JP7375355B2 - Estimation method, estimation device, computer program, and method for generating a trained model - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act Published at the 41st Information, Systems, Utilization, and Technology Symposium on December 6, 2018

The present invention relates to an estimation method, an estimation device, a computer program, and a method for generating a trained model.

A configuration has been proposed that performs various estimations based on the amount of power consumed in a building. For example, Patent Document 1 discloses a configuration for estimating living conditions based on the amount of power consumed in a building. Living conditions include the number of people in the building, whether people are cooking or not, or whether they are taking a bath. Based on the estimation results, support or advice regarding daily life can be provided.

Japanese Patent Application Publication No. 2015-153156

In buildings, electricity is consumed for multiple purposes, such as powering air conditioners, water heaters, and lights. In order to save electricity in a building, it is necessary to know which uses consume a lot of electricity. For this reason, for example, it is necessary to know a plurality of individual consumption amounts that indicate the breakdown of the total amount of power consumed in a building for each use in one day.

As a configuration for knowing a plurality of individual consumption amounts, a configuration can be considered in which a wattmeter for measuring power consumption is installed for each purpose, and the consumption amounts indicated by the plurality of wattmeters are obtained. However, when actually measuring power consumption for each application, it is necessary to install a plurality of power meters, so the cost of realizing this configuration is high.

The present invention has been made in view of the above circumstances, and its purpose is to provide an estimation method, an estimation device, a computer program, and The goal is to provide trained models.

In the estimation method according to one aspect of the present invention, the total amount of power consumed within a predetermined period in a target building that is an estimation target is obtained, and the obtained total amount of consumption is calculated as the amount of power consumed in the building. The relationship between the total consumption amount and a plurality of individual consumption amounts showing the breakdown of the total consumption amount for each use of the building is input to the trained model, and based on the output of the trained model, the relationship is applied to the target building. Determine whether the estimated individual consumption amount is excessive by estimating a plurality of corresponding individual consumption amounts and determining whether the occupancy rate of the estimated individual consumption amount is greater than or equal to a second threshold value. The computer executes the process.

In one aspect of the above, the total amount of power consumed in a target building within a predetermined period of time, such as one day or one week, is input into the trained model. Based on the output of the trained model, estimate multiple individual consumption amounts broken down by usage. Therefore, there is no need to measure individual consumption amounts for each use, and a configuration for estimating a plurality of individual consumption amounts can be realized at low cost.

According to the above aspect, it is possible to estimate a plurality of individual consumption amounts that show the breakdown for each use.

1 is a schematic diagram of an estimation system in Embodiment 1. FIG. FIG. 2 is a block diagram showing the configuration of main parts of a server. FIG. 2 is an explanatory diagram of a consumption estimation model. FIG. 3 is an explanatory diagram of teacher data. 3 is a flowchart showing the procedure of model generation processing. It is a flowchart which shows the procedure of an estimation process. FIG. 7 is an explanatory diagram of a consumption amount estimation model in Embodiment 2. FIG. FIG. 3 is an explanatory diagram of teacher data. FIG. 3 is a block diagram showing the configuration of main parts of a server in Embodiment 3. FIG. 3 is a flowchart showing the procedure of model generation processing. It is an explanatory diagram of a selection table. It is a flowchart which shows the procedure of an estimation process. FIG. 7 is an explanatory diagram of a selection table in Embodiment 4; FIG. 7 is an explanatory diagram of a consumption amount estimation model in Embodiment 5; 3 is a flowchart showing the procedure of model generation processing. It is a flowchart which shows the procedure of an estimation process.

A specific example of an estimation system according to an embodiment of the present invention will be described below with reference to the drawings.
(Embodiment 1)
FIG. 1 is a schematic diagram of an estimation system 1 in the first embodiment. The estimation system 1 includes a server 2 and a terminal 3. Server 2 and terminal 3 are connected to network N. The server 2 is an information processing device that performs various information processing, data transmission and reception, and the like. Based on the total power consumption, the server 2 estimates a plurality of individual consumption amounts showing a breakdown for each use. The target building 4 that is the estimation target is an office, a factory, a commercial facility, a residence, or the like. Applications include power supply to outdoor units of air conditioners, indoor units of air conditioners, water heaters, electric lights, office equipment, etc. The office equipment is, for example, a personal computer.

In the target building 4, a power meter (not shown) is installed that measures the total amount of power consumed per unit time such as one hour or one day within a predetermined period such as one day or one week. The terminal 3 acquires target data indicating the total amount of power consumed per unit time in the target building 4 within a predetermined period in time series. The target data is data showing, in time series, a plurality of total consumption amounts measured by a power meter for each unit time within a predetermined period. The user causes the terminal 3 to acquire the target data, for example, via communication or using a portable memory. The user causes the target data of the total consumption amount to be transmitted to the server 2 via the network N. The terminal 3 is, for example, a personal computer. In addition, 2 hours or 2 days etc. are also mentioned as a unit time.

Based on the target data received from the terminal 3, the server 2 estimates a plurality of individual consumption amounts indicating a breakdown of each usage with respect to the average value or the total of the plurality of total consumption amounts indicated by the target data. As an example, if the predetermined period is one week and the unit time is one day, the server 2 calculates the breakdown of each usage for the average value of multiple total consumption amounts (power consumption per day). Estimate multiple individual consumption amounts that indicate. As another example, if the predetermined period is one day and the unit time is one hour, the server 2 calculates the amount of electricity for each use for the sum of the 24 total consumption amounts (one day's power consumption). Estimate multiple individual consumption amounts with a breakdown. Server 2 functions as an estimation device. The server 2 transmits estimation data indicating a plurality of estimated individual consumption amounts to the terminal 3 via the network N. When receiving the estimated data, the terminal 3 displays a plurality of individual consumption amounts indicated by the received estimated data.

The server 2 stores time-series data showing the total amount of power consumed per unit time within a predetermined period in a building 5 such as an office, factory, commercial facility, or residence, and a plurality of data indicated by the time-series data. Machine learning is performed on the relationship between the average value of the total consumption amount or the relationship between the multiple individual consumption amounts that show the breakdown by usage against the total amount. As a result, the server 2 inputs the target data and outputs a plurality of individual consumption amounts indicating the average value or the breakdown of the total consumption amount for each purpose of the plurality of total consumption amounts indicated by the target data (Fig. 2 reference). The server 2 performs machine learning for each of the plurality of buildings 5.

Therefore, the consumption estimation model 60 is a trained model that has learned the relationship between time-series data and a plurality of individual consumption amounts indicating a breakdown of each use for the average value or total of a plurality of total consumption amounts indicated by the time-series data. It is. The server 2 estimates a plurality of individual consumption amounts regarding the target building 4 using the consumption amount estimation model 60.
Note that it is preferable that the target building 4 and the building 5 have the same type. For example, the types of target buildings 4 and 5 may be offices.

FIG. 2 is a block diagram showing the main configuration of the server 2. As shown in FIG. The server 2 includes a communication section 20, a storage section 21, and a control section 22. These are connected to an internal bus 23. The communication unit 20 is further connected to a network N.

The communication unit 20 receives data via the network N. One of the data that the communication unit 20 receives from the terminal 3 is target data. The communication unit 20 transmits data to the terminal 3 via the network N according to instructions from the control unit 22. One of the data that the communication unit 20 transmits to the terminal 3 is estimated data.
The storage unit 21 is a nonvolatile memory. A consumption estimation model 60 is stored in the storage unit 21 .

FIG. 3 is an explanatory diagram of the consumption estimation model 60. As described above, the server 2 generates the consumption estimation model 60 by performing machine learning. The consumption estimation model 60 is a neural network related to LSTM (Long-Short Term Memory). A neutral network related to LSTM is a type of RNN (Recurrent Neural Network). The consumption estimation model 60 inputs data indicating the total amount of power consumed per unit time in a time series, that is, target data or time series data, and calculates the amount of power that the target data or time series data indicates. This is a neutral network that outputs a plurality of individual consumption amounts showing a breakdown for each use with respect to the average value or total of a plurality of total consumption amounts. FIG. 3 shows an example in which the predetermined period is K days and the unit time is one day. K is an integer of 2 or more. The unit time may be, for example, one hour. In this case, the predetermined period is, for example, one day.

As a first example, when the predetermined period is K days and the unit time is one day, the consumption estimation model 60 calculates the average value of multiple total consumption amounts (per day). Outputs multiple individual consumption amounts that show a breakdown of each usage (power consumption amount). As a second example, when the predetermined period is one day and the unit time is one hour, the consumption estimation model 60 calculates the sum of 24 total consumption amounts (per day). Outputs multiple individual consumption amounts that show a breakdown of each usage (power consumption amount). As a third example, when the predetermined period is 30 days and the unit time is 1 day, the consumption estimation model 60 calculates the total consumption of 30 items (per month). Outputs multiple individual consumption amounts showing the breakdown for each use (power consumption amount). As a fourth example, when the predetermined period and unit time are one hour, the consumption estimation model 60 calculates the average value of multiple total consumption amounts (power consumption per hour). Outputs multiple individual consumption amounts showing a breakdown by usage.

The consumption estimation model 60 has an input layer, a middle layer, and an output layer. The input layer has a plurality of neurons that accept input values at each time point, ie, total consumption amount, in time series. The output layer has neurons that output a plurality of individual consumption amounts that are broken down by usage. The intermediate layer has neurons for calculating a plurality of individual consumption amounts based on input values to a plurality of neurons included in the input layer. Neurons in the intermediate layer are called LSTM Blocks, and use calculation results of the intermediate layer regarding input values at past points in time to perform calculations regarding input values at the next point in time. Thereby, the neurons in the middle layer calculate the value at the next point in time from the time series data up to the most recent point in time.

Note that the configuration of the neutral network related to LSTM shown in FIG. 3 is an example. The configuration of the neutral network related to LSTM is not limited to the configuration shown in FIG. 3. The number of intermediate layers is not limited to one, and may be two or more. The number of neurons in the output layer is also not limited to one, and may be two or more.

The control unit 22 shown in FIG. 2 includes a processing element such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), or a GPU (Graphics Processing Unit). Calculations in the consumption estimation model 60 are executed by a processing element (computer) of the control unit 22. The control unit 22 inputs target data or time series data to the input layer of the consumption estimation model 60.

Specifically, the control unit 22 inputs a plurality of total consumption amounts indicated by the target data or time series data to each of the plurality of neurons included in the input layer. As described above, in the target data or time series data, the total amount of power consumed per unit time within a predetermined period is shown in time series. In the data shown in FIG. 3, the total amount of power consumed per day within K days is shown in time series. Note that the target data or time-series data may be data in which the total amount of power consumed per hour within K hours is shown in a time-series manner.

The control unit 22 calculates a plurality of individual consumption amounts as outputs of the consumption amount estimation model 60 by inputting target data or time series data to the input layer of the consumption amount estimation model 60. The control unit 22 performs machine learning using the teacher data and generates the consumption estimation model 60. In machine learning, the control unit 22 optimizes parameters used for calculation in the intermediate layer. Here, the parameter is a weight (coupling coefficient) between neurons, a coefficient of an activation function, or the like.

FIG. 4 is an explanatory diagram of the teacher data. In the training data, correct answer data indicating the correct output of the consumption estimation model 60 corresponds to time series data indicating the total amount of power consumed per unit time in the common building 5 within a predetermined period. It is attached. The correct data indicates, for the common building 5, a plurality of individual consumption amounts indicating a breakdown for each use with respect to the average value or the total of the plurality of total consumption amounts indicated by the time series data. The correct data indicates, for example, a plurality of individual consumption amounts showing the actual breakdown.

In the example of FIG. 4, the predetermined period is K days, and the unit time is one day. The time-series data shows the total amount consumed per day within K days in a time-series manner. When the predetermined period is one day and the unit time is one hour, the time-series data shows the total amount consumed per hour within one day in a time-series manner. The time series data and the correct data correspond to the data input to the consumption estimation model 60 and the output of the consumption estimation model 60, respectively. The predetermined period and unit time related to the data input to the consumption estimation model 60 respectively match the predetermined period and unit time related to time series data. When the consumption estimation model 60 outputs a plurality of individual consumption amounts that indicate a breakdown of each use with respect to the average value of a plurality of total consumption amounts, the correct answer data indicates a breakdown of each use with respect to the average value of a plurality of total consumption amounts. This is data showing multiple individual consumption amounts. When the consumption estimation model 60 outputs a plurality of individual consumption amounts indicating a breakdown of each use for the total of multiple total consumption amounts, the correct data is a plurality of individual consumption amounts indicating a breakdown of each use for the total of multiple total consumption amounts This data shows individual consumption.

The control unit 22 inputs the time series data of the teacher data to the input layer of the consumption estimation model 60, and calculates a plurality of individual consumption amounts as outputs of the output layer. The control unit 22 compares the plurality of calculated individual consumption amounts with the plurality of individual consumption amounts indicated by the correct answer data, and optimizes the parameters used for calculation in the intermediate layer based on the comparison result.

As shown in FIG. 2, the storage unit 21 stores a computer program P1. The processing element (computer) of the control unit 22 executes the teacher data storage process, model generation process, and estimation process by executing the computer program P1. The teacher data storage process is a process of storing teacher data in the storage unit 21. The model generation process is a process of generating the consumption estimation model 60 by performing machine learning. The estimation process is a process of estimating a plurality of individual consumption amounts showing a breakdown for each use based on the total power consumption amount.

Note that the number of processing elements included in the control unit 22 may be two or more. In this case, the plurality of processing elements may cooperate to execute various processes including teacher data storage processing, model generation processing, and estimation processing according to the computer program P1. Further, calculations performed by the consumption amount estimation model 60 may be executed cooperatively by a plurality of processing elements.

Furthermore, the computer program P1 may be provided to the server 2 using a non-transitory storage medium A1 that readably records the computer program P1. The storage medium A1 is, for example, a portable memory. Examples of the portable memory include a CD-ROM, a USB (Universal Serial Bus) memory, an SD card, a micro SD card, or a Compact Flash (registered trademark). When the storage medium A1 is a portable memory, the processing element of the control unit 22 reads the computer program P1 from the storage medium A1 using a reading device (not shown), and stores the read computer program P1 in the storage unit 21. good. Furthermore, the computer program P1 may be provided to the server 2 through communication via the communication unit 20 of the server 2. In this case, the processing element of the control unit 22 may acquire the computer program P1 through the communication unit 20 and store the acquired computer program P1 in the storage unit 21.

The control unit 22 executes the teacher data storage process when the communication unit 20 receives the teacher data or when the teacher data is input to the input unit of the server 2 (not shown). The control unit 22 acquires the teacher data received by the communication unit 20 or input to the input unit, and stores it in the acquired storage unit 21 . Thereafter, the control unit 22 ends the teacher data storage process.

FIG. 5 is a flowchart showing the procedure of model generation processing. For example, the control unit 22 executes the model generation process when instructed to execute the model generation process while the teacher data is stored in the storage unit 21 . The instruction to execute the model generation process is realized by transmitting instruction data instructing the execution to the communication unit 20, or by operating an operation unit (not shown) included in the server 2.

In the model generation process, the control unit 22 selects one piece of teacher data from a plurality of pieces of teacher data stored in the storage unit 21 (step S1). If the number of teacher data stored in the storage unit 21 is one, the control unit 22 selects one piece of teacher data stored in the storage unit 21 in step S1. Next, the control unit 22 inputs the time series data of the total consumption included in the teacher data selected in step S1 to the consumption estimation model 60 (step S2), and outputs the time-series data for each application as the output of the consumption estimation model 60. A plurality of individual consumption amounts indicating the breakdown of the amounts are calculated (step S3).

Next, the control unit 22 compares the plurality of individual consumption amounts calculated in step S2 with the plurality of individual consumption amounts indicated by the correct answer data included in the teacher data selected in step S1 (step S4). Next, the control unit 22 updates the parameters used for calculation in the intermediate layer so that the output of the consumption amount estimation model 60 approximates the plurality of individual consumption amounts indicated by the correct data (step S5). As a result, a new consumption estimation model 60 is generated.

After executing step S5, the control unit 22 determines whether all the teacher data stored in the storage unit 21 have been selected (step S6). If the control unit 22 determines that all the teacher data have not been selected (S6: NO), it executes step S1 and updates the parameters again. In step S1, the control unit 22 selects teacher data that has not been selected in the model generation process. The control unit 22 continues to update the parameters until all the teacher data are selected.

If the control unit 22 determines that all the teacher data have been selected (S6: YES), it deletes all the teacher data stored in the storage unit 21 (step S7), and ends the model generation process.

As described above, in the model generation process, the control unit 22 generates the consumption estimation model 60 based on the teacher data acquired in the teacher data storage process.
Note that in the model generation process, the number of times a parameter is updated using one teaching data is not limited to one. The control unit 22 may update the parameters two or more times using one piece of teacher data.

FIG. 6 is a flowchart showing the procedure of estimation processing. The control unit 22 periodically executes the estimation process. In the estimation process, the control unit 22 first determines whether the communication unit 20 has received target data from the terminal 3 (step S11). As described above, the target data is data indicating the total amount of power consumed within a predetermined period in time series in the target building 4. The target data received by the communication unit 20 is acquired by the control unit 22. The control unit 22 functions as an acquisition unit.

When the control unit 22 determines that the communication unit 20 has not received the target data (S11: NO), the control unit 22 ends the estimation process. When the next cycle has arrived, the control unit 22 executes step S11 again. The control unit 22 repeatedly executes the estimation process and waits until the communication unit 20 receives the target data.

When the control unit 22 determines that the communication unit 20 has received the target data (S11: YES), the control unit 22 inputs the target data received by the communication unit 20 into the consumption amount estimation model 60 (step S12). The control section 22 also functions as an input section. Next, the control unit 22 calculates, as an output of the consumption estimation model 60, a plurality of individual consumption amounts indicating the average value or the breakdown of the total consumption amount for each use indicated by the target data (step S13). . Thereby, a plurality of individual consumption amounts corresponding to the target building 4 are estimated. The control unit 22 also functions as an estimation unit.

The plurality of individual consumption amounts calculated by the control unit 22 in step S13 are the average value of the plurality of total consumption amounts measured every day (power consumption amount per day), and the 24 individual consumption amounts measured every hour. (power consumption per day), the sum of 30 total consumptions measured every day (power consumption per month), or the sum of 30 total consumptions measured every hour (power consumption per month). These are a plurality of individual consumption amounts corresponding to the average value (power consumption per hour) of a plurality of total consumption amounts.

Next, the control unit 22 determines whether each individual consumption amount calculated in step S13 is excessive (step S14). As a first example of step S14, a plurality of first threshold values corresponding to a plurality of uses are set in advance, and the control unit 22 sets each individual consumption amount calculated in step S13 as the first threshold value for a common use. compare. The control unit 22 determines that the individual consumption amount is excessive when the individual consumption amount is greater than or equal to the first threshold value. The control unit 22 determines that the individual consumption amount is not excessive when the individual consumption amount is less than the first threshold value. As an example, for a common application, each first threshold is set to the individual consumption consumed in a standard building. As another example, each first threshold value is set to the average value of a plurality of individual consumption amounts corresponding to a plurality of buildings 5 and having a common purpose.

As a second example of step S14, the control unit 22 calculates the occupancy rate occupied by each individual consumption amount with respect to the average value or total of a plurality of total consumption amounts indicated by the target data. Here, when the consumption estimation model 60 outputs a plurality of individual consumption amounts indicating a breakdown for each use with respect to the average value of a plurality of total consumption amounts, the control unit 22 outputs , calculate the share occupied by each individual consumption amount. When the consumption estimation model 60 outputs a plurality of individual consumption amounts indicating a breakdown for each use with respect to the total of a plurality of total consumption amounts, the control unit 22 outputs each individual consumption amount with respect to the total of a plurality of total consumption amounts. Calculate the occupancy rate occupied by

A plurality of second threshold values corresponding to a plurality of uses are set in advance, and the control unit 22 compares each calculated occupancy rate with the second threshold value for the common use. The control unit 22 determines that the individual consumption amount is excessive when the occupancy rate is greater than or equal to the second threshold. The control unit 22 determines that the individual consumption amount is not excessive when the occupancy rate is less than the second threshold. As an example, for a common use, each second threshold is set to the occupancy of the individual consumption consumed in a standard building. As another example, each second threshold value is set to the average value of the occupancy rates of a plurality of individual consumption amounts corresponding to a plurality of buildings 5 and having a common purpose.

As a third example of step S14, the control unit 22 compares each individual consumption amount calculated in step S13 with a plurality of individual consumption amounts consumed by a plurality of buildings 5 for a common use. Based on the comparison result, the control unit 22 determines whether each individual consumption amount calculated in step S13 is excessive.

After executing step S14, the control unit 22 instructs the communication unit 20 to transmit the determination result data and estimation data to the terminal 3 via the network N (step S15). The determination result data indicates the result of the determination made in step S14 for each individual consumption amount calculated in step S13, that is, whether or not the individual consumption amount is excessive. The estimated data indicates a plurality of individual consumption amounts calculated by the control unit 22 in step S13. When receiving the determination result data and estimation data from the communication unit 20, the terminal 3 displays the contents of the received data.

Based on the contents of the estimated data, the user can confirm the breakdown of the average value or total of the plurality of total consumption amounts indicated by the target data for each use. The user can confirm the purpose of the excessive individual consumption amount based on the content of the determination result data.

After executing step S15, the control unit 22 ends the estimation process. When the next cycle arrives, the control unit 22 executes step S11 of the estimation process again, and waits until the communication unit 20 receives the target data.

As described above, the server 2 estimates a plurality of individual consumption amounts based on the target data. For this reason, it is not necessary to measure individual consumption amounts for each purpose of the target building 4, and a configuration for estimating a plurality of individual consumption amounts can be realized at low cost. Since the target data, that is, a plurality of total consumption amounts shown in time series are input into the consumption estimation model 60 generated using time-series data, a plurality of individual consumption amounts can be accurately estimated.

(Embodiment 2)
In the first embodiment, data showing only the total amount of power consumed within a predetermined period in the target building 4 in time series is used as the target data used for estimating a plurality of individual consumption amounts. The target data may be data showing not only the total consumption amount but also other information in chronological order.
In the following, differences between the second embodiment and the first embodiment will be explained. Other configurations other than those described below are the same as in Embodiment 1, so the same reference numerals as in Embodiment 1 are given to the components common to Embodiment 1, and the explanation thereof is omitted. do.

FIG. 7 is an explanatory diagram of the consumption estimation model 60 in the second embodiment. As described in the description of the first embodiment, target data or time series data is input to the input layer of the consumption estimation model 60. In Embodiment 2, as shown in FIG. 7, in each of the target data and time series data, in addition to the total power consumption per unit time within a predetermined period, weather information and floor area are is shown. The input values accepted by each of the plurality of neurons in the input layer are total consumption, weather information, and floor area. FIG. 7 shows an example in which the predetermined period is K days and the unit time is one day. The target data and time-series data each indicate total consumption, weather information, and floor area at daily intervals. As described in the description of the first embodiment, the predetermined period may be K hours and the unit time may be one hour. In this case, the target data and time series data each indicate total consumption, weather information, and floor area at one hour intervals.

The total consumption of each of the target data and time-series data is the total consumption of power consumed in the target buildings 4 and 5 within a predetermined period per unit time, as in the first embodiment. The weather information for each of the target data and time-series data is information regarding the weather around the target building 4 and building 5. In the example of FIG. 7, the average temperature, average humidity, and weather around the target building 4 or building 5 during a predetermined period are shown as the weather information. The floor area of each of the target data and time series data is the floor area of the target building 4 and building 5.

The temperature around the target building 4 is measured, for example, by a thermometer installed around the target building 4. The humidity around the target building 4 is measured, for example, by a hygrometer installed around the target building 4. A camera, for example a camera, may be installed to take an image of the sky above the target building 4. In this case, the weather around the target building 4 may be determined based on the image taken by the camera. The temperature, humidity, and weather around the target building 4 may be information on the Internet. The terminal 3 acquires the target data in the same manner as in the first embodiment.

In the consumption estimation model 60 in the second embodiment, similarly to the first embodiment, the plurality of neurons included in the input layer receive input values at each time point in time series. Here, the input values that the neuron accepts are the total consumption at each point in time, weather information, and floor area. In the consumption estimation model 60 in the second embodiment, the intermediate layer and the output layer operate in the same manner as in the first embodiment.

FIG. 8 is an explanatory diagram of the teacher data. In the teacher data in the second embodiment, as in the first embodiment, correct answer data is associated with time series data. In the second embodiment, the control unit 22 executes teacher data storage processing and model generation processing in the same manner as in the first embodiment. As described in the description of the first embodiment, in the model generation process, the control unit 22 generates the consumption estimation model 60 based on the teacher data acquired in the teacher data storage process. Therefore, the consumption estimation model 60 is based on the total amount of power consumed per unit time in the building 5 within a predetermined period, weather information around the building 5, and a time series that shows the floor area of the building 5 in chronological order. This is a trained model that has learned the relationship between data and a plurality of individual consumption amounts that indicate a breakdown of each use for the average value or total of a plurality of total consumption amounts indicated by time-series data.

In the second embodiment, the control unit 22 executes the estimation process in the same manner as in the first embodiment. Therefore, the control unit 22 acquires the target data and inputs the acquired target data to the consumption amount estimation model 60. The control unit 22 calculates, as an output of the consumption estimation model 60, a plurality of individual consumption amounts indicating a breakdown for each use with respect to the average value or the total of the plurality of total consumption amounts indicated by the target data. Thereby, a plurality of individual consumption amounts corresponding to the target building 4 are estimated.

The server 2 in the second embodiment has the same effects as the server 2 in the first embodiment. Furthermore, the server 2 in the second embodiment further includes weather information regarding the weather around the target building 4 and the floor area of the target building 4 as elements used for estimating multiple individual consumption amounts. Quantities are estimated more accurately.

Note that the weather information is not limited to average temperature, average humidity, or weather. As the weather information, instead of the average temperature, one temperature measured every unit time, the highest temperature within a unit time, the lowest temperature within a unit time, etc. may be used. Similarly, instead of the average humidity, one humidity measured within a unit time, the highest humidity within a unit time, the lowest humidity within a predetermined period, etc. may be used. The number of weather information is not limited to 3, and may be 1, 2, or 4 or more.

In each of the target data and the time series data, the total amount of power consumed per unit time within a predetermined period and weather information are shown in time series, and the floor area may be excluded. In this case, the consumption estimation model 60 uses time-series data showing the total amount of electricity consumed in the building 5 in a predetermined period per unit time, and weather information around the building 5 in a chronological order. This is a trained model that has learned the relationship between the average value of multiple total consumption amounts indicated by the data or the relationship between the multiple individual consumption amounts indicating the breakdown for each use with respect to the total. Even with this configuration, since weather information regarding the weather around the target building 4 is further included as an element used for estimating the plurality of individual consumption amounts, the plurality of individual consumption amounts can be estimated more accurately.

Similarly, in each of the target data and time series data, the total amount of power consumed per unit time within a predetermined period and the floor area are shown in time series, and weather information may be excluded. In this case, the consumption estimation model 60 includes time-series data showing the total amount of power consumed per unit time in the building 5 within a predetermined period and the floor area of the building 5, and time-series data. This is a trained model that has learned the relationship between the average value of the plurality of total consumption amounts shown by , or the relationship between the plurality of individual consumption amounts showing the breakdown for each use with respect to the average value or total of the plurality of total consumption amounts. Even with this configuration, the floor area of the target building 4 is further included in the elements used for estimating a plurality of individual consumption amounts, so that a plurality of individual consumption amounts can be estimated more accurately.

(Embodiment 3)
In the first embodiment, the server 2 has consumption information that has learned the relationship between time-series data and a plurality of individual consumption amounts indicating the average value of a plurality of total consumption amounts indicated by the time-series data or a breakdown of each usage for the total. The number of quantity estimation models (trained models) is one. However, the number of consumption estimation models is not limited to one, and may be two or more.
In the following, differences between the third embodiment and the first embodiment will be explained. Other configurations other than those described below are the same as in Embodiment 1, so the same reference numerals as in Embodiment 1 are given to the components common to Embodiment 1, and the explanation thereof is omitted. do.

FIG. 9 is a block diagram showing the main configuration of the server 2 in the third embodiment. A plurality of consumption estimation models 60a, 60b, . . . are stored in the storage unit 21 in the third embodiment. These are configured similarly to the consumption estimation model 60 in the first embodiment. Therefore, each of the plurality of consumption estimation models 60a, 60b, . The relationship between the average value of a plurality of total consumption amounts shown by , or the relationship between a plurality of individual consumption amounts showing a breakdown for each use with respect to the total amount is learned. Target data and time series data are input to the input layer of each of the plurality of consumption estimation models 60a, 60b, . . . . As described in the description of the first embodiment, each of the target data and the time series data indicates the total amount of power consumption within a predetermined period in time series. The output layer of each of the plurality of consumption estimation models 60a, 60b, . Output.

The plurality of consumption estimation models 60a, 60b, . . . correspond to a plurality of types and periods regarding the building 5. As described in the description of the first embodiment, the types of buildings 5 include offices, factories, commercial facilities, residences, and the like. The month is mentioned as a time.

In the third embodiment, the control unit 22 executes the teacher data storage process similarly to the first embodiment. In the third embodiment, the teacher data includes model data in addition to time series data and correct answer data. The model data indicates a consumption estimation model that should be generated among the consumption estimation models 60a, 60b, . . . using time series data and correct answer data included in the own teacher data.

FIG. 10 is a flowchart showing the procedure of model generation processing. Similarly to the first embodiment, the control unit 22 executes the model generation process when instructed to execute the model generation process with the teacher data stored in the storage unit 21 .
A part of the model generation process in the third embodiment is common to the model generation process in the first embodiment. Therefore, in the model generation process in the third embodiment, a detailed explanation of the parts common to the model generation process in the first embodiment, that is, steps S1 to S7 will be omitted.

In the model generation process, after executing step S1, the control unit 22 selects a consumption estimation model indicated by the model data of the teacher data selected in step S1 from among the plurality of consumption estimation models 60a, 60b, . is selected (step S21). After executing step S21, the control unit 22 executes step S2. In step S2 of the third embodiment, the control unit 22 inputs the time series data of the total consumption included in the teacher data selected in step S1 to the consumption estimation model selected in step S21. After that, the control unit 22 sequentially executes steps S3 to S5. As a result, the parameters used for calculation in the intermediate layer of the consumption estimation model selected by the control unit 22 in step S21 are updated, and a new consumption estimation model is generated.

As described above, in the model generation process, the control unit 22 generates at least one of the plurality of consumption estimation models 60a, 60b, . . . based on the teacher data acquired in the teacher data storage process.
Note that in the model generation process in the third embodiment as well, the number of times a parameter is updated using one teaching data is not limited to one. The control unit 22 may update the parameters two or more times using one piece of teacher data.

In the third embodiment, the control unit 22 executes estimation processing similarly to the first embodiment. In the estimation process, the control unit 22 selects a consumption estimation model from among the plurality of consumption estimation models 60a, 60b, . As shown in FIG. 9, the storage unit 21 further stores a selection table T1 used for selecting a consumption estimation model.

FIG. 11 is an explanatory diagram of the selection table T1. The selection table T1 is provided with a type field, a period field, and a consumption amount estimation model field. The type field indicates the type of building 5. In the example of FIG. 11, the types of buildings 5 include offices, factories, commercial facilities, and the like. In the period field, the period is shown in association with each of the plurality of types indicated by the type field. In the example of FIG. 11, the months from January to December are shown as the period. In the consumption amount estimation model field, the consumption amount estimation model to be selected is shown in association with each combination of type and time indicated by the type field and the time field. The consumption estimation model indicated in the consumption estimation model field is one of the consumption estimation models 60a, 60b, .

In addition to the wattmeter, the target building 4 is equipped with a storage unit storing type data indicating the type of the target building 4 and a clock unit indicating the year, month, day, and time. The terminal 3 in the third embodiment acquires target data as in the first embodiment, and also acquires selection data necessary for selecting a consumption estimation model. The selection data indicates the type of the target building 4 and the time when the plurality of total consumption amounts indicated by the target data were measured. Here, the target data is data acquired together with the selection data. The period is the month indicated by the clock section at the time when the wattmeter measures the total consumption. A consumption estimation model can be selected based on the selection table T1 and selection data.

The user causes the terminal 3 to acquire the target data and selection data, for example, via communication or using a portable memory. The user causes the target data and selection data to be transmitted to the server 2 via the network N.

FIG. 12 is a flowchart showing the procedure of estimation processing. The control unit 22 periodically executes the estimation process as in the first embodiment.
A part of the estimation process in the third embodiment is common to the estimation process in the first embodiment. Therefore, in the estimation processing in the third embodiment, a detailed explanation of the parts common to the estimation processing in the first embodiment, that is, steps S12 to S15 will be omitted.

In the estimation process, the control unit 22 first determines whether the communication unit 20 has received target data and selection data from the terminal 3 (step S31). When the control unit 22 determines that the communication unit 20 has not received the target data and the selection data (S31: NO), the control unit 22 ends the estimation process. When the next cycle has arrived, the control unit 22 executes step S31 again. The control unit 22 repeatedly executes the estimation process and waits until the communication unit 20 receives the target data and the selection data.

When the control unit 22 determines that the communication unit 20 has received the target data and the selection data (S31: YES), the control unit 22 selects the plurality of consumption estimation models 60a, 60b, . . , a consumption estimation model corresponding to the type and period indicated by the selection data received by the communication unit 20 is selected (step S32). For example, when the selection data indicates an office and January as the type and time, in step S32, the control unit 22 selects the consumption estimation model 60a based on the selection table T1. After executing step S32, the control unit 22 sequentially executes steps S12 to S15.

In step S12, the control unit 22 inputs the target data received by the communication unit 20 into the consumption amount estimation model selected in step S32. In step S13, the control unit 22 calculates, as the output of the consumption estimation model selected in step S32, a plurality of individual consumption amounts indicating the average value of the plurality of total consumption amounts indicated by the target data or a breakdown of the total consumption by use. do.

In the third embodiment, the control unit 22 executes step S14 in the same manner as in the first embodiment. In the first example of step S14, a plurality of first threshold values corresponding to the purpose, type, and time are set in advance. The control unit 22 compares each individual consumption amount calculated in step S13 with the first threshold value for the common purpose, type, and time. As an example, for a common use, type and time, the first threshold is set to the individual consumption consumed in a standard building. As another example, each first threshold value is set to the average value of a plurality of individual consumption amounts corresponding to a plurality of buildings 5 of a common type and having a common purpose and time.

In the second example of step S14, a plurality of second threshold values corresponding to the purpose, type, and time are set in advance. The control unit 22 compares the occupancy rate of each individual consumption amount calculated in step S13 with the second threshold value for the common purpose, type, and time. As an example, for common uses, types and times, the second threshold is set to the occupancy of individual consumption consumed in a standard building. As another example, each second threshold value is set to the average value of the occupancy rates of a plurality of individual consumption amounts corresponding to a plurality of buildings 5 of a common type and having a common purpose and time.

In the estimation process in the third embodiment, in order to estimate a plurality of individual consumption amounts, the type of the target building 4 and the plurality of total consumptions indicated by the target data are used among the plurality of consumption estimation models 60a, 60b,... Since a consumption estimation model corresponding to the time when the amount was measured is used, multiple individual consumption amounts can be estimated more accurately. The server 2 in the third embodiment has the same effects as the server 2 in the first embodiment.

Note that in the third embodiment, the time is not limited to the month, and may be, for example, the season. If the period is seasonal, for example, the period of January, February and December corresponds to winter, the period from March to May corresponds to spring, and the period from June to August corresponds to summer. Correspondingly, the period from September to November corresponds to autumn. In this case, in the selection table T1, the consumption estimation models for January, February, and December corresponding to the common type are common, and the consumption estimation models for March to May corresponding to the common type are common. They have something in common. Furthermore, consumption estimation models from June to August corresponding to common types are common, and consumption estimation models from September to November corresponding to common types are common. If the period is a season, one of spring, summer, autumn, and winter may be indicated in the period field of selection table T1.

(Embodiment 4)
In the third embodiment, the time is the month or season corresponding to the time when the power meter measures the total consumption. However, time is not limited to months or seasons.
In the following, differences between the fourth embodiment and the third embodiment will be explained. Other configurations other than those to be described later are the same as in the third embodiment, so the same reference numerals as in the third embodiment are given to the same components as in the third embodiment, and the explanation thereof is omitted. do.

The storage unit of the target building 4 stores calendar information indicating days that are weekdays and days that are holidays. The terminal 3 in the fourth embodiment acquires target data and selection data similarly to the third embodiment. The period indicated by the selection data is either a weekday or a holiday. If the day on which the wattmeter measures the total consumption is a weekday, the selection data indicates a weekday as the time. If the day on which the power meter measures the total consumption is a holiday, the selection data indicates the holiday as the time.

FIG. 13 is an explanatory diagram of the selection table T1 in the fourth embodiment. In the fourth embodiment, the time field of the selection table T1 indicates either weekdays or holidays as the time.

In the fourth embodiment, the control unit 22 of the server 2 executes teacher data storage processing, model generation processing, and estimation processing in the same manner as in the third embodiment. The server 2 in the fourth embodiment has the same effects as the server 2 in the third embodiment.

In addition, in Embodiments 3 and 4, the content indicated by the selection data may be either the type or the time. It is assumed that the selection data indicates the type of target building 4. In this case, the selection table T1 is configured by a type field and a consumption amount estimation model field. In the selection table T1, one of the plurality of consumption estimation models 60a, 60b, . . . is associated with each of the plurality of types regarding the target building 4. In step S32 of the estimation process, the control unit 22 of the server 2 selects the consumption estimation model corresponding to the type indicated by the selection data received by the communication unit 20 from among the plurality of consumption estimation models 60a, 60b, . Select a model. Even in this case, the consumption estimation model corresponding to the type of the target building 4 is used among the plurality of consumption estimation models 60a, 60b, . . . to estimate the plurality of individual consumptions. individual consumption is estimated more accurately. A plurality of first threshold values and a plurality of second threshold values corresponding to the purpose and type are set in advance.

It is assumed that the selection data indicates when multiple total consumption amounts were measured. In this case, the selection table T1 is configured by a period field and a consumption estimation model field. In the selection table T1, one of the plurality of consumption estimation models 60a, 60b, . . . is associated with each period in which a plurality of total consumption amounts were measured. In step S32 of the estimation process, the control unit 22 of the server 2 selects a consumption estimation model corresponding to the period indicated by the selection data received by the communication unit 20 from among the plurality of consumption estimation models 60a, 60b, . Select a model. Even in this case, when estimating multiple individual consumption amounts, among the multiple consumption estimation models 60a, 60b,... Since a quantity estimation model is used, multiple individual consumption quantities are estimated more accurately. A plurality of first threshold values and a plurality of second threshold values corresponding to the purpose and time are set in advance.

Further, in the third and fourth embodiments, each of the consumption estimation models 60a, 60b, . . . may be configured similarly to the consumption estimation model 60 in the second embodiment. Therefore, each of the target data and time-series data may be data that shows weather information and/or floor area in a time-series manner in addition to the total amount of electricity consumed per unit time within a predetermined period. good. As described in the description of the third embodiment, each of the consumption estimation models 60a, 60b, . This is a trained model that has learned the relationship with multiple individual consumption amounts that show the breakdown. In the third and fourth embodiments, if each of the consumption estimation models 60a, 60b, . . . is configured similarly to the consumption estimation model 60 in the second embodiment, the server 2 in the third and fourth embodiments This similarly produces the same effect as the server 2 in the second embodiment.

Furthermore, in Embodiments 3 and 4, examples of types include offices, factories, commercial facilities, and residences. However, types may be further subdivided. For example, instead of commercial facilities, multiple types of buildings such as department stores, museums, and art galleries may be shown.

(Embodiment 5)
In the first embodiment, the consumption estimation model 60 is not limited to a neutral network related to LSTM.
Below, the differences from the first embodiment will be explained regarding the fifth embodiment. Other configurations other than those described below are the same as in Embodiment 1, so the same reference numerals as in Embodiment 1 are given to the components common to Embodiment 1, and the explanation thereof is omitted. do.

FIG. 14 is an explanatory diagram of the consumption estimation model 60 in the fifth embodiment. The consumption estimation model 60 in the fifth embodiment is, for example, a CNN (Convolutional Neural Network). The consumption estimation model 60 has an input layer, a plurality of intermediate layers, and an output layer. FIG. 14 shows an example in which the number of intermediate layers is two. Note that the number of intermediate layers may be three or more.

Each of the input layer, hidden layer, and output layer has one or more neurons. Neurons in each layer are unidirectionally connected to neurons in the previous and following layers with desired weights and biases. A vector having the same number of components as the number of neurons in the input layer is input to the consumption estimation model 60 as input data. The input layer of the consumption estimation model 60 receives target data or time series data, that is, a plurality of total consumption amounts.

Note that the plurality of intermediate layers may include a layer that performs convolution. In this layer, an appropriate filter including a convolution filter is applied to the input data, and a plurality of data obtained after applying the filter is output to the next layer.

The input values input to each neuron of the input layer are input to the first hidden layer. In this intermediate layer, the output is calculated using an activation function that includes weights and biases. The calculated value is input to the next intermediate layer. In the same manner, the output is successively transmitted to subsequent layers until the output of the output layer is calculated. Note that parameters such as weights and biases that connect neurons are learned using an error backpropagation method.

As in the first embodiment, the consumption estimation model 60 calculates the relationship between time-series data and a plurality of individual consumption amounts indicating a breakdown of each usage with respect to the average value or total of a plurality of total consumption amounts indicated by the time-series data. This is a trained model that has learned . The control unit 22 of the server 2 performs machine learning using teacher data including time series data and correct answer data to generate a consumption estimation model 60. In machine learning, the control unit 22 optimizes parameters used for calculation in the intermediate layer. The teacher data in the fifth embodiment is the same as the teacher data in the first embodiment, an example of which is shown in FIG.

The output layer of the consumption estimation model 60 in Embodiment 5 is an average value of a plurality of total consumption amounts indicated by the target data or time series data input to the input layer for each of a plurality of combinations regarding a plurality of individual consumption amounts. Outputs the probability corresponding to a combination of multiple individual consumption amounts showing the breakdown of each usage for the total.

As described in the description of the first embodiment, the application is to supply power to the outdoor unit of the air conditioner, the indoor unit of the air conditioner, the water heater, electric lights, office equipment, etc. In the example of FIG. 14, the top neuron of the output layer outputs the probability that the individual consumption amounts of the outdoor unit, indoor unit, etc. are 0 kWh, 10 kWh, etc. “kWh” is a unit related to power consumption.

The control unit 22 of the server 2 calculates the probability of a combination regarding a plurality of individual consumption amounts by inputting the target data into the input layer of the consumption amount estimation model 60. For example, the control unit 22 calculates a plurality of individual consumption amounts corresponding to the combination with the highest probability for each use with respect to the average value or total of a plurality of total consumption amounts indicated by the target data or time series data input to the input layer. It is estimated that there are multiple individual consumption amounts showing the breakdown.

In the fifth embodiment, the processing element of the control unit 22 of the server 2 executes the teacher data storage process, model generation process, and estimation process by executing the computer program P1, as in the first embodiment. The teacher data storage process in the fifth embodiment is similar to the teacher data storage process in the first embodiment. In the teacher data storage process, the control unit 22 acquires the teacher data received by the communication unit 20 or input to the input unit, and stores it in the acquired storage unit 21 . Thereafter, the control unit 22 ends the teacher data storage process.

FIG. 15 is a flowchart showing the procedure of model generation processing. In the fifth embodiment, similarly to the first embodiment, the control unit 22 of the server 2 executes the model generation process when instructed to execute the model generation process with the teacher data stored in the storage unit 21. Execute.
A part of the model generation process in the fifth embodiment is similar to the model generation process in the first embodiment. In the model generation process in the fifth embodiment, a detailed explanation of the parts common to the model generation process in the first embodiment, that is, steps S1, S2, and S6 will be omitted.

In step S2 of the model generation process, similarly to the first embodiment, the control unit 22 inputs the time series data included in the teacher data selected in step S1 to the consumption amount estimation model 60. Next, the control unit 22 calculates the probabilities of a plurality of combinations regarding a plurality of individual consumption amounts as outputs of the consumption estimation model 60 (step S41). These probabilities are calculated based on the average value of multiple total consumption amounts indicated by the time-series data input to the input layer, or multiple individual consumption amounts showing the breakdown of each use for the total, for each of multiple combinations regarding multiple individual consumption amounts. This is the probability that the combination of

Next, the control unit 22 updates the parameters used for calculation in the intermediate layer based on the correct answer data of the teacher data selected in step S1 and the calculation result in step S41 (step S42). Specifically, in step S<b>42 , the control unit 22 adjusts the output of the consumption estimation model 60 so that the probability of a combination of individual consumption amounts closest to the individual consumption amounts indicated by the correct data is increased. Update parameters. After executing step S42, the control unit 22 executes step S6.

The consumption estimation model 60 corresponds to a combination of a plurality of individual consumption amounts that indicates a breakdown for each use of the average value of a plurality of total consumption amounts indicated by the target data or time series data for each combination of a plurality of individual consumption amounts. Suppose we want to output probabilities. In this case, the correct data is data indicating a plurality of individual consumption amounts that show a breakdown for each use with respect to the average value of a plurality of total consumption amounts.

Similarly, for each combination of a plurality of individual consumption amounts, the consumption estimation model 60 generates a combination of a plurality of individual consumption amounts that shows a breakdown of each use for the total of a plurality of total consumption amounts indicated by the target data or time series data. Assume that the corresponding probability is output. In this case, the correct data is data indicating a plurality of individual consumption amounts that indicate a breakdown of the total consumption amount for each use.

Also in the model generation process in the fifth embodiment, the control unit 22 generates the consumption estimation model 60 based on the teacher data acquired in the teacher data storage process.
Note that in the model generation process in the fifth embodiment as well, the number of times parameters are updated using one teaching data is not limited to one. The control unit 22 may update the parameters two or more times using one piece of teacher data.

FIG. 16 is a flowchart showing the procedure of estimation processing. In the fifth embodiment, the control unit 22 periodically executes the estimation process as in the first embodiment.
A part of the estimation process in the fifth embodiment is similar to the estimation process in the first embodiment. In the estimation process in the fifth embodiment, detailed explanations of the parts similar to the estimation process in the first embodiment, that is, steps S11, S12, S14, and S15 will be omitted.

In step S12 of the estimation process, similarly to the first embodiment, the control unit 22 inputs the target data received by the communication unit 20 into the consumption amount estimation model 60. Next, the control unit 22 calculates the probabilities of a plurality of combinations regarding a plurality of individual consumption amounts as outputs of the consumption estimation model 60 (step S51). These probabilities are calculated based on the average value of multiple total consumption amounts indicated by the target data input to the input layer, or the breakdown of multiple individual consumption amounts for each use for the total, for each of multiple combinations regarding multiple individual consumption amounts. This is the probability that the combination applies.

Next, based on the probability calculated in step S51, the control unit 22 determines, for each of the plurality of combinations regarding the plurality of individual consumption amounts, the average value or the sum of the plurality of total consumption amounts indicated by the target data input to the input layer. A plurality of individual consumption amounts showing the breakdown for each use are estimated (step S52). In step S52, for example, the control unit 22 converts the plurality of individual consumption amounts corresponding to the combination with the highest probability into the plurality of individual consumption amounts indicating the average value of the plurality of total consumption amounts indicated by the target data or the breakdown of each use with respect to the total. Estimate as consumption amount.

After executing step S52, the control unit 22 sequentially executes steps S14 and S15. In step S14, the control unit 22 determines whether each individual consumption amount estimated in step S52 is excessive. In step S15, similarly to Embodiment 1, the control unit 22 instructs the communication unit 20 to transmit the determination result data and estimation data to the terminal 3 via the network N. The determination result data indicates the determination result made in step S14 for each individual consumption amount estimated in step S52. The estimated data indicates a plurality of individual consumption amounts estimated by the control unit 22 in step S52.

By executing the estimation process, the control unit 22 allows the user to determine the average value or total of the plurality of total consumption amounts indicated by the target data for each use based on the contents of the estimation data, as in the first embodiment. You can check the breakdown. The user can confirm the purpose of the excessive individual consumption amount based on the content of the determination result data.
The server 2 in the fifth embodiment has the same effects as the server 2 in the first embodiment.

Note that in the second embodiment, the consumption estimation model 60 may be configured in the same manner as in the fifth embodiment. In this case, in each of the target data and the time series data, in addition to the total amount of power consumed per unit time within a predetermined period, both or one of the weather information and the floor area is shown in the time series. In the generation process, similarly to the fifth embodiment, parameters are updated based on a plurality of probabilities output by the consumption estimation model 60. In the estimation process, as in the fifth embodiment, a plurality of individual consumption amounts are estimated based on a plurality of probabilities output by the consumption estimation model 60.

Further, in the third and fourth embodiments, each of the plurality of consumption estimation models 60a, 60b, . . . may be configured similarly to the fifth embodiment. In the generation process, as in the fifth embodiment, the selected consumption amount is determined based on the plurality of probabilities output by the consumption amount estimation model selected from the plurality of consumption estimation models 60a, 60b, . Update the estimated model parameters. In the estimation process, as in the fifth embodiment, a plurality of individual consumption amounts are estimated based on a plurality of probabilities output by a consumption amount estimation model selected from among the plurality of consumption amount estimation models 60a, 60b, . Estimate.

Note that in the fifth embodiment, the consumption estimation model 60 is not limited to a trained model generated using time-series data, but is, for example, a model generated using the average value or sum of a plurality of total consumption amounts. It may be a trained model. In this case, the teacher data includes consumption data and correct answer data indicating an average value of a plurality of total consumption amounts measured in time series in the building 5 or a representative value of these total consumption amounts. In the model generation process, consumption data is used instead of time series data. The target data indicates an average value of a plurality of total consumption amounts measured in time series in the target building 4, a representative value of these total consumption amounts, or the like. Even if the consumption estimation model 60 is not a trained model generated using time-series data, there is no need to measure individual consumption for each use, and a configuration for estimating multiple individual consumptions is possible. Realized at low cost.

Further, in the case where the consumption estimation model 60 in the second embodiment and the plurality of consumption estimation models 60a, 60b in the third and fourth embodiments are configured in the same manner as in the fifth embodiment, these consumption The quantity estimation model may be a trained model generated using data different from time-series data, for example, consumption data.

In the first to fifth embodiments, the device that estimates a plurality of individual consumption amounts and generates the consumption amount estimation model 60 or the consumption amount estimation models 60a, 60b, . . . may be an information processing device. However, it is not limited to server 2. A personal computer may be used instead of the server 2. Furthermore, the estimation device that estimates a plurality of individual consumption amounts and the generation device that generates the consumption amount estimation model 60 or the consumption amount estimation models 60a, 60b, . . . may be different. In this case, the consumption estimation model stored in the estimation device is updated to the consumption estimation model generated by the generation device.

The technical features (constituent features) described in Embodiments 1 to 5 can be combined with each other, and new technical features can be formed by combining them.
The disclosed embodiments 1 to 5 are illustrative in all respects and should not be considered restrictive. The scope of the present invention is indicated by the claims rather than the above-mentioned meaning, and is intended to include meanings equivalent to the claims and all changes within the scope.

2 Server (estimation device)
4 Target building 5 Building 22 Control unit (acquisition unit, input unit, estimation unit)
60, 60a, 60b,... Consumption estimation model (trained model)
P1 Computer program

Claims (11)

Obtain the total amount of electricity consumed within a predetermined period in the target building that is the target of estimation,
The obtained total consumption amount is input into a trained model that has learned the relationship between the total amount of electricity consumed in the building and multiple individual consumption amounts that indicate the breakdown of the total consumption amount for each use of the building. ,
Estimating a plurality of individual consumption amounts corresponding to the target building based on the output of the trained model ,
Determine whether the estimated individual consumption amount is excessive by determining whether the occupancy rate of the estimated individual consumption amount is greater than or equal to a second threshold value.
An estimation method in which the processing is performed by a computer. obtaining target data showing the total amount of power consumed per unit time in the target building within the predetermined period in time series;
The acquired target data is divided into time-series data showing the total power consumption per unit time in the building, and multiple individual consumption data showing the breakdown of the total power consumption of the building by usage. 2. The estimation method according to claim 1, wherein a computer executes the process of inputting the relationship into the learned model. Obtaining target data showing the total amount of electricity consumed per unit time in the target building within the predetermined period and weather information regarding the weather around the target building in chronological order;
The acquired target data is divided into time-series data showing the total amount of electricity consumed by the building per unit time, weather information around the building in chronological order, and a breakdown of the total amount of electricity consumed by the building by usage. 3. The estimation method according to claim 1 or 2, wherein a computer executes the process of inputting the relationship between the individual consumption amount and the individual consumption amount to the learned model. Obtaining target data indicating the total amount of power consumed per unit time in the target building within the predetermined period and the floor area of the target building in chronological order;
The acquired target data is divided into time-series data showing the total amount of electricity consumed in the building per unit time, the floor area of the building in chronological order, and a breakdown of the total consumption of the building by use. The estimation method according to any one of claims 1 to 3 , wherein a computer executes the process of inputting the relationship with a plurality of individual consumption amounts shown into a learned model. Selecting a trained model corresponding to the type of target building from among a plurality of trained models that correspond to a plurality of types of buildings and have learned the relationship;
The estimation method according to any one of claims 1 to 4, wherein a computer executes the process of inputting the obtained total consumption into the selected trained model. Selecting a trained model corresponding to a period in which the total consumption of the target building was measured from among a plurality of trained models that correspond to a plurality of periods and have learned the relationship;
The estimation method according to any one of claims 1 to 5, wherein a computer executes the process of inputting the obtained total consumption amount into the selected trained model. an acquisition unit that acquires the total amount of electricity consumed within a predetermined period in the target building that is the estimation target;
A trained device that has learned the relationship between the total consumption amount acquired by the acquisition unit, the total amount of electricity consumed in the building, and a plurality of individual consumption amounts showing a breakdown of the total consumption amount for each use of the building. an input section for inputting into the model;
an estimation unit that estimates a plurality of individual consumption amounts corresponding to the target building based on the output of the learned model ;
a determination unit that determines whether the estimated individual consumption is excessive by determining whether the estimated individual consumption occupancy is equal to or higher than a second threshold;
An estimation device comprising: Obtain the total amount of electricity consumed within a predetermined period in the target building that is the target of estimation,
The obtained total consumption amount is input into a trained model that has learned the relationship between the total amount of electricity consumed in the building and multiple individual consumption amounts that indicate the breakdown of the total consumption amount for each use of the building. ,
Estimating a plurality of individual consumption amounts corresponding to the target building based on the output of the trained model ,
Determine whether the estimated individual consumption amount is excessive by determining whether the occupancy rate of the estimated individual consumption amount is greater than or equal to a second threshold value.
A computer program that causes a computer to perform a process. Time-series data showing the total amount of electricity consumed in the building per unit time and the floor area of the building in chronological order, as well as multiple individual consumption amounts showing a breakdown of the total amount of electricity consumed by the building for each use. Get the teacher data associated with
Based on the acquired teacher data, the target building is estimated by inputting target data indicating the total amount of electricity consumed per unit time in the target building , which is the estimation target, and the floor area of the target building in time series. A trained model generation method in which a computer executes the process of generating a trained model used in the process of estimating multiple individual consumption amounts corresponding to . Obtaining target data showing the total amount of power consumed per unit time in a target building, which is the estimation target, within a predetermined period, and the floor area of the target building in chronological order;
The acquired target data is divided into time-series data showing the total amount of electricity consumed in the building per unit time, the floor area of the building in chronological order, and a breakdown of the total consumption of the building by use. Input the relationships with multiple individual consumption amounts shown in the learned model into the learned model,
Estimating a plurality of individual consumption amounts corresponding to the target building based on the output of the trained model.
An estimation method in which the processing is performed by a computer. Obtain the total amount of electricity consumed within a predetermined period in the target building that is the target of estimation,
Multiple learned models that correspond to multiple types of buildings and have learned the relationship between the total amount of electricity consumed in the building and multiple individual consumption amounts that show the breakdown of the total consumption of the building by use. Select a trained model corresponding to the type of target building from among the models,
Input the obtained total consumption into the selected trained model,
Estimating a plurality of individual consumption amounts corresponding to the target building based on the output of the trained model.
An estimation method in which the processing is performed by a computer.

Images