No. 2015-0115063 (published on October 14, 2015)

The technical problem to be solved by the present invention is to provide a power demand analysis device capable of accurately predicting power consumption for buildings.

Another technical problem to be solved by the present invention is to provide a power management system that enables efficient power operation of buildings.

Another technical problem to be solved by the present invention is to provide a computer-readable recording medium storing a power demand analysis program that enables efficient power management of buildings.

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

A power demand analysis device according to some embodiments of the present invention for achieving the above technical problem includes a processor and a storage unit storing a regression coefficient estimator and a classification variable determiner performed using the processor, wherein the regression coefficient estimator and the classification variable are stored. The decision unit receives external data including building power usage data and weather information, and creates a regression tree model using at least one of the external data as a classification variable.

In an embodiment, the Regression Tree model may include a quantile regression tree model.

In an embodiment, the classification variable includes a continuous classification variable and a categorical classification variable, and the classification variable determiner determines the classification variable using different methods for the continuous classification variable and the categorical classification variable. You can decide.

In an embodiment, the classification variable determiner, once the classification variable is determined, may select a classification point for the determined classification variable.

In an embodiment, generating the regression tree model includes estimating quantile regression coefficients for a target node, selecting the classification variable for the target node, and selecting a classification point for the classification variable. can do.

In an embodiment, generating the regression tree model may further include repeating the quantile regression coefficient estimation, selecting the classification variable, and selecting the classification point until the target node satisfies a stopping condition. there is.

In an embodiment, the building's power usage data may be classified according to the day of the week, time of day, and the weather information and used to generate the regression tree model.

A power management system according to some embodiments of the present invention for achieving the other technical problems is provided with first external data including building power usage data and weather information, and classifies at least one of the first external data. It includes a power demand analysis device that generates a quantile regression tree model using variables, and a building energy management device that performs building energy management based on power demand prediction information predicted using the quantile regression tree model.

In some embodiments, the power demand analysis device is provided with second external data that is different from the first external data and includes weather information, and uses a quantile regression tree model generated based on the first external data. The power demand prediction information that matches second external data can be generated.

In some embodiments, the first external data and the second external data may be used separately in different time units.

In some embodiments, the building's power usage data may be provided for each group related to power consumption within the building.

A power management system according to some embodiments of the present invention for achieving the above other technical problems includes a power meter installed at the contact point where power consumption equipment and wiring in the building meet to measure power usage data of the building, and a power meter at a specific point in time. A building energy management device that controls the power consumption appliances based on power demand prediction information of the building, wherein the power demand prediction information of the building is based on external data including power usage data of the building and weather information. This is determined by creating a quantile regression tree model.

In some embodiments, the power usage data of the building includes power usage data of the building for a certain period of time in the past, and the external data includes first external data including weather information for a certain period of time in the past; It may include second external data including weather information at the specific point in time.

A computer-readable recording medium storing a power demand analysis program according to some embodiments of the present invention for achieving the other technical problem above includes a regression tree model based on external data including building power usage data and weather information. A power demand analysis program that generates, (a) estimating a regression coefficient for a target node, (b) selecting the classification variable for the target node, and (c) classifying the classification variable. It includes the step of selecting a point.

In some embodiments, estimating a regression coefficient for the target node includes estimating a quantile regression coefficient for the target node, and the regression tree model may include a quantile regression tree model.

In some embodiments, the step of repeating steps (a) to (c) may be further included until the target node satisfies a stop condition.

Specific details of other embodiments are included in the detailed description and drawings.

According to embodiments of the present invention, it is possible to accurately predict power consumption by group (e.g., heating appliances, lighting appliances, heating and cooling appliances, etc.) related to power consumption within the building, rather than the total power consumption for the entire building. Thus, efficient power operation of the building may be possible.

In addition, by using a bottom-up demand forecasting method that is different from the power company-centered top-down power demand forecasting method, the accuracy of predicting power usage can be improved.

Figure 1 is a diagram to explain problems with the power demand prediction model for buildings.
Figure 2 is a diagram to explain the problem of the power demand prediction model based on the power usage for the temperature of the building heating equipment.
Figure 3 is a diagram to explain the problems of the power demand prediction model based on the power usage relative to the temperature of building lighting equipment.
Figure 4 is a diagram to explain the problems of the power demand prediction model based on the power usage relative to the temperature of the building's heating and cooling equipment.
Figure 5 is a diagram to explain the problems of the power demand prediction model based on the power consumption by time period in relation to temperature in the building.
Figure 6 is a conceptual diagram of a power management system according to some embodiments of the present invention.
FIG. 7 is a block diagram of the power demand analysis device shown in FIG. 6.
FIG. 8 is a flowchart for explaining the operation of the power management system shown in FIG. 6.
FIG. 9 is a diagram illustrating an example of a quantile regression tree model generated by the power demand analysis device shown in FIG. 6.
FIG. 10 is an example of external data at a prediction time provided by the power demand analysis device shown in FIG. 6.
FIG. 11 is a diagram illustrating an example of power demand prediction performed by the power demand analysis device shown in FIG. 6.
FIG. 12 is a flowchart for explaining the operation of generating a quantile regression tree model of the power demand analysis device shown in FIG. 6.
FIG. 13 is a diagram for explaining the operation of the regression coefficient estimator shown in FIG. 7.
FIG. 14 is a diagram showing an exemplary installation of the power meter shown in FIG. 6.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and will be implemented in various different forms. The present embodiments only serve to ensure that the disclosure of the present invention is complete and that common knowledge in the technical field to which the present invention pertains is not limited. It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims. The sizes and relative sizes of components shown in the drawings may be exaggerated for clarity of explanation. Like reference numerals refer to like elements throughout the specification, and “and/or” includes each and all combinations of one or more of the referenced items.

The terminology used herein is for describing embodiments and is not intended to limit the invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context. As used in the specification, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other elements in addition to the mentioned elements.

Although first, second, etc. are used to describe various elements or components, these elements or components are of course not limited by these terms. These terms are merely used to distinguish one device or component from another device or component. Therefore, of course, the first element or component mentioned below may also be a second element or component within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings that can be commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless clearly specifically defined.

The detailed description that follows is not intended to be taken in a limiting sense, and the scope of the invention is limited only by the appended claims, together with all equivalents to what those claims assert, if properly described. Similar reference numbers in the drawings refer to identical or similar functions across various aspects.

Figure 1 is a diagram to explain problems with the power demand prediction model for buildings.

Figure 1 is a graph showing the relationship between the total hourly power consumption of a specific building over a year and the building's surrounding temperature. The X-axis of the graph represents temperature, the Y-axis represents power usage by hour, and each dot represents the distribution of total power usage by temperature and hour over a year. The solid line shown in the graph is the nonlinear regression estimator (RL) estimated using the LOWESS (LOcally WEighted Scatterplot Smoothing) method.

Referring to Figure 1, when the temperature is mild, such as the section marked A, the distribution of power usage is small, but when the temperature is very high or low, such as the section marked B or C, there is a large deviation in the building's hourly power usage. Able to know. In other words, it can be seen that there is heteroscedasticity in the building's hourly power usage.

Therefore, when simply generating a regression estimate (RL) for the hourly power consumption of a building, there is a problem in that it cannot be used to accurately predict the building's power demand. For example, the generated regression estimator (RL) is likely not representative of the total power consumption of the building in the section marked B.

Furthermore, this heteroskedasticity can be seen to mean that variables other than temperature must be considered when predicting a building's power usage, but the model in Figure 1 that does not take this into account can be considered to be able to accurately predict the building's power demand. I can't.

In addition, as shown in FIGS. 2 to 4, the power consumption of each group of groups related to power consumption in the building (e.g., electric heating equipment (FIG. 2), lighting equipment (FIG. 3), heating and cooling equipment (FIG. 4), etc.) The patterns are different, and are different for each day of the week and time as shown in FIG. 5, so there is a problem that the consumption pattern cannot be accurately predicted based on the total power consumption of the building alone, as shown in FIG. 1.

Hereinafter, a power demand analysis device and a power management system including the same according to embodiments of the present invention that can solve these problems will be described.

6 is a conceptual diagram of a power management system according to some embodiments of the present invention. FIG. 7 is a block diagram of the power demand analysis device shown in FIG. 6.

Referring to FIG. 6, the power management system is a building energy management device 400, and the building energy management device 400 includes a power demand analysis device 100 and a power meter 410.

According to one embodiment of the present invention, the building energy management device 400 is a real-time cloud server system and is always accessible through the network 300. However, the idea of the present invention is not limited to this, and it is possible to install the power demand analysis device 100 inside a building and implement a system using data generated inside the building.

The power demand analysis device 100 receives power usage data of the building 500 resulting from the power consumption device 510 and external data, for example, from the external data providing device 200, and models a quantile regression tree. (quantile regression tree) can be created.

And, based on the generated quantile regression tree model, the power demand analysis device 100 provides power demand prediction information that matches external data at the time power demand prediction is needed, to, for example, the building energy management device 400. can do.

In this embodiment, external data may include, for example, weather information. Specifically, external data may include, for example, temperature, humidity, wind speed, etc., but the technical idea of the present invention is not limited thereto. In addition, external data includes all factors that can affect power usage, such as public holidays, discomfort index, perceived temperature or weather conditions on the day (snow, rain, cloudy, clear, etc.), air pollution index, and light pollution index. It may include etc.

Referring to FIG. 7 , the power demand analysis device 100 may include a controller 110, a regression coefficient estimator 120, and a classification variable determiner 130. The controller 110 may control the operations of the regression coefficient estimator 120 and the classification variable determiner 130.

In some embodiments, the power demand analysis device 100 may be implemented as, for example, a server type computing system. In this case, the controller 110 may be implemented in the form of a processor that performs calculations, and the regression coefficient estimator 120 and classification variable determiner 130 may be implemented in the form of software, for example. .

In this way, when the regression coefficient estimator 120 and the classification variable determiner 130 are implemented in software form, the regression coefficient estimator 120 and the classification variable determiner 130 can be modularized and stored in the storage unit 140. That is, the regression coefficient estimator 120 and the classification variable determiner 130 can perform operations to be described later using the controller 110 implemented in the form of a processor.

Although the regression coefficient estimator 120 and the classification variable determiner 130 are shown separately in the drawings for convenience of explanation, the technical idea of the present invention is not limited thereto. If necessary, the regression coefficient estimator 120 and the classification variable determiner 130 may be implemented as one software module. Also, conversely, the regression coefficient estimator 120 and the classification variable determiner 130 may be implemented separately as software modules to implement more detailed functions as needed.

The regression coefficient estimator 120 may be provided with power usage data and external data of the building 500 and use them to estimate the quantile regression coefficient for the target node.

The classification variable determiner 130 may receive power usage data and external data of the building 500 and use them to determine a classification variable for the target node and further determine a classification point for the classification variable. Accordingly, in some embodiments, the classification variable determiner 130 may be implemented by being divided into a classification variable determination module and a classification point determination module.

More detailed operations of the regression coefficient estimator 120 and the classification variable determiner 130 will be described later.

Additionally, the power demand analysis device 100 may further include a control signal generator 150. The control signal generator 150 is linked to the power control device 520 in the building and can generate signals required for remote control of the power control device 520. The control signal generator 150 generates a signal for controlling the power-consuming device 510, such as turning off lighting or lowering the indoor temperature by controlling an air conditioner or electric heater.

The generated control signal is received by the power control device 520 of the building 500 through the network 300. The power control device 520 may control the power consuming device 510 based on the received control signal.

Referring again to FIG. 6, the building energy management device 400 may perform energy management for the building 500. In some embodiments, building energy management device 400 may include a power meter 410 that measures the amount of power used by power consuming appliances 510 in building 500 .

According to one embodiment of the present invention, the power meter 410 is a contact point (for example, an outlet, etc.) where the power consumption device 510 and the wiring in the building 500 meet, or the building 500 or the building 500. It may be installed at a point where power is supplied to each floor (for example, a distribution board).

Referring to FIG. 14, the power meter 410 may be installed within the distribution panel 1000.

In general, the distribution board 1000 is composed of a main circuit breaker 1010, a branch circuit breaker 1020, and a busbar 1030. In this case, the branch breaker 1020 is connected to one branch 1031 of the busbar 1030 that supplies power from the main breaker 1010.

In one embodiment of the present invention, the power meter 410 is installed instead of the branch breaker 1020 at a location where the branch breaker 1020 can be installed, and the current sensor 1040 is installed at the main breaker 1010 for precise power measurement. ) is installed on the conductor connected to the However, the technical idea of the present invention is not limited to the above embodiment, and the power meter 410 is installed at various locations, for example, the location of the main breaker 1010 or the power consumption device 510 such as an outlet and the building. It can also be installed at the point where wiring meets.

Referring again to FIG. 6 , the power consumption devices 510 may be classified into several groups, such as a heating device group, a lighting device group, an air conditioning device group, etc. At this time, the power meter 410 may be placed separately for each group to measure the power usage of each group.

For example, the first power meter 410 measures the amount of power used by a group of heating devices, the second power meter 410 measures the amount of power used by a group of lighting devices, and the third power meter 410 measures the amount of power used by a group of heating and cooling devices. You can measure the amount of power used by the group. The power usage for each group measured in this way can be provided to the power demand analysis device 100.

According to one embodiment of the present invention, the building energy management device 400 is a real-time cloud system that is always accessible. Accordingly, the power usage for each group measured by the power meter 410 may be transmitted to the power demand analysis device 100 that is physically distant through the network 300.

In addition, the building energy management device 400 receives groups related to power consumption in the building 500 (e.g., electric heating devices, lighting devices, heating and cooling devices, etc.) from the power demand analysis device 100 through the network 300. ) It is possible to receive power usage prediction information for each star and provide it to the user terminal 600, etc.

Here, the user terminal 600 may include all terminals that provide various types of displays, such as wearable devices such as mobile phones, tablet computers, PCs, PDAs, and smart watches.

Although not shown in detail in the drawing, the building energy management device 400 provides an interface, for example, a graphical user interface, for providing various information about the power usage of the building 500 to the user terminal 600. It may include a web server, etc.

In addition, the building energy management device 400 is linked to the power control device 520 in the building to obtain a group (e.g., , electric heating devices, lighting devices, heating and cooling devices, etc.), it is possible to receive power usage prediction information and control the power consumption of the power consumption device 510 of the building 500 based on this information.

Meanwhile, in some other embodiments, the building energy management device 400 may be directly connected to the power demand analysis device 100 and the network 300 without going through them, or may be integrated and implemented as a single system. That is, the technical idea of the present invention is not limited to the details shown.

The external data providing device 200 may include, for example, an external data providing server. Examples of such an external data providing device 200 include a server that can be accessed using an API provided by the Korea Meteorological Administration, a server that can be accessed using a calendar API, etc. However, the technical idea of the present invention is also limited thereto. It is not limited.

External data including, for example, weather information, etc. provided from the external data providing device 200 to the power demand analysis device 100 may be transmitted through the network 300 as shown. However, the technical idea of the present invention is not limited thereto, and if necessary, the external data provision device 200 and the power demand analysis device 100 may be integrated and implemented as one, or the external data provision device 200 and the power demand analysis device 100 may be integrated into one. The demand analysis device 100 may be connected directly without going through the network 300.

Hereinafter, with reference to FIGS. 6 and 8 to 11, the operation of the power management system according to some embodiments of the present invention will be described.

FIG. 8 is a flowchart for explaining the operation of the power management system shown in FIG. 6. FIG. 9 is a diagram illustrating an example of a quantile regression tree model generated by the power demand analysis device shown in FIG. 6. FIG. 10 is an example of external data at a prediction time provided by the power demand analysis device shown in FIG. 6. FIG. 11 is a diagram illustrating an example of power demand prediction performed by the power demand analysis device shown in FIG. 6.

Referring to Figure 8, a quantile regression tree model is created (S100). Specifically, for example, the power demand analysis device 100 may receive external data including power usage data of the building 500 and weather information and generate a quantile regression tree model.

Here, power usage data of the building 500 may be provided from, for example, the building energy management device 400. Specifically, power usage data of the building 500 may be provided, for example, from the power meter 410 included in the building energy management device 400. More specifically, the power meter 410 measures power usage by groups related to power consumption (e.g., electric heating devices, lighting devices, heating and cooling devices, etc.) within the building 500, and measures the power usage by the power demand analysis device (100). ), but the technical idea of the present invention is not limited thereto.

In some embodiments, the power demand analysis device 100 may store the received power usage data separately, for example, in 15-minute increments, in order to create a quantile regression tree model. Additionally, in some other embodiments, power usage data provided to the power demand analysis device 100 may be provided separately, for example, in 15-minute increments. Additionally, in some other embodiments, power usage data provided to the power demand analysis device 100 may be provided separately, for example, on an hourly basis. Additionally, in some other embodiments, the power usage data provided to the power demand analysis device 100 may be stored divided by day of the week, for example.

External data provided to the power demand analysis device 100 may be provided, for example, from the external data providing device 200. Such external data may include, for example, information such as temperature (observed temperature), humidity, wind speed, expected precipitation, and sky conditions, but the technical idea of the present invention is not limited thereto.

In some embodiments, such weather information may be provided on a 1-hour or 3-hour basis, for example. When the external data is provided in this way on a 1-hour or 3-hour basis, the power demand analysis device 100 uses the provided external data, for example, by interpolation, in 15-minute units to create a quantile regression tree model. It can be saved separately. In some other embodiments, external data provided to the power demand analysis device 100 may be stored separately in units of, for example, one hour. In some other embodiments, external data provided to the power demand analysis device 100 may be stored separately by, for example, days of the week.

Here, it is assumed that the power demand analysis device 100 receives external data including power usage data of the building 500 and weather information and generates the quantile regression tree model shown in FIG. 9, and continues the explanation. do. That is, the power demand analysis device 100 may, for example, use power usage data for the building 500 from December 1, 2014 to May 31, 2015, and external data including weather information for the same period. Based on this, we will continue the explanation assuming that the quantile regression tree model shown in Figure 9 has been created.

Referring to FIG. 9, the quantile regression tree model may include a plurality of nodes (1 to 11). A plurality of nodes (1 to 11) may have a parent-child relationship with each other. For example, node 1 may be the parent node of node 2 and node 9, and node 2 and node 9 may be child nodes of node 1. The same relationship holds true for other nodes.

A node without a parent (1) can be a root node, and a node without children (4, 6, 7, 8, 10, 11) can be a leaf node.

The total amount of usage data included in node 1 can be divided into node 2 and node 9 by being divided by a partition variable called temperature. Specifically, node 2 includes power usage data with a temperature of 14.1°C or lower among the power usage data included in node 1, and node 9 includes power usage data with a temperature of 14.1°C or lower among the power usage data included in node 1. May include power usage data exceeding ℃.

In addition, node 3 includes power usage data with a temperature of 5°C or lower among the power usage data included in node 2, and node 8 includes power usage data with a temperature of 5°C or lower among the power usage data included in node 2. May include excess power usage data.

In addition, node 4 includes power usage data with a humidity of 48% or less among the power usage data included in node 3, and node 5 includes power usage data with a humidity of 48% or less among the power usage data included in node 3. May include excess power usage data.

In addition, node 6 includes power usage data with a wind speed of 2.1 m/s or less among the power usage data included in node 5, and node 7 includes power usage data with a wind speed of 2.1 m/s or less among the power usage data included in node 5. May include power usage data exceeding 2.1 m/s.

In addition, the node 10 includes power usage data with a humidity of 86% or less among the power usage data included in the node 9, and the node 11 includes power usage data with a humidity of 86% or less among the power usage data included in the node 9. May include excess power usage data.

Once all leaf nodes (4, 6, 7, 8, 10, 11) without children are determined, quantile regression is performed on the power usage data contained in each leaf node (4, 6, 7, 8, 10, 11). The coefficient can be estimated.

The specific operation of generating a quantile regression tree model based on the power usage data of the building 500 and external data collected in this way will be described later.

Next, referring again to FIG. 8, data necessary for prediction is obtained (S200). Specifically, for example, the power demand analysis device 100 may receive external data including weather information at a time when prediction is needed, for example, from the external data providing device 200.

Even in this case, when the provided external data is divided into 1-hour or 3-hour units, the power demand analysis device 100 can store the provided external data by dividing them into 1-hour units using, for example, an interpolation method. there is. Additionally, the provided external data can be stored separately by day of the week. Alternatively, the provided external data can be divided and stored by time. Figure 10 shows an example of external data provided from the external data providing device 200, but the technical idea of the present invention is not limited thereto.

For example, if the time when power usage prediction is needed is 15:30 on October 15, 2015, the power demand analysis device 100 sends external data including weather information at the time of prediction to the external data providing device 200. It can be provided from . At this time, we will continue the explanation assuming that the temperature at the time when prediction is needed is expected to be 18℃ and humidity is expected to be 50%.

Meanwhile, in FIG. 8, for convenience of explanation, an operation (S200) of obtaining data necessary for prediction is shown after an operation of generating a quantile regression tree model (S100). However, the technical idea of the present invention is limited to the matters shown. That is not the case. In some other embodiments, the operation of obtaining data necessary for prediction (S200) may be performed before the operation of generating a quantile regression tree model (S100), and the operation of obtaining data necessary for prediction (S200) may be performed before the operation of generating a quantile regression tree model (S100). It may also be performed in the same manner as the operation to create a model (S100).

Next, referring again to FIG. 8, power demand prediction is performed (S300). Specifically, the power demand analysis device 100 generates leaf nodes (4, 6, 7, 8, 10, 11) can be selected.

Here, since the temperature at the time when prediction is required is expected to be 18°C and humidity is expected to be 50%, node 10 can be selected from leaf nodes (4, 6, 7, 8, 10, and 11). Also, power demand prediction can be performed using the regression estimator of the node 10.

Specifically, referring to FIG. 11, at 15:30 on October 15, 2015, when power usage prediction is needed, using the quantile regression estimator (QRL) estimated based on the power usage data included in the node 10. It can be seen that 98Wh of power is predicted to be used per minute.

Next, referring to FIG. 8 again, information may be provided to the user terminal 600 or power consumption mechanisms may be controlled based on the power demand prediction information (S400).

For example, the power demand analysis device 100 may provide this predicted power use information to the building energy management device 400, and the building energy management device 400 may provide this to the user terminal 600.

Additionally, the building energy management device 400 may control the power consumption of the power consumption device 510 based on the provided power use prediction information. For example, when considering the appropriate indoor temperature, the building energy management device 400 is a power consumption device to minimize the power usage of the building 500 by reducing the power of the air conditioner by about 5% and the use of lights by about 10%. A control signal capable of controlling (510) can be generated.

In some embodiments, the building energy management device may predict peak times of power consumed by the power consuming appliance 510. A detailed description of this is disclosed in Republic of Korea Patent Application No. 10-2015-0122845, and the disclosure may be incorporated herein by reference.

In this way, in the case of the power management system according to embodiments of the present invention, accurate power consumption prediction for the building 500 may be possible by generating and using the quantile regression tree model. Additionally, by controlling the power consumption mechanism 510 of the building 500 based on this, efficient power operation of the building 500 may be possible.

Hereinafter, with reference to FIGS. 6, 9, and 12 to 13, the operation of the power demand analysis device 100 to generate a quantile regression tree model (S100 in FIG. 4) will be described in more detail.

FIG. 12 is a flowchart for explaining the operation of generating a quantile regression tree model of the power demand analysis device shown in FIG. 6. FIG. 13 is a diagram for explaining the operation of the regression coefficient estimator shown in FIG. 7.

Referring to Figure 12, data necessary for learning the power demand model is provided (S110).

For example, the power demand analysis device 100 receives power usage data of the building 500 from the building energy management device 400 and receives external data including weather information from the external data providing device 200. It can be provided. Additionally, the power demand analysis device 100 may store the provided data in a form necessary for generating a quantile regression tree model.

In some embodiments, the power usage data for building 500 may include power usage data for building 500 for a sufficient period of time (e.g., three months or more) immediately prior to the power usage forecast date for building 500. there is.

In addition, external data during the same period may also include weather information for a sufficient period (for example, 3 months or more) immediately before the power usage forecast date of the building 500.

If weather information is provided in 1-hour or 3-hour increments, such weather information can be divided into 15-minute increments and stored, for example, using interpolation.

In some embodiments, power usage data may be stored separately by day of the week. Figure 12 shows an example of data provided to the power demand analysis device 100 to generate a quantile regression tree model, but the technical idea of the present invention is not limited to this example.

Referring again to FIG. 12, the quantile regression coefficient is estimated for the target node (S120).

Before explaining the following process, terms are first defined as follows.

Y: Response variable (power usage per 15 minutes defined in Euclidean space)

X: Fitted variable (used in the quantile regression model at each node) dimensional explanatory variable (e.g. 1-dimensional temperature)

Z: Split variable (d-dimensional explanatory variable used for data classification, such as weather information and day of the week information, etc.)

t: any node in the model, (e.g. t=1 for the root node)

& : Child nodes of node t (for example, in Figure 5, t _L of node 1 is node 2, t _R is node 9)

N: Total number of observations of root nodes (total number of samples subject to analysis)

: Number of observations at node t (number of samples of that node)

: regression estimator for quantile τ at node t

First, the regression coefficient estimator (120 in FIG. 7) of the power demand analysis device 100 may estimate the quantile regression coefficient for node 1 in FIG. 9. Specifically, in some embodiments, the regression coefficient estimator (120 in FIG. 7) of the power demand analysis apparatus 100 may estimate a quantile regression coefficient for power usage data included in the node 1.

Specifically, the regression coefficient estimator 120 calculates the quantile regression coefficient at node 1. can be estimated.

The 100τ% conditional quantile function of Y given X for the p-dimensional fit variable It can be defined as follows.

<Equation 1>

Here, the linear quantile function can be defined as <Equation 2> below.

<Equation 2>

=

The objective function may be as shown in <Equation 3> below, and the check loss function may be as in <Equation 4>.

<Equation 3>

<Equation 4>

For example, quantiles If is determined to be 0.1, the quantile regression coefficient can be ② in Figure 13, and the quantile If is determined to be 0.5, the quantile regression coefficient can be ① in Figure 13 and the quantile If is determined to be 0.9, the quantile regression coefficient can be ③ in Figure 13.

Referring again to FIG. 12, it is determined whether the target node satisfies the stop condition (S130).

Specifically, it is determined whether the target node satisfies one of the rules below.

Rule 1: Sample size of node t is small enough to not fit the data.

(For example, small enough might mean 5% of the sample size, while if the sample size is large you might use 1%)

Rule 2: When the model explanatory power of node t is large enough to sufficiently explain the data

(For example, if the coefficient of determination (R-squared) value (range: 0~1), which is a measure of explanatory power, is 0.8 or higher, or an empirical probability distribution can be used)

Rule 3: When the significance probability of the classification variable selected at node t is greater than the predefined significance level α

As a result of the judgment, if node t satisfies the stopping condition, the creation of the quantile regression tree model is completed. If node t does not satisfy the stopping condition, the next step is performed.

Since node 1 does not satisfy the stopping condition, the next step is performed.

Referring again to FIG. 12, a classification variable for the target node is selected (S140). For example, the classification variable determiner 130 of the power demand analysis device 100 determines all the division variables A stability test of the regression coefficient can be performed, and the most unstable partition variable among these can be designated as the data classification variable.

Specifically, the classification variable determiner 130 determines all split variables that can be split at node t through the following process. About By testing the volatility of , the most volatile partition variable Z _j can be designated as the classification variable for the node. (Here, Z _j may include weather conditions such as humidity and wind speed, or other status information such as discomfort index, perceived temperature, weather conditions on the day, or whether it is a public holiday.)

Random split variable using the score-based fluctuation test proposed by Zeileis and Hornik (2007) The degree of volatility can be measured using the empirical fluctuation process of <Equation 5> below.

<Equation 5>

(here, is the score function of the objective function, Is of the second split variable second observation, is the outer-product-of gradient estimates

And since W _j (t) converges to the k-dimensional Brownian bridge under the null hypothesis of the volatility test, the partition variable with the smallest significance probability means the variable with the largest variation at node t. You can.

Growing the tree structure by selecting the most significant variable in the volatility test has the advantage of dividing the entire data into heterogeneous data between nodes and homogeneous data within the node.

The significance probability is calculated using a test statistic, and the test statistic for continuous partition variables (e.g. humidity, etc.) is calculated using the supLM statistic of Andrew (Andrew (1993)) <Equation 6> It can be calculated as follows.

<Equation 6>

.

( Since the limiting distribution of is a distribution for the upper limit of the tied-down Bessel process, the probability of significance can be calculated)

If the jth split variable is categorical (i.e., categorical split variable go If there are categories (for example, whether it is a public holiday, etc.), it can be calculated using the chi-square statistic as shown in <Equation 7>.

<Equation 7>

(here Is Increment of experience variation process in the second category, The limiting distribution of Since it is a chi-square distribution with degrees of freedom, significance probability can be calculated)

Through this process, the split variable with the lowest significance probability among all split variables can be designated as the classification variable of node t.

Here, the classification variable determiner 130 according to this embodiment can select classification variables separately depending on whether the division variable is a continuous division variable or a categorical classification variable.

Referring again to FIG. 12, a classification point for a classification variable is selected (S150).

For example, the classification variable determiner 130 of the power demand analysis device 100 refers to node t as a child node. class In order to divide, the data classification point can be selected using local optimization methodology.

Here, two child nodes are used for data interpretation and calculation convenience. class The focus is on a classification point selection algorithm based on binary segmentation for division, but the technical idea of the present invention is not limited to this.

The error function at node t can be defined as shown in Equation 8 below.

<Equation 8>

And, node t and child nodes as in <Equation 9> class The amount of reduction in the error function between The classification rule that shows the greatest error reduction by evaluating (continuous classification variable) or (Categorical classification variable) can be selected.

<Equation 9>

To do this, we explore all possible cases to find child nodes class A candidate classification point C or a classification set A can be formed.

Referring again to FIG. 12, when the classification variable and classification point are selected in this way, the target node can be divided into child nodes and the child nodes can be set as target nodes. (S160). Then, the processes from S120 to S160 can be repeated until the target node satisfies the stop condition (S130).

At least some of the embodiments of the present invention described above may be implemented as computer-readable code on a computer-readable recording medium. For example, the operation of the power demand analysis device 100 shown in FIG. 6 to generate a quantile regression tree model (S110 to S160 of FIG. 8) is implemented as computer-readable code on a computer-readable recording medium. It may be possible to do so.

Computer-readable recording media may include all types of recording devices that store data that can be read by a computer system.

Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage devices, and can also be implemented in the form of a carrier wave (e.g., transmitted via the Internet). It can be included.

In addition, computer-readable recording media are distributed and implemented in computer systems connected to a network, so that computer-readable codes can be stored and executed in a distributed manner.

Although embodiments of the present invention have been described above with reference to the attached drawings, the present invention is not limited to the above embodiments and can be manufactured in various different forms, and can be manufactured in various different forms by those skilled in the art. It will be understood by those who understand that the present invention can be implemented in other specific forms without changing its technical spirit or essential features. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive.

100: Power demand analysis device
200: External data provision device
300: Network
400: Building energy management device

Claims (16)

processor; and
A storage unit storing a regression coefficient estimator and a classification variable determiner performed using the processor,
The regression coefficient estimator and classification variable determiner are,
A power demand analysis device that receives external data including building power usage data and weather information and generates a regression tree model using at least one of the external data as a classification variable. According to clause 1,
The regression tree model is a power demand analysis device including a quantile regression tree model. According to clause 1,
The classification variables include continuous classification variables and categorical classification variables,
The classification variable determiner is a power demand analysis device that determines the classification variable using different methods for the continuous classification variable and the categorical classification variable. According to clause 3,
The classification variable determiner is a power demand analysis device that, when the classification variable is determined, selects a classification point for the determined classification variable. According to clause 1,
Creating the regression tree model involves:
estimating the quantile regression coefficient for the target node;
selecting the classification variable for the target node;
Power demand analysis device comprising selecting a classification point for the classification variable. According to clause 5,
Creating the regression tree model involves:
Power demand analysis device further comprising repeating the quantile regression coefficient estimation, selecting the classification variable, and selecting the classification point until the target node satisfies a stopping condition. According to clause 1,
A power demand analysis device in which the power consumption data of the building is classified according to the day of the week, time, and the weather information and used to generate the regression tree model. A power demand analysis device that receives first external data including building power usage data and weather information and generates a quantile regression tree model using at least one of the first external data as a classification variable; and
A power management system including a building energy management device that performs building energy management based on power demand prediction information predicted using the quantile regression tree model. According to clause 8,
The power demand analysis device receives second external data that is different from the first external data and includes weather information, and uses a quantile regression tree model generated based on the first external data to determine the second external data. A power management system that generates the corresponding power demand prediction information. According to clause 9,
A power management system in which the first external data and the second external data are used separately in different time units. According to clause 8,
A power management system in which power usage data of the building is provided for each group related to power consumption within the building. A power meter installed at the contact point where power consumption equipment and wiring within the building meet to measure power usage data of the building; and
It includes a building energy management device that controls the power consumption devices based on predicted power demand information of the building at a specific point in time,
A power management system in which the building's power demand prediction information is determined by generating a quantile regression tree model based on external data including the building's power usage data and weather information. According to clause 12,
The power usage data of the building includes power usage data of the building over a certain period of time in the past,
The external data includes first external data including weather information for a certain period of time in the past, and second external data including weather information at the specific point in time. A power demand analysis program that generates a regression tree model based on external data including building power usage data and weather information,
(a) estimating the regression coefficient for the target node,
(b) selecting a classification variable for the target node,
(c) A computer-readable recording medium storing a power demand analysis program including the step of selecting a classification point for the classification variable. According to clause 14,
The step of estimating a regression coefficient for the target node includes estimating a quantile regression coefficient for the target node,
The regression tree model is a computer-readable recording medium storing a power demand analysis program including a quantile regression tree model. According to clause 14,
A computer-readable recording medium storing a power demand analysis program, further comprising repeating steps (a) to (c) until the target node satisfies a stop condition.