WO2020235150A1

WO2020235150A1 - Power consumption prediction device

Info

Publication number: WO2020235150A1
Application number: PCT/JP2020/005099
Authority: WO
Inventors: 玄太吉村; 裕希川野; 浩之安田; 利宏妻鹿; 智祐成井; 修一村山
Original assignee: 三菱電機株式会社
Priority date: 2019-05-21
Filing date: 2020-02-10
Publication date: 2020-11-26
Also published as: JP2020190876A; JP7277254B2

Abstract

A typical pattern extraction unit (102) extracts, on a per electrical device basis, on period typical patterns compiled on the basis of past on times and off times for each of the electrical devices. A co-occurrence group extraction unit (104) extracts an on period typical pattern which is co-occurrent with a prescribed on period typical pattern of a prescribed electrical device and groups that on period typical pattern with the prescribed on period typical pattern into one co-occurrence group. A co-occurrence group occurrence prediction unit (106) carries out an occurrence prediction for each of the co-occurrence groups. A power consumption prediction unit (108) derives the power consumption for each of the co-occurrence groups and predicts the power consumption of a facility on the basis of the power consumption of the co-occurrence group which is predicted to occur.

Description

Power consumption predictor

The present invention relates to a power consumption prediction device.

For example, in a facility such as a building, a power management system (EMS, Energy Management System) that monitors and controls the power consumption of electrical equipment installed in the facility is used.

For example, one of the functions of the power management system is power consumption prediction. The power consumption forecast is also called a power demand forecast on the supply side of a power plant or the like. For example, the transition (time change) of the power consumption of the entire facility is predicted the next day or several days later.

For example, if a power consumption forecast is made in advance at a predetermined facility, if a storage battery is installed in the facility, the charging / discharging timing of the storage battery can be appropriately set. For example, the storage battery is discharged at a time when the power consumption of the entire facility is predicted to exceed the contract power. In addition, the storage battery will be charged during the time when the power consumption of the entire facility is predicted to be much lower than the contracted power.

Further, for example, in the so-called demand response (DR), it is necessary to make a decision in advance whether to participate in the demand response (Opt-In) or not (Opt-Out).
Even in this case, if the power consumption forecast is made in advance, it is possible to make a rational decision as to which of Opt-In and Opt-Out should be selected.

For example, in Patent Document 1, a consumption pattern determined for each electric device is used in predicting the power consumption of the entire building. The consumption pattern is obtained by clustering the operation results showing the transition of the power consumption of each electric device.

Further, for example, Patent Document 2 discloses a load prediction device that predicts the load of a device. In this device, a load pattern indicating a transition of a load for a predetermined period acquired by measuring a load to be managed at a predetermined time is obtained, and a degree of similarity between a plurality of load patterns is obtained. The degree of similarity is obtained based on the actual load in the load pattern, that is, the transition of power consumption of each electric device.

Further, for example, Patent Document 3 discloses a demand forecasting system for an electric power plant or the like. This system is provided with a function to extract a pattern that seems appropriate from the past actual values saved in the database and the demand pattern registered in advance by the user, and set this as the demand forecast value. ..

Japanese Unexamined Patent Publication No. 2011-176984 Japanese Unexamined Patent Publication No. 2017-21497 Japanese Unexamined Patent Publication No. 2004-164388

By the way, when predicting power consumption, the number of input points for prediction calculation is reduced by predicting power consumption by grouping a plurality of electric devices into one group. That is, it is possible to reduce the calculation load in the prediction calculation. Here, when a plurality of electric devices are grouped based on the transition of power consumption for each electric device, the power consumption changes depending on, for example, the number of years of operation of the electric device (in other words, depending on the degree of deterioration). Further, for example, in an air conditioner, the power consumption differs between the air conditioner installed on the south side and the air conditioner installed on the north side. In this way, if grouping is performed based on power consumption, it may be difficult to include many electric devices in one group.

Therefore, an object of the present invention is to provide a power consumption prediction device capable of reducing the calculation load at the time of power consumption prediction as compared with the conventional one while suppressing a decrease in the accuracy of power consumption prediction.

The present invention relates to a power consumption prediction device that predicts the power consumption of a facility in which a plurality of electric devices are installed. The device includes a typical pattern extraction unit, a co-occurrence group extraction unit, a co-occurrence group generation prediction unit, and a power consumption prediction unit. The typical pattern extraction unit extracts a typical pattern for an on-period based on the past on-time and off-time of each electric device for each electric device. The co-occurrence group extraction unit extracts an on-period typical pattern having co-occurrence that occurs together with a predetermined on-period typical pattern of a predetermined electric device, and combines the above-mentioned predetermined on-period typical pattern into one co-occurrence group. Summarize. The co-occurrence group occurrence prediction unit predicts the occurrence of each co-occurrence group. The power consumption prediction unit obtains the power consumption of each co-occurrence group and predicts the power consumption of the facility based on the power consumption of the co-occurrence group predicted to occur.

According to the above configuration, the typical pattern of the on-period is grouped based on the on-time and the off-time of the electric device, and a plurality of co-occurrence typical patterns of the on-period are grouped into the co-occurrence group.
In this way, since each electric device is grouped only by the on / off information of the electric device, filtering based on the power consumption is eliminated as compared with the case where the power consumption change is added to the on / off information. Therefore, it is possible to increase the number of electrical devices that can be put together as one group. In addition, by grouping a plurality of electric devices so that they can be regarded as a single electric device co-occurring, it is possible to suppress a decrease in the accuracy of power consumption prediction.

Further, in the above invention, the power consumption prediction device may include a display unit. The display unit displays the on period from the on time to the off time of each co-occurrence group predicted to occur and the facility power consumption which is the sum of the power consumption of each co-occurrence group in the on period. Will be done. Further, the display unit may be able to display the names of the electric devices grouped together as the co-occurrence group for each co-occurrence group.

Further, in the above invention, the display unit may be able to display the distribution of the on time and the off time summarized as a single on-period typical pattern.

Further, in the above invention, the typical pattern extraction unit may extract the typical pattern of the on-period of each electric device from the past on-time and off-time by unsupervised learning.

Further, in the above invention, the co-occurrence group extraction unit may extract a plurality of on-period typical patterns having co-occurrence by unsupervised learning and combine them into a single co-occurrence group.

Further, in the above invention, the power consumption prediction unit may predict the co-occurrence group that occurs on a predetermined day from the past electrical equipment operation history by supervised learning.

Further, in the above invention, an input unit that can input a correction instruction for the on period of each on period typical pattern displayed on the display unit may be provided.

Further, in the above invention, the input unit may be able to input the integration instruction for a plurality of co-occurrence groups displayed on the display unit.

Further, in the above invention, a prediction result correction unit may be provided to correct the occurrence prediction for each co-occurrence group based on the control plan set for each electric device.

According to the present invention, it is possible to reduce the calculation load at the time of power consumption prediction as compared with the conventional case, while suppressing the decrease in the accuracy of power consumption prediction.

It is a figure which illustrates the electric power system diagram of the facility including the power consumption prediction apparatus which concerns on this embodiment. It is a figure which illustrates the functional block of the power consumption prediction apparatus. It is a figure which illustrates the power consumption prediction flow. It is a time chart which exemplifies the on-time / off-time of a predetermined electric device by day. It is a figure which illustrates the extraction process of a typical pattern in an on-period. It is a figure which illustrates the extraction process of the co-occurrence group. It is a figure which exemplifies the on-period typical pattern classified by the co-occurrence group, and the power consumption forecast of a facility, which are displayed on the display part. It is a figure which shows the display example of the display part which displayed the distribution of the operation state of an electric device in a predetermined on-period typical pattern in addition to the on-period typical pattern and the power consumption forecast of a facility. It is a daily chart showing the distribution of co-occurrence groups and the power consumption forecast of facilities. It is a figure which shows the 1st example of the power consumption prediction flow. It is a figure which illustrates the correction work (time adjustment) of a typical pattern of an on-period. It is a figure which illustrates the correction work (pattern deletion) of a typical pattern in an on-period. It is a figure which exemplifies the correction work (group integration) of a co-occurrence group. It is a figure which shows the 2nd example of the power consumption prediction flow. It is a figure which shows the 3rd alternative example of the power consumption prediction flow.

Hereinafter, in order to explain the present invention in more detail, a mode for carrying out the present invention will be described with reference to the accompanying drawings.
FIG. 1 illustrates an electric power system of a facility including the power consumption prediction device 100 according to the present embodiment. This facility is a multi-story building such as a building, and a plurality of electric devices are installed in the facility. It should be noted that these electric devices may be referred to as electric facilities as if they were installed in the facility. As will be described later, the power consumption prediction device 100 predicts the power consumption of the entire facility.

These electrical devices (electrical equipment) are monitored and controlled by the BEMS (Building and Energy Management System), which is a power management system.

The power management system includes a power consumption prediction device 100, a power management device 10 (B-OWS), sub-controllers 14A to 14C (B-BC),

digital controllers

16A and 16B (DDC), and a remote station 18 (RS). These are connected to the bus. The

digital controllers

16A and 16B and the remote station 18 are connected to the electric devices 20A to 20D and various sensors 22A to 22F.

Although the power management device 10 and the power consumption prediction device 100 are both shown as independent devices in FIG. 1, they may be integrated into one device (for example, a computer).
Further, FIG. 1 illustrates a part of devices such as the sub controller 14 connected to the lower level of the power management device 10 due to space limitations, and various devices are connected in addition to the illustrated configuration. You may be.

Electrical equipment 20A to 20D are various equipment installed in a building, and include, for example, lighting equipment, air conditioning equipment, elevators, sanitary equipment, disaster prevention equipment, crime prevention equipment, and the like. In the example of FIG. 1, the electric device 20A is a lighting device, the electric device 20B is a lighting operation panel, the electric device 20C is an air conditioner, and the electric device 20D is an elevator control panel.

Further, as sensors placed under the power management device 10, an illuminance sensor 22A, an illumination power meter 22B, an air conditioner sensor 22C, an air conditioner power meter 22D, an elevator power meter 22E, and a high pressure power meter 22F are provided. The high-voltage watt-hour meter 22F is provided in, for example, a high-voltage power receiving facility supplied from an electric power company to a facility (building), and the power consumption of the entire facility is measured.

The power management device 10 receives each measured value from various sensors 22A to 22E under its control. Further, in the power management device 10, for example, on / off signals of various electric devices 20A to 20D are transmitted from the sub-controllers 14A to 14C. As will be described later, on / off signals of the various electric devices 20A to 20D are used in predicting the power consumption.

The power management device 10 is composed of, for example, a so-called B-OWS (BACnet Operator Workstation), and has a function as a client PC whose operation is monitored by an administrator or the like and a function as a server which performs data storage, application processing, and the like. I have. In the power management device 10, for example, screen display and setting operations are performed.

The sub controller 14 mainly plays a control function. The sub controller 14 is composed of, for example, a so-called B-BC (Building Controller), communicates with a terminal transmission device such as a digital controller 16 or a remote station 18, and manages point data, schedule control, and the like. For example, one sub-controller 14 is provided for each functional system (subsystem) such as an air-conditioning equipment system, a lighting equipment system, an elevator system, a sanitary equipment system, and a crime prevention equipment system.

The digital controller 16 may be a so-called DDC (Digital Digital Controller), and has a function as a controller for realizing distributed control in BEMS. For example, the digital controller 16 controls the connected

electric devices

20C and 20D by program control based on the schedule setting sent from the sub controller 14 and feedback control based on the target value also sent from the sub controller 14. Further, the digital controller 16 transmits the measured values of the sensors 22C to 22E, warnings of the

electric devices

20C and 20D, and the like to the above system and other digital controllers 16.

The remote station 18 is also called an out station or a local station, and monitors and controls the

sensors

22A and 22B and the electrical devices 20A and 20B to be connected. Since it functionally overlaps with the digital controller 16, one of the digital controller 16 and the remote station 18 is appropriately selected according to the connected electric devices 20A to 20D and the sensors 22A to 22E.

The power consumption prediction device 100, the power management device 10, the sub controller 14, the digital controller 16, and the remote station 18 are composed of computers. For example, as typically shown in the power management device 10, each of them is provided with a CPU 26, a memory 28, a hard disk drive (HDD) 30, an input unit 32, a display unit 34, and an input / output interface 36.

A functional block diagram of the power consumption prediction device 100 is illustrated in FIG. The power consumption prediction device 100 is communicated with the power management device 10, and utilizes various data stored in the power management device 10 to obtain a power consumption prediction for the entire facility.

For example, by executing a program stored in a non-transient storage medium such as a DVD or a hard disk of a computer, the computer functions as a power consumption prediction device 100 including various functional blocks illustrated in FIG. To do.

Specifically, the power consumption prediction device 100 includes an input unit 32, a display unit 34, a typical pattern extraction unit 102, a co-occurrence group extraction unit 104, a co-occurrence group generation prediction unit 106, a power consumption prediction unit 108, and an extraction result correction. A unit 112 and a prediction result correction unit 114 are provided. Further, the power management device 10 from which various data are extracted from the power consumption prediction device 100 includes a device control plan storage unit 12A, a device operation history storage unit 12B, and a power consumption history storage unit 12C as its functional blocks. ..

FIG. 3 illustrates a flowchart of the power consumption prediction process by the power consumption prediction device 100 according to the present embodiment. In this process, the time change of the power consumption of the entire facility such as a building is predicted.

With reference to FIG. 2, the typical pattern extraction unit 102 extracts an on-period typical pattern compiled based on the past on-time and off-time of each electric device for each electric device.
For example, the typical pattern extraction unit 102 acquires the operation history of the electric equipment under the control of the power management device 10 (under monitoring and control control) from the device operation history storage unit 12B of the power management device 10, and also records the operation history in the operation history. Based on this, a typical on-period pattern for each electrical device is extracted.

The device operation history storage unit 12B stores on / off information of each electric device under the power management device 10. FIG. 4 illustrates on / off information of a predetermined electric device (for example, an indoor unit of an air conditioner) under the power management device 10.

With reference to FIG. 4, the device operation history storage unit 12B has an on time of the electric device, that is, a time when the off state is switched to the on state, and an off time, that is, a time when the on state is switched to the off state. It is memorized for multiple days.

The typical pattern extraction unit 102 extracts a typical (or typical) on / off pattern for each electric device based on the device operation history, and sets this as a typical pattern during the on period (FIG. 3S10). ..

For example, the on time and the off time illustrated in FIG. 4 are plotted on a Cartesian coordinate plane as illustrated in FIG. This coordinate plane may be called a typical pattern extraction plane, and the horizontal axis indicates the on time and the vertical axis indicates the off time.

For example, referring to FIG. 4 and referring to the on / off status of 12/1, the on time is 3:00 and the off time is 18:00, and the typical pattern extraction unit 102 is a typical pattern extraction plane (FIG. 5). (See) Plot the coordinate points at the coordinates corresponding to the above on time and off time. In this way, the coordinate points are sequentially plotted on the plane of FIG. 5 along the coordinates represented by the set of the on time and the off time from 0:00 to 24:00 on each date.

The broken line drawn diagonally upward from the origin indicates the boundary line. That is, the lower right region from the boundary line (broken line) is an region in which the off time appears earlier than the on time (for example, the off time is 10:00 with respect to the on time of 12:00), and cannot occur realistically. Therefore, the area lower right than the boundary line is excluded from the extraction area of the typical pattern during the on-period.

When the operation history (on / off history) of a predetermined electric device as shown in FIG. 4 is plotted on the typical pattern extraction plane of FIG. 5 over the entire schedule, the typical pattern extraction unit 102 Extract (group) these point clouds.

In extracting the typical pattern of the on-period of the electric device, for example, when supervised learning is used, the true pattern of the on-period is the teacher signal. However, the typical on-period pattern changes according to the building, season, region, etc., and it becomes difficult to appropriately select the true on-period typical pattern. Therefore, the extraction of the typical pattern during the on-period of the electric device may be performed by, for example, clustering based on unsupervised learning or frequent pattern mining. For example, in the case of clustering, clustering using a mixed Gaussian distribution may be performed, and a single Gaussian distribution corresponds to one cluster.

When the point cloud on the typical pattern extraction plane is grouped into any of the on-period typical patterns TP1 to TP5, the typical pattern extraction unit 102 represents the on-time and off-time representative values of the on-period typical patterns TP1 to TP5, respectively. Ask for. For example, the average value and the mode of each of the plurality of on-time and off-time included in one on-period typical pattern are defined as the on-time and off-time of the respective on-period typical patterns TP1 to TP5.

Such an on-period typical pattern required for one electric device is extracted for a plurality of electric devices under the power management device 10.

When the on-period typical pattern is extracted for each electric device, the co-occurrence group extraction unit 104 (see FIG. 2) extracts the co-occurrence group from the on-period typical pattern for each electric device (FIG. 3S12).

Co-occurrence means that a certain on-period typical pattern and another on-period typical pattern occur together. This co-occurrence does not have to have strict simultaneity, for example, it is sufficient that one on-period typical pattern and another on-period typical pattern occur frequently in one day.

Further, one co-occurrence group may include a predetermined on-period typical pattern of a predetermined electric device and another on-period typical pattern of the same electric device. Similarly, a predetermined on-period typical pattern of a predetermined electric device and an on-period typical pattern of another electric device may be included in one co-occurrence group.

Typically, the indoor unit and outdoor unit of the air conditioner are included in one co-occurrence group because when one is turned on, the other is also turned on. In addition, in a predetermined tenant in a building, lighting equipment and air conditioning equipment (indoor unit, outdoor unit) installed in the tenant are also included in one co-occurrence group in summer and winter, for example.

By grouping multiple electrical devices into a co-occurrence group based on a typical pattern during the on-period, multiple electrical devices can be regarded as a single electrical facility that co-occurs with each other, and calculations related to power consumption prediction. The load can be reduced.

In addition, when extracting co-occurrence groups, typical patterns of on-periods are summarized based on the on-time and off-time of electrical equipment. In other words, each electric device is grouped only by the on / off information of the electric device. Therefore, as compared with the case where the power consumption change is added to the on / off information, the filtering based on the power consumption is eliminated, so that the number of electric devices that can be grouped together can be increased.

The co-occurrence group extraction unit 104 extracts an on-period typical pattern (having co-occurrence) that occurs together with a predetermined on-period typical pattern of a predetermined electric device, and performs one co-occurrence group together with the predetermined on-period typical pattern. Summarize in the Ki group.

FIG. 7 exemplifies the power consumption forecast on a predetermined date. In the upper time chart, typical patterns of the on-periods of the electric devices A to Z are shown, and each co-occurrence group is shown by hatching.

For example, as the on-period typical pattern included in the co-occurrence group CO-Gr2, the on-period typical patterns TP3 and TP4 of the electric device A and the on-period typical pattern TP1 of the electric device C are included.

When supervised learning is used to extract such co-occurrence groups, the co-occurrence pattern changes according to the building, season, region, etc., in the same manner as the above-mentioned on-period typical pattern, and is true. Proper selection of co-occurrence groups becomes difficult. Therefore, in extracting the co-occurrence group, operations such as matrix factorization and frequent pattern mining may be performed by unsupervised learning.

The co-occurrence group extraction unit 104 (see FIG. 2) executes matrix factorization as shown in FIG. 6, for example. This matrix factorization may be, for example, a non-negative matrix factorization.

First, as shown in the table on the upper left of FIG. 6, a matrix in which dates are arranged and typical patterns (TP1, TP2, ...) Of the on-period of each electric device are arranged as columns is generated by the co-occurrence group extraction unit 104. Will be created. This matrix is also called the observation data matrix Y, the number of rows is the number of days, and the number of columns is the product of the number of typical patterns during the on-period of each electrical device and the number of electrical devices.

For example, in the non-negative matrix factorization, the observation data matrix Y is approximately decomposed into the product (Y≈AB) of the basis matrix A and the coefficient matrix B. Here, the observation data matrix Y, the components _y ij basis matrix A, and the coefficient matrix _B, a _{ij, b ij} is expressed as follows in Equation (11) below (1).

Regarding the mathematical formula (1), in light of the present embodiment, the basis matrix A is a matrix representing which co-occurrence group occurred every day, and the coefficient matrix B is which on-period typical pattern is grouped as a co-occurrence group. It becomes a matrix representing.

In this way, the co-occurrence group extraction unit 104 (see FIG. 2) summarizes a plurality of on-period typical data co-occurring with each other as a co-occurrence group, and the collected data of each co-occurrence group is a co-occurrence group. It is transmitted to the occurrence prediction unit 106.

The co-occurrence group occurrence prediction unit 106 predicts the occurrence of each co-occurrence group.
For example, the co-occurrence group occurrence prediction unit 106 obtains a co-occurrence group predicted to occur by supervised learning from information such as the day of the week, weather, season, month, and scheduler of the prediction target day.
Here, regarding the occurrence of the co-occurrence group, it may be possible to obtain a teacher signal from past performance data. By performing supervised learning, prediction accuracy can be improved as compared with unsupervised learning.

In supervised learning, for example, one of a generalized linear model, a neural network model, and an SVM (Support vector machine) is used. For example, in supervised learning using a neural network, for example, in the learning phase, information such as the day of the week, weather, season, month, and scheduler of any date in the past is given to the input layer as training data. Further, the output layer is given a co-occurrence group generated on the date as correct answer data. By inputting such training data and correct answer data from actual past data, appropriate weighting and hidden layers are generated. Based on such a learning phase, the co-occurrence group occurrence prediction unit 106 learns a model for predicting the presence or absence of occurrence of each co-occurrence group from information such as the day of the week and the weather (FIG. 3S14).

When the learning phase is completed, for example, all the past data of the facility is input, the prediction phase is started. In the prediction phase, the co-occurrence group occurrence prediction unit 106 predicts the occurrence of each future co-occurrence group using the model after learning (FIG. 3S16).

Specifically, information such as the day of the week, weather, season, month, scheduler, etc. of the power consumption prediction target day is input to the above model. A co-occurrence group that is predicted to occur on the prediction target date is output to the output layer.

The upper part of FIG. 7 shows an example of a co-occurrence group predicted to occur on a predetermined prediction target date. In this example, typical patterns (TP1, TP2, ...) During the on-period are displayed for each of the device A, the device B, the device C, and the device Z. Each on-period typical pattern has different hatching for each co-occurrence group (CO-Gr1, CO-Gr2, ...).

When the occurrence prediction of the co-occurrence group is completed, the prediction data is sent to the power consumption prediction unit 108 (see FIG. 2). The power consumption prediction unit 108 obtains the power consumption of each co-occurrence group and predicts the power consumption of the facility based on the power consumption of the co-occurrence group predicted to occur (FIG. 3S18).

In this step, for example, power consumption prediction using a simulator is performed. For example, when extracting a typical pattern during the on-period, past actual data, that is, on-time and off-time, as illustrated in FIG. 5, are grouped. By tracing the past actual data that is the basis of this grouping, it is possible to obtain the actual value of the past power consumption corresponding to the grouped on time and off time.

For example, with reference to FIG. 5, the actual power consumption corresponding to the four coordinate points grouped in the typical pattern TP2 during the on-period can be obtained from the power consumption history storage unit 12C (see FIG. 2). For example, the average value or the median value of the actual power consumption can be used as the power consumption of the typical pattern TP2 during the on-period.

When the power sensor is not provided in each electric device, that is, the actual power consumption of each electric device is unknown, for example, the power consumption detected by the high-voltage power meter 22F (see FIG. 1) is used. By apportioning according to the rated power of each electric device that is on at the time, the actual power consumption of each electric device can be obtained.

As illustrated in the lower part of FIG. 7, the power consumption prediction unit 108 adds (integrates) the power consumption corresponding to each on-period typical pattern predicted to occur, thereby adding (accumulating) the power consumption of the entire facility in which the electrical equipment is installed. Predict power consumption (facility power consumption).

The prediction result as described above is displayed on the display unit 34, for example, as illustrated in the upper and lower rows of FIG. 7 (FIG. 3S20). The display unit 34 is composed of, for example, a display.

The upper part of FIG. 7 is a time chart showing the on period from the on time to the off time of the co-occurrence group predicted to occur. As described above, the names of the electric devices grouped as the co-occurrence group are displayed for each co-occurrence group. By displaying such a time chart on the display unit 34, the breakdown of the co-occurrence group can be confirmed. Further, in the lower part of FIG. 7, the facility power consumption, which is the sum of the power consumption of each co-occurrence group during the on period, is displayed.

The display content of the display unit 34 can be changed by an operation from the input unit 32. The input unit 32 is, for example, a mouse or a keyboard, and the pointer 120 can be moved on the display screen by operating the input unit 32, for example, as illustrated in FIG. When the on-period typical pattern TP4 is selected by the pointer 120, the distribution of past actual data grouped (grouped) as the on-period typical pattern TP4 is shown at the bottom. For example, when the electric device A is an indoor unit of an air conditioner, its past actual data includes on-time, off-time, set temperature distribution, and air volume distribution.

Further, in FIGS. 7 and 8, the forecast data on a daily basis was displayed, but the display format can be changed. For example, the power consumption prediction device 100 can output a switching command to the display unit 34 in response to the operation of the input unit 32, for example, to switch from the daily unit to the display of the forecast data for several weeks as illustrated in FIG. It has become.

With reference to FIG. 9, the vertical axis shows the co-occurrence group number and the horizontal axis shows the date. In the lower row, the vertical axis shows the power consumption, and in the horizontal axis, the date synchronized with the upper row is shown.

As described above, in the power prediction device according to the present embodiment, the co-occurrence group and the on-period typical pattern that are the basis of the power prediction are shown on the display unit. By showing the prediction basis in this way, the administrator or the like who browses the display unit can examine the validity of the prediction. Further, as will be described later, it is possible to revise the power prediction for each co-occurrence group and for each on-period typical pattern.

<First example of power consumption prediction process>
FIG. 10 illustrates a flowchart of the first alternative example of the power consumption prediction process according to the present embodiment. In this figure, the same steps as in FIG. 3 are designated by the same reference numerals, and the description thereof will be omitted as appropriate.

In the example of FIG. 10, when the occurrence prediction and the power consumption prediction of the co-occurrence group as illustrated in FIG. 8 are displayed in step S20, it is possible to correct these prediction results.

For example, as illustrated in the upper part of FIG. 8, the pointer 120 is superimposed on the on-period typical pattern TP4 for the time chart of the co-occurrence group displayed by the display unit 34. Correspondingly, the distribution of past actual data grouped as the on-period typical pattern TP4 is shown.

Furthermore, the pointer 120 is aligned with this past actual data, and corrections are made. For example, as illustrated in FIG. 11, the pointer 120 is aligned with the distribution graph of the off time, the off time of the on-period typical pattern TP4 is set as a specified value, and the mode is the mode from 17:00 to 18 of the graph. It is corrected at: 00.

In this way, when a correction instruction for the on period of the typical pattern of the on period is input from the input unit 32, the extraction result correction unit 112 (see FIG. 2) selects the selected electricity as illustrated in the upper part of FIG. The off time of the typical pattern TP4 during the on period of the device A is changed from 17:00 to 18:00 (FIG. 10S22). Further, in the flowchart of FIG. 10, the process returns to step S12, and the extraction of the co-occurrence group is executed again.

By re-extracting the co-occurrence group, the power consumption forecast will also be revised. Specifically, as shown in the lower part of FIG. 11, the end point of the power consumption of the co-occurrence group CO-Gr2 is extended as shown by the broken line.

Further, the power consumption prediction device 100 can input, for example, a correction instruction of the co-occurrence group from the input unit 32. For example, as illustrated in the upper part of FIG. 12, the on-period typical pattern TP4 of the electrical equipment A grouped in the co-occurrence group CO-Gr2 is erased by the operation of the pointer 120. For example, when the correction instruction of the co-occurrence group is input from the input unit 32 (FIG. 10S24), the extraction result correction unit 112 (see FIG. 2) erases the selected on-period typical pattern. Further, the process returns to step S14, and the occurrence prediction of the co-occurrence group is executed again.

The power consumption forecast will be revised by executing the occurrence forecast again. Specifically, as shown in the lower part of FIG. 12, the power consumption corresponding to the on-period typical pattern TP4 of the electric device A is deleted as shown by the broken line.

FIG. 13 illustrates a correction operation of a co-occurrence group different from that of FIG. In FIG. 13, a plurality of on-period typical patterns once set as different co-occurrence groups are integrated into a single co-occurrence group. For example, the co-occurrence group CO-Gr3 selected by the pointer 120 is integrated into another co-occurrence group CO-Gr2.

When such an integration instruction is input from the input unit 32 of the power consumption prediction device 100, the extraction result correction unit 112 (see FIG. 2) executes the integration process as described above. Further, the flow illustrated in FIG. 13 returns to step S14, and the occurrence prediction of the co-occurrence group is executed again. When the occurrence prediction is executed again, the power consumption prediction is also corrected. Specifically, as shown in the lower part of FIG. 13, the power consumption grouped in the co-occurrence group CO-Gr3 is changed to the co-occurrence group CO-Gr2.

<Second example of power consumption prediction process>
FIG. 14 illustrates a flowchart of a second alternative example of the power consumption prediction process according to the present embodiment. In this figure, the same steps as in FIG. 3 are designated by the same reference numerals, and the description thereof will be omitted as appropriate.

In this example, after the occurrence prediction of the co-occurrence group is performed (S16), the occurrence prediction is corrected by using the control plan of the electric device (S26). With reference to FIG. 2, the power management device 10 is provided with a device control plan storage unit 12A. The device control plan storage unit 12A stores the control plans of the electric devices installed in the facility and under the control of the power management device 10. For example, the date and time zone in which the on operation is prohibited are stored.

With reference to FIG. 2, the prediction result correction unit 114 of the power consumption prediction device 100 predicts by the co-occurrence group generation prediction unit 106 based on the control plan of each electric device stored in the device control plan storage unit 12A. Correct the occurrence predictions made. Based on the revised occurrence prediction, the power consumption prediction (S18) and the display of the prediction result (S20) are performed.

<Third example of power consumption prediction process>
FIG. 15 illustrates a flowchart of another third example of the power consumption prediction process according to the present embodiment. In this figure, the same steps as in FIG. 3 are designated by the same reference numerals, and the description thereof will be omitted as appropriate.

In this example, after the occurrence prediction (S16) of the co-occurrence group, the power consumption prediction (S18) is executed through the step (S28) of learning the model for predicting the power consumption. In learning a model for predicting power consumption, for example, supervised learning is performed using the past power consumption history (actual results) as a teacher.

With reference to FIG. 2, the power consumption history storage unit 12C of the power management device 10 stores the power consumption history of each of the electric devices installed in the facility and under the control of the power management device 10. Alternatively, if each electrical device is not provided with a power sensor, the power consumption detected by the high-voltage power meter 22F (see FIG. 1) is apportioned according to the rated power of each electrical device that is on at that time. By doing so, the power consumption history of each electric device can be obtained. The power consumption prediction unit 108 acquires the power consumption history of each electric device from the power consumption history storage unit 12C, and executes supervised learning using this as a teacher.

For example, in supervised learning using a neural network, for example, in the learning phase, the period from the on time to the off time of each electric device in the past is input to the input layer as training data. The output layer is given power consumption corresponding to the period as correct answer data. By inputting such training data and correct answer data from actual past data, appropriate weighting and hidden layers are generated.

For the model obtained through such a learning phase, the power consumption prediction unit 108 inputs the co-occurrence group predicted to occur in step S16. As a result, the output layer is required to predict the power consumption corresponding to the co-occurrence group.

The power consumption prediction device according to the present invention includes a typical pattern extraction unit, a co-occurrence group extraction unit, a co-occurrence group generation prediction unit, and a power consumption prediction unit, and power consumption prediction while suppressing a decrease in accuracy of power consumption prediction. It is possible to reduce the calculation load at that time, which is suitable for predicting power consumption.

10 Power management device, 12A device control plan storage unit, 12B device operation history storage unit, 12C power consumption history storage unit, 20A to 20D electrical equipment, 22A to 22F sensor, 100 power consumption prediction device, 102 typical pattern extraction unit, 104 Co-occurrence group extraction unit, 106 co-occurrence group generation prediction unit, 108 power consumption prediction unit, 112 extraction result correction unit, 114 prediction result correction unit, 120 pointer.

Claims

It is a power consumption prediction device that predicts the power consumption of a facility in which a plurality of electric devices are installed, and the electricity is a typical pattern of an on-period that is summarized based on the past on-time and off-time of each of the electric devices. A typical pattern extraction unit that extracts each device,
The co-occurrence pattern having the co-occurrence that occurs together with the predetermined on-period typical pattern of the predetermined electric device is extracted and combined with the predetermined on-period typical pattern into one co-occurrence group. Ki group extraction department and
A co-occurrence group occurrence prediction unit that predicts the occurrence of each of the co-occurrence groups,
A power consumption prediction unit that obtains the power consumption of each co-occurrence group and predicts the power consumption of the facility based on the power consumption of the co-occurrence group predicted to occur.
A power consumption forecasting device.
The power consumption prediction device according to claim 1.
The on-period from the on-time to the off-time of each of the co-occurrence groups predicted to occur and the facility power consumption of the sum of the power consumption of each of the co-occurrence groups in the on-period are displayed. Equipped with a display unit
The display unit can display the names of the electric devices grouped as the co-occurrence group for each co-occurrence group.
Power consumption forecaster.
The power consumption prediction device according to claim 2.
The display unit can display the distribution of the on-time and the off-time summarized as a single typical pattern of the on-period.
Power consumption forecaster.
The power consumption prediction device according to claim 3.
The typical pattern extraction unit extracts the on-period typical pattern of each of the electric devices from the past on-time and off-time by unsupervised learning.
Power consumption forecaster.
The power consumption prediction device according to claim 4.
The co-occurrence group extraction unit extracts a plurality of the co-occurrence typical patterns of the on-period by unsupervised learning and puts them together in a single co-occurrence group.
Power consumption forecaster.
The power consumption prediction device according to claim 5.
The power consumption prediction unit is a power consumption prediction device that predicts the co-occurrence group that occurs on a predetermined day from the past electrical equipment operation history by supervised learning.
The power consumption prediction device according to any one of claims 2 to 6.
The display unit includes an input unit capable of inputting a correction instruction for the on-period of each of the on-period typical patterns displayed on the display unit.
Power consumption forecaster.
The power consumption prediction device according to claim 7.
The input unit can input integration instructions for a plurality of the co-occurrence groups displayed on the display unit.
Power consumption forecaster.
The power consumption prediction device according to claim 1.
A power consumption prediction device including a prediction result correction unit that corrects the occurrence prediction for each co-occurrence group based on a control plan set for each of the electric devices.