JP6904072B2 - Learning device, attribute estimation device, learning method, attribute estimation method and program - Google Patents

Description

Embodiments of the present invention relate to a learning device, an attribute estimation device, a learning method, an attribute estimation method, and a program.

Consider promoting the distribution of existing homes. In order to promote the distribution of used homes, it is necessary to acquire the quality and performance of used homes. Field surveys, infrared thermal image diagnosis, indoor and outdoor temperature measurement, etc. are known as methods for acquiring the quality and performance of existing homes.
In addition, the quality and performance of existing homes can be estimated to some extent from the attribute information of consumers such as family composition, living dynamics, and lifestyle of consumers living in existing homes.

Regarding a technique for estimating the attribute information of a consumer, a technique for estimating the user attribute of the user with higher reliability based on the power consumption by the user is known (see, for example, Patent Document 1). According to this technology, weather-related values are acquired, the amount of activity power of the user is calculated, a feature time representative vector representing the amount of activity power in a feature time zone in which a feature can occur according to the weather is generated, and the activity power is generated. An estimation model is generated by generating a feature vector containing a vector element, a feature time representative vector element, and a weather-related value as elements, and performing machine learning using a teacher dataset containing a pair of the feature vector and a user attribute value as teacher data. Is constructed, the feature vector related to the estimation target user is input to the estimation model, and the estimated value of the user attribute of the estimation target user is acquired.

JP 2015-162229

There are the following problems in conducting field surveys, infrared thermal image diagnosis, indoor / outdoor temperature measurement, etc. in order to acquire attribute information such as quality and performance of used houses.
In order to conduct a field survey, it is necessary to go to the site where the building is built. In addition, in order to perform thermal image diagnosis using infrared rays and indoor / outdoor temperature measurement, it is necessary to go to the site where the building is built and install a measuring device inside or outside the building for a certain period of time. .. In either case, the quality and performance of existing homes cannot be easily evaluated.
In addition, in order to estimate the quality and performance of a building from the attribute information of consumers, it is necessary to formulate a hypothesis and verify the hypothesis each time. Hypothesis planning, etc. largely depends on the experience and skills of the analyst.
These problems are not limited to promoting the distribution of existing homes, but also apply to the acquisition of home attribute information in order to promote home remodeling. In short, these issues apply to pre-owned homes, homes and other buildings.
The present invention has been made in view of the above points, and an object of the present invention is a time-series data of household electricity usage and a period during which each of a plurality of electricity usages included in the time-series data is acquired. It is to estimate the attribute information of the household based on the temperature information of.

One aspect of the present invention is the time-series data of the electricity usage of the household and the temperature information of the period in which each of the plurality of electricity usages included in the time-series data is acquired and the household in which the time-series data is obtained. Based on the relationship between the acquisition unit that acquires the attribute information, the time series data acquired by the acquisition unit, and the temperature information, the first electricity information indicating the amount of electricity used at the representative temperature in winter and the summer A calculation unit that calculates the second electric information indicating the amount of electricity used at the representative temperature of the above and the third electric information indicating the minimum value of the amount of electricity used, the first electric information, the second electric information, and the above. It is a learning device including a learning unit that performs machine learning using the attribute information of the household as teacher data based on the third electrical information.
In the learning device of one aspect of the present invention, the acquisition unit acquires the attribute information of the building in which the consumers constituting the household live.
In the learning device of one aspect of the present invention, the learning unit performs machine learning using any or a combination of the first electrical information, the second electrical information, and the third electrical information as explanatory variables. ..
In the learning device of one aspect of the present invention, the learning device performs decision tree learning.
In the learning device of one aspect of the present invention, each of the plurality of electricity consumptions included in the time series data is acquired every period of one week or more and one month or less.
In the learning device of one aspect of the present invention, the air temperature information is an average value of the air temperature during the period for acquiring each of the plurality of electricity usages included in the time series data.
In the attribute estimation device of one aspect of the present invention, the time-series data of the electricity usage of the household whose attribute information is unknown and the temperature information of the period in which each of the plurality of electricity usages included in the time-series data are acquired. Based on the relationship between the acquisition unit that acquires the data, the time-series data acquired by the acquisition unit, and the temperature information, the first electricity information indicating the amount of electricity used at the representative temperature in winter and the representative temperature in summer. In the time-series data and the time-series data including a plurality of information indicating the amount of electricity used at home, a calculation unit that calculates the second electric information indicating the amount of electricity used at the time of, and the third electric information indicating the minimum value. The amount of electricity used at the representative winter temperature, the amount of electricity used at the representative temperature of summer, and the amount of electricity used, which were obtained from the relationship with the temperature information during the period when each of the multiple electricity usages included was acquired. based on the minimum value of the quantity, the attribute information of the home time series data is obtained as teacher data, using the results obtained by machine learning, the said first electric information and the second electric information first (Iii) An attribute estimation device including an estimation unit that estimates the attribute information of the household from which the time-series data acquired by the acquisition unit has been obtained based on the electrical information.
In the learning method of one aspect of the present invention, the temperature information and the time-series data of the period in which each of the time-series data of the household electricity usage and the plurality of electricity usages included in the time-series data are acquired can be obtained. First electricity information indicating the amount of electricity used at the representative temperature in winter based on the relationship between the step of acquiring the attribute information of the household and the time series data acquired in the acquisition step and the temperature information. And the step of calculating the second electric information indicating the amount of electricity used at the representative temperature in summer and the third electric information indicating the minimum value of the amount of electricity used, and the first electric information and the second electric information. based on said third electrical information, the attribute information of the home as teacher data, and a step of machine learning, a learning method learning device performs.
In the attribute estimation method of one aspect of the present invention, the time-series data of the electricity usage of the household whose attribute information is unknown and the temperature information of the period in which each of the plurality of electricity usages included in the time-series data are acquired. Based on the relationship between the step of acquiring the data, the time-series data acquired in the acquisition step, and the temperature information, the first electrical information indicating the amount of electricity used at the representative temperature in winter and the representative temperature in summer. The step of calculating the second electric information indicating the amount of electricity used at the time and the third electric information indicating the minimum value, and the time-series data including a plurality of information indicating the amount of electricity used at home and the time-series data included in the time-series data. The amount of electricity used at the representative winter temperature, the amount of electricity used at the representative temperature of summer, and the amount of electricity used, which were obtained from the relationship with the temperature information during the period when each of the multiple electricity usages was acquired. based of the and the minimum value, time series data using the result obtained by the machine learning the attribute information of the home was obtained as teacher data, wherein a first electric information and the second electric information third electric It is an attribute estimation method executed by an attribute estimation device, which includes a step of estimating the attribute information of the household from which the time series data acquired in the acquisition step is obtained based on the information.
One aspect of the present invention is the temperature information and the time series during the period in which each of the time-series data of the household electricity usage and the plurality of electricity usages included in the time-series data are acquired by the computer provided in the learning device. Based on the relationship between the step of acquiring the attribute information of the household from which the data was obtained and the time-series data acquired in the step of acquiring the data and the temperature information, the amount of electricity used at the representative temperature in winter is shown. A step of calculating the first electric information, the second electric information indicating the amount of electricity used at the representative temperature in summer, and the third electric information indicating the minimum value of the amount of electricity used, the first electric information and the above. It is a program that executes a step of machine learning using the attribute information of the household as teacher data based on the second electric information and the third electric information.
One aspect of the present invention is a period during which each of the time-series data of the electricity usage of a household whose attribute information is unknown and the plurality of electricity usages included in the time-series data are acquired by the computer provided in the attribute estimation device. Based on the step of acquiring the temperature information of the above, and the relationship between the time-series data acquired in the acquisition step and the temperature information, the first electricity information indicating the amount of electricity used at the representative temperature in winter, and the first electricity information. A step to calculate the second electric information indicating the amount of electricity used at the representative temperature in summer and the third electric information indicating the minimum value, and time-series data including a plurality of information indicating the amount of electricity used at home and the time. The amount of electricity used at the representative temperature in winter and the amount of electricity used at the representative temperature in summer obtained from the relationship with the temperature information during the period when each of the multiple electricity usages included in the series data was acquired. , based on the minimum value of the electricity consumption, using the results obtained by the machine learning the attribute information of the home time series data is obtained as the teacher data, and the first electric information and the second electric information It is a program that executes a step of estimating the attribute information of the household from which the time-series data acquired in the acquisition step is obtained based on the third electrical information.

According to the embodiment of the present invention, the attribute information of the household is provided based on the time-series data of the electricity usage of the household and the temperature information of the period in which each of the plurality of electricity usages included in the time-series data is acquired. Can be estimated.

It is a figure which shows the learning system which concerns on embodiment. It is a figure which shows an example of the hardware configuration of the learning apparatus which concerns on embodiment. It is a figure which shows an example of the attribute information estimation table. It is a functional block diagram of the learning apparatus which concerns on embodiment. It is a figure which shows an example (the 1) of the relationship between the air temperature and the amount of electricity used. It is a figure which shows an example (the 1) of the machine learning executed by the learning apparatus which concerns on embodiment. It is a figure which shows an example (the 2) of the relationship between the air temperature and the amount of electricity used. It is a figure which shows an example (the 2) of the machine learning executed by the learning apparatus which concerns on embodiment. It is a figure which shows an example (the 3) of the machine learning executed by the learning apparatus which concerns on embodiment. It is a figure which shows an example (the 4) of the machine learning executed by the learning apparatus which concerns on embodiment. It is a flowchart which shows an example of the operation of the learning apparatus which concerns on embodiment. It is a figure which shows an example of the relationship between the ratio of all customers and the cumulative ratio with gas floor heating. It is a figure which shows an example of the hardware composition of the attribute estimation apparatus which concerns on embodiment. It is a functional block diagram of the attribute estimation device which concerns on embodiment. It is a flowchart which shows an example of the operation of the attribute estimation apparatus which concerns on embodiment. It is a figure which shows the operation of the attribute estimation system which concerns on embodiment.

Next, a mode for carrying out the present invention will be described with reference to the drawings. The embodiments described below are merely examples, and the embodiments to which the present invention is applied are not limited to the following embodiments.
In all the drawings for explaining the embodiment, the same reference numerals are used for those having the same function, and the repeated description will be omitted.

(First Embodiment)
(Learning system)
In the learning system according to the embodiment, the time-series data of the electricity usage of each of the plurality of households, the temperature information of the period in which each of the plurality of electricity usages included in the time-series data is acquired, and the time-series data are obtained. Acquire the obtained attribute information of the house or building. Based on the relationship between the acquired time-series data and the acquired temperature information, the learning system provides information indicating the amount of electricity used at the representative temperature in winter (hereinafter referred to as "first electricity information") and the representative temperature in summer. Information indicating the amount of electricity used at that time (hereinafter referred to as "second electricity information") and information indicating the minimum value of electricity consumption (hereinafter referred to as "third electricity information") are calculated. An example of a typical winter temperature is the lowest temperature, and an example of a typical summer temperature is the highest temperature. The learning system obtains variables to be used for explaining the attribute information by machine learning the attribute information as teacher data based on the first electric information, the second electric information, and the third electric information. The learning system associates and stores the attribute information with the variables used to explain the attribute information.

FIG. 1 is a diagram showing a learning system according to an embodiment. The learning system 1 includes a learning device 100 and an electricity usage data server 200. FIG. 1 also shows the attribute estimation device 300. The attribute estimation device 300 will be described later. The learning device 100, the electricity usage data server 200, and the attribute estimation device 300 are connected via a communication network 50 such as the Internet.
The electricity usage data server 200 stores information indicating the past electricity usage. Specifically, the electricity usage data server 200 acquires information indicating the electricity usage from a system capable of outputting the electricity usage to the outside, such as a smart meter and HEMS, at a predetermined cycle. The electricity usage data server 200 stores the acquired information indicating the electricity usage in association with the acquisition time and the location information of the building in which the system capable of outputting electricity to the outside is installed.
The learning device 100 acquires time-series data of the electricity usage of each of the plurality of households. Here, the information indicating the amount of electricity used is acquired every period of one week or more and one month or less, and is continuously acquired for a period of about one year. The learning device 100 creates information requesting electricity usage (hereinafter referred to as "electricity usage request information"), and transmits the created electricity usage request information to the electricity usage data server 200. Here, the electricity usage request information includes information for identifying a building in which a system capable of outputting electricity to the outside is installed, information indicating a period for acquiring time-series data of electricity usage, and electricity. It contains information indicating the cycle in which the usage amount is acquired.
The electricity usage data server 200 receives the electricity usage request information transmitted by the learning device 100, and acquires the information for identifying the building included in the received electricity usage request information and the time series data of the electricity usage. When the time-series data of the corresponding electricity usage is acquired from the stored information indicating the electricity usage based on the information indicating the period and the information indicating the cycle in which the electricity usage is acquired. The series data is transmitted to the learning device 100.
The learning device 100 receives the series data transmitted by the electricity usage data server 200.

In addition, the learning device 100 acquires temperature information during the period in which each of the plurality of electricity usage amounts included in the acquired time-series data of the electricity usage amount is acquired. Here, the air temperature information is the average value of the air temperature during the period when each of the electricity usage is acquired. Specifically, the learning device 100 acquires the temperature information included in the past meteorological data by accessing a site that provides meteorological information such as the Japan Meteorological Agency.
In addition, the learning device 100 acquires the attribute information of the home or building from which the acquired time series data has been obtained. Here, an example of home or building attribute information is home attribute information such as household information, or building attribute information such as housing performance, housing equipment, gas usage, and window type.
Further, the learning device 100 obtains the relationship between the acquired time series data and the acquired temperature information. The learning device 100 calculates the first electric information, the second electric information, and the third electric information from the relationship between the time series data and the temperature information.
The learning device 100 performs machine learning using attribute information as teacher data based on the first electrical information, the second electrical information, and the third electrical information. The learning device 100 stores the machine learning result and the attribute information in association with each other.

(Configuration of learning device)
FIG. 2 is a diagram showing an example of the hardware configuration of the learning device according to the embodiment.
The learning device 100 includes a CPU (Central Processing Unit) 102, a ROM (Read Only Memory) 104, a RAM (Random Access Memory) 106, a communication I / F 108, a storage unit 110, an operation unit 112, and a display unit. It is equipped with 114 and a bus 150.
The CPU 102 reads the program 1102 or the like that controls the operation of the learning device 100 from the storage unit 110, expands it into the RAM 106, and executes it.
The communication I / F 108 is composed of a communication module such as an NFC (Near field radio communication) module and a wireless LAN module. Further, the communication I / F 108 may be configured by USB (Universal Serial Bus). The communication I / F 108 communicates with an external device by NFC or wireless LAN. The communication I / F 108 is connected to a smart meter (not shown) and an attribute estimation device 300 described later.
The storage unit 110 is composed of a flash memory, an HDD (Hard Disk Drive), an SSD (Solid State Drive), an SD (Secure Digital) card, or other non-volatile memory. The storage unit 110 stores the program 1102 executed by the CPU 102, the electricity usage time series data 1104, the temperature information 1106, the attribute information 1108, and the attribute information estimation table 1110.

The electricity usage time-series data 1104 is time-series data of the electricity usage of each of a plurality of households. The time series data includes a plurality of information indicating the amount of electricity used. As described above, the plurality of electricity consumptions are acquired every one week or more and one month or less, and are continuously acquired for a period of about one year.
The air temperature information 1106 is the air temperature information during the period in which the electricity usage of each of the plurality of households included in the electricity usage time series data 1104 is acquired. The air temperature information is stored in association with the information indicating the electricity usage of each of the plurality of households included in the electricity usage time series data 1104.
Attribute information 1108 includes household attribute information such as household information and building attribute information such as housing performance, housing equipment, gas usage, and window type. The household attribute information and the building attribute information are stored in association with the information indicating the electricity usage of each of the plurality of households included in the electricity usage time series data 1104. The attribute information estimation table 1110 will be described later.

The operation unit 112 is an input device that accepts user operations, and includes a pointing device such as a touch panel, buttons, dials, touch sensors, touch pads, and the like.
The display unit 114 is configured by, for example, a liquid crystal display or the like, and displays time-series data of electricity usage, home attribute information from which the time-series data has been obtained, building attribute information, and the like.

(Attribute information estimation table)
FIG. 3 is a diagram showing an example of the attribute information estimation table. The attribute information estimation table 1110 is table format data in which the attribute information and the variables used for explaining the attribute information are associated with each other. The variables used to explain the attribute information are based on the first electrical information, the second electrical information, and the third electrical information acquired from the relationship between the temperature information and the amount of electricity used, and the attribute information is used as teacher data. It is obtained by performing machine learning such as decision tree learning, neural network, and deep learning. In this embodiment, as an example of machine learning, a case where decision tree learning is performed will be described as an example.
In the example shown in FIG. 3, the attribute information “gas floor heating possession” is associated with the variables “AAAAA”, “BBBBBB”, and “CCCCC” used to explain the attribute information. This is as shown below when the time-series data of the unknown electricity usage and the temperature information of the period in which each of the information indicating the plurality of electricity usages contained in the time-series data was acquired are given. Indicates that the attribute information is estimated. The relationship between the time-series data and the temperature information is obtained, and the first electric information, the second electric information, and the third electric information are obtained from the obtained relationship between the time-series data and the temperature information. Then, two nodes (points) are formed by differentiating into two edges (branches) by "AAAAA" with the obtained first electric information, second electric information, and third electric information as roots. .. Further, of the two nodes, one of the preset nodes is differentiated into two edges by "BBBBBB" to form two nodes. Further, of the two nodes, one of the preset nodes is differentiated into two edges by "CCCCC" to form two nodes. If one of the two nodes corresponds to a preset node, the building in which the consumers who make up the household from which the unknown time-series data has been obtained resides has gas floor heating. It is presumed that it is.

(Functional configuration of learning device)
FIG. 4 is a functional block diagram of the learning device according to the embodiment.
The learning device 100 functions as an acquisition unit 152, a calculation unit 154, a learning unit 156, and an output unit 158 by executing the program 1102 expanded on the RAM 106 from the storage unit 110 by the CPU 102.

(Each functional configuration of the learning device)
Each functional configuration of the learning device 100 will be described in detail with reference to FIGS. 2 and 4. In the following description of each function of the learning device 100, the relationship with the main hardware for realizing each functional block of the learning device 100 shown in FIG. 4 will also be described.
The acquisition unit 152 is realized by a command from the CPU 102 and a communication I / F 108. The acquisition unit 152 acquires time-series data of the electricity usage of each of the plurality of households. The acquisition unit 152 stores the acquired time-series data of the electricity usage of each of the plurality of households in the electricity usage time-series data 1104 of the storage unit 110. Specifically, the acquisition unit 152 creates the electricity usage request information, and transmits the created electricity usage request information from the communication I / F 108 to the electricity usage data server 200. Then, the acquisition unit 152 acquires the information indicating the electricity usage transmitted by the electricity usage data server 200 from the communication I / F 108.
In addition, the acquisition unit 152 acquires the temperature information during the period in which the acquired time series data is acquired. The acquisition unit 152 stores the acquired air temperature information in the air temperature information 1106 of the storage unit 110. Specifically, the acquisition unit 152 acquires the temperature information included in the past meteorological data from the communication I / F 108 by accessing a site that provides meteorological information such as the Japan Meteorological Agency via the communication network 50. To do. The acquisition unit 152 acquires the temperature information from the temperature information of the area corresponding to the area where the time series data is acquired and during the acquisition period.
In addition, the acquisition unit 152 acquires the attribute information of the home from which the time-series data of the amount of electricity used has been obtained and the attribute information of the building. The acquisition unit 152 stores the acquired home attribute information and building attribute information in the attribute information 1108 of the storage unit 110.

The calculation unit 154 is realized by an instruction from the CPU 102. The calculation unit 154 acquires time-series data of electricity usage for each of the plurality of households from the electricity usage time-series data 1104 of the storage unit 110. In addition, the calculation unit 154 acquires the air temperature information during the period in which the time-series data of the amount of electricity used is acquired for each of the plurality of households. Then, the calculation unit 154 obtains the relationship between the time-series data of the amount of electricity used and the temperature information for each of the plurality of households.
Hereinafter, the process of obtaining the relationship between the time-series data of electricity consumption and the temperature information for one household will be described. The calculation unit 154 obtains the average value of the air temperature from the previous acquisition to the current acquisition of the air temperature when each of the plurality of electricity usages included in the time series data of the electricity usage is acquired. Ask by. Specifically, the calculation unit 154 specifies the date and time when each of the plurality of electricity usages included in the time-series data of the electricity usage was acquired. Then, the calculation unit 154 calculates the average value of the air temperature acquired from the date and time acquired last time to the date and time acquired this time based on the specified date and time, and calculates the calculation result of the average value of the air temperature this time. It is the temperature at the time of acquisition.
FIG. 5 is a diagram showing an example (No. 1) of the relationship between the air temperature and the amount of electricity used. In the example shown in FIG. 5, the monthly usage amount is shown as an example of the electricity usage amount. As shown in FIG. 5, the relationship between the air temperature and the amount of electricity used can be approximated by a downwardly convex function.
The calculation unit 154 calculates the first electric information, the second electric information, and the third electric information based on the relationship between the air temperature and the amount of electricity used. In FIG. 5, the first electrical information is indicated by "A", the second electrical information is indicated by "C", and the third electrical information is indicated by "B". When calculating the third electrical information, the calculation unit 154 approximates the relationship between the air temperature and the electrical information with a curve. The calculation unit 154 calculates the third electrical information by obtaining a minimum value based on the result approximated by the curve. The calculation unit 154 outputs the first electric information, the second electric information, and the third electric information to the learning unit 156.
In this way, the calculation unit 154 obtains the relationship between the time-series data of electricity usage and the temperature for one household, and from the relationship between the time-series data of the obtained electricity usage and the temperature, the first electric information And the second electric information and the third electric information are calculated. By performing the same processing as the above-mentioned processing, the calculation unit 154 obtains the relationship between the time-series data of the electricity usage and the temperature information for each of the plurality of households, and obtains the time-series data of the electricity usage obtained. Based on the relationship with the temperature information, the first electric information, the second electric information, and the third electric information are calculated.

The learning unit 156 is realized by a command from the CPU 102. The learning unit 156 acquires the first electric information, the second electric information, and the third electric information output by the calculation unit 154. Further, the learning unit 156 is associated with the time-series data of the amount of electricity used obtained from the attribute information 1108 stored in the storage unit 110, the first electric information, the second electric information, and the third electric information. Get attribute information. The learning unit 156 performs machine learning using the acquired attribute information as teacher data based on the first electric information, the second electric information, and the third electric information.
FIG. 6 is a diagram showing an example (No. 1) of machine learning executed by the learning device according to the embodiment. FIG. 6 shows an example of learning whether or not the attribute information of the building “has gas floor heating”.
At node N1-01, the composition ratio of buildings that have gas floor heating is about bb%. The learning unit 156 generates a condition of "whether or not the new year is 2000 or later". The learning unit 156 forms a node N1-03 and a node N1-04 by differentiating the node N1-01 into two edges under the condition of "whether or not the new year is 2000 or later". If the new year is before 2000, it is differentiated into nodes N1-03, and if the new year is after 2000, it is differentiated into nodes N1-0. Node N1-03 has a building composition ratio of about dd% that has gas floor heating. Node N1-04 has a building composition ratio of about ff% that has gas floor heating. Here, it is assumed that the node N1-04 has a higher composition ratio of the building having the gas floor heating, and the description will be continued in the direction of differentiating the nodes N1-04.

The learning unit 156 generates a condition of "whether or not the difference between the first electric information and the second electric information is xx or more". The learning unit 156 differentiates the node N1-04 into two edges under the condition of "whether or not the difference between the first electrical information and the second electrical information is xx or more", thereby making the node N1-11 and the node N1-11. It forms nodes N1-12. If the difference between the first electrical information and the second electrical information is less than xx, it is differentiated into nodes N1-11, and if the difference between the first electrical information and the second electrical information is xx or more, the node N1 It is differentiated into -12. At node N1-11, the composition ratio of the building having gas floor heating is about hh%. At nodes N1-12, the composition ratio of buildings that have gas floor heating is about jj%. Here, it is assumed that the node N1-11 has a higher composition ratio of the building having the gas floor heating, and the description will be continued in the direction of differentiating the node N1-11.
The learning unit 156 generates a condition of "whether or not the value obtained by dividing the first electric information by the second electric information is yy or more". The learning unit 156 divides the node N1-11 into two edges under the condition of "whether or not the value obtained by dividing the first electric information by the second electric information is yy or more", thereby causing the node N1-13. And nodes N1-14. If the value obtained by dividing the first electric information by the second electric information is less than yy, it is differentiated into nodes N1-13, and if the value obtained by dividing the first electric information by the second electric information is greater than or equal to yy, it is differentiated into nodes N1-13. It is differentiated into nodes N1-14. At node N1-13, the composition ratio of the building having gas floor heating is about ll%. At nodes N1-14, the composition ratio of the building having gas floor heating is about nn%. It is assumed that the node N1-13 has a higher composition ratio of the building having gas floor heating and the composition ratio is as high as about ll%, and the learning ends here. Hereinafter, the node used when learning is completed and it is determined whether or not it has attribute information is referred to as a "final node".
In the example shown in FIG. 6, the objective variable is "gas floor heating possession", and the explanatory variables are "whether or not the new year is 2000 or later" and "difference between the first electric information and the second electric information". Is xx or more ”and“ whether or not the value yy or more obtained by dividing the first electric information by the second electric information ”. Here, the explanatory variables include any or a combination of the first electric information, the second electric information, and the third electric information. In this way, by using a combination of one or at least two differences, ratios, etc. of the first electric information, the second electric information, and the third electric information as explanatory variables, the temperature that has a great influence on the amount of electricity used The impact can be corrected, and a stable estimation model can be created regardless of regional differences and recording years.

The learning unit 156 associates the objective variable with the explanatory variable obtained by machine learning and outputs the output to the output unit 158.
The output unit 158 is realized by a command from the CPU 102. The output unit 158 stores the objective variable and the explanatory variable output by the learning unit 156 in the attribute information estimation table 1110 of the storage unit 110. Specifically, when the determination tree learning shown in FIG. 6 is performed by the learning unit 156, the output unit 158 sets the objective variable “gas floor heating possession” output by the learning unit 156 to the attribute of the attribute information estimation table 1110. Stored in the information column, the explanatory variables "whether the new year is 2000 or later", "whether the difference between the first electrical information and the second electrical information is xx or more", and "first Whether or not the electric information is equal to or greater than the value yy obtained by dividing the electric information by the second electric information ”is stored in the column of the variable used for explaining the attribute information in the attribute information estimation table 1110.
Here, an example in which any or a combination of the first electric information, the second electric information, and the third electric information is generated as explanatory variables will be described.
FIG. 7 is a diagram showing an example (No. 2) of the relationship between the air temperature and the amount of electricity used. In the example shown in FIG. 7, the relationship between the air temperature of a building having good heat insulation and the amount of electricity used and the relationship between the temperature of a building having poor heat insulation and the amount of electricity used are shown.
In a building with good heat insulation, even if the temperature is low, the increase in monthly electricity usage is small, so it is assumed that the slope of the line segment connecting A and B is small or the value obtained by dividing A by C is small. Will be done. On the other hand, in a building with poor heat insulation, the monthly amount of electricity used increases as the temperature decreases, so the slope of the line segment connecting A and B is large, or the value obtained by dividing A by C is large. Is assumed. However, when the heat source for heating is other than electricity, in a building with poor heat insulation, the slope of the line segment connecting A and B is small, the value obtained by dividing C by B is large, or C is larger than A. Expected to be small.
In this way, the attribute information of the household and the attribute information of the building can be explained by any or a combination of the first electric information, the second electric information, and the third electric information.

FIG. 8 is a diagram showing an example (No. 2) of machine learning executed by the learning device according to the embodiment. FIG. 8 shows an example of learning the household attribute information “number of households”.
In node N2-01, the composition ratio of households with one household is about oo%, the composition ratio of households with two households is about pp%, and the number of households is three to four. The composition ratio of a household is about qq%, and the composition ratio of a household with five or more households is about rr%.
The learning unit 156 generates a condition of "whether or not the base power consumption is less than zzkWh". The learning unit 156 forms node N2-02 and node N2-03 by differentiating node N2-01 into two edges under the condition of "whether or not the base power consumption is less than zzkWh". .. When the base power consumption is less than zzkWh, it is differentiated into nodes N2-02, and when the base power consumption is zzkWh or more or is a missing value, it is differentiated into nodes N2-03. In node N2-02, the composition ratio of a household with one household is about ss%, the composition ratio of a household with two households is about tt%, and the number of households is three to four. The composition ratio of a certain household is about uu%, and the composition ratio of a household with five or more households is about vv%. In node N2-03, the composition ratio of a household with one household is about ww%, the composition ratio of a household with two households is about xxx%, and the number of households is three to four. The composition ratio of a certain household is about yy%, and the composition ratio of a household having five or more households is about zz%.
Here, the description will be continued in the direction of differentiating the nodes N2-03 under the preset conditions.

The learning unit 156 generates a condition of "whether or not it is before 1992". The learning unit 156 forms a node N2-06 and a node N2-07 by differentiating into two edges under the condition of "whether or not it is before 1992". If it is before 1992, it is differentiated into node N2-06, and if it is after 1992, it is differentiated into node N2-07. In node N2-06, the composition ratio of households with one household is about ab%, the composition ratio of households with two households is about ac%, and the number of households is three to four. The composition ratio of a certain household is about ad%, and the composition ratio of a household with five or more households is about ae%. In node N2-07, the composition ratio of a household with one household is about af%, the composition ratio of a household with two households is about ag%, and the number of households is three to four. The composition ratio of a certain household is about ah%, and the composition ratio of a household with five or more households is about ai%. The learning unit 156 ends learning with node N2-07 as the final node.
In the example shown in FIG. 8, the objective variable is "number of households" and the explanatory variables are "whether the base power consumption is less than zzkWh" and "whether it is before 1992". .. In this case, the explanatory variables do not include any or a combination of the first electrical information, the second electrical information, and the third electrical information.

FIG. 9 is a diagram showing an example (No. 3) of machine learning executed by the learning device according to the embodiment. FIG. 9 shows an example of learning the home attribute information “presence or absence of infants”.
In node N3-01, the composition ratio of households with infants is about ax%, and the composition ratio of households without infants is about aw%.
The learning unit 156 generates a condition of "whether or not it was newly established before 2011". The learning unit 156 differentiates node N3-01 into two edges under the condition of "whether or not it was newly established before 2011" to form node N3-02 and node N3-03. To form. If it was newly established before 2011 or if it is a missing value, it will be differentiated into node N3-02, and if it was newly established after 2011, it will be differentiated into node N3-03. .. In node N3-02, the composition ratio of households with infants is about al%, and the composition ratio of households without infants is about aj%. In node N3-03, the composition ratio of households with infants is about an%, and the composition ratio of households without infants is about am%. The learning unit 156 ends learning with node N3-02 as the final node.
In the example shown in FIG. 9, the objective variable is "presence or absence of infants" and the explanatory variable is "whether or not it was newly established before 2011". In this case, the explanatory variables do not include any or a combination of the first electrical information, the second electrical information, and the third electrical information.
Here, the case of differentiating the node N3-03 will be further described.

The learning unit 156 generates a condition of "whether or not the value obtained by dividing A by B is smaller than xy". The learning unit 156 forms a node N3-06 and a node N3-07 by differentiating into two edges under the condition of "whether or not the value obtained by dividing A by B is smaller than xy". When the value obtained by dividing A by B is smaller than xy, it is differentiated into nodes N3-06, and when the value obtained by dividing A by B is greater than or equal to xy, it is differentiated into nodes N3-07. In node N3-06, the composition ratio of households with infants is ap%, and the composition ratio of households without infants is ao%. At node N3-07, the composition ratio of households with infants is about ar%, and the composition ratio of households without infants is about aq%.
Here, the description will be further continued in the direction of differentiating the node N3-07. The learning unit 156 forms nodes N3-10 and nodes N3-11 by differentiating into two edges under the condition of "whether or not the value obtained by subtracting A by B is smaller than xz". When the value obtained by subtracting A by B is smaller than xz, it is differentiated into nodes N3-10, and when the value obtained by subtracting A by B is xz or more, it is differentiated into nodes N3-11. In node N3-10, the composition ratio of households with infants is at%, and the composition ratio of households without infants is as%. At node N3-11, the composition ratio of households with infants is about av%, and the composition ratio of households without infants is about au%.
By comparing node N3-07 and node N3-10, the amount of electricity used in winter is xy times or more of the base power consumption, but the amount of AB used is due to the small number of households and household appliances. Profiling can be done if the difference is not large.

FIG. 10 is a diagram showing an example (No. 4) of machine learning executed by the learning device according to the embodiment. FIG. 10 shows an example of learning the home attribute information “presence or absence of elderly people”.
In node N4-01, the composition ratio of households with elderly people is about bc%, and the composition ratio of households without elderly people is about bab%.
The learning unit 156 generates a condition of "whether or not it was newly established before 1979". The learning unit 156 differentiates node N4-01 into two edges under the condition of "whether or not it was newly established before 1979" to form node N4-02 and node N4-03. To form. If it was newly established before 1979 or if it is a missing value, it is differentiated into node N4-02, and if it was newly established after 1979, it is differentiated into node N4-03. .. In node N4-02, the composition ratio of households with elderly people is about be%, and the composition ratio of households without elderly people is about bd%. In node N4-03, the composition ratio of households with elderly people is about bg%, and the composition ratio of households without elderly people is about bf%.
The learning unit 156 generates a condition of "whether or not B is less than yxkWh". The learning unit 156 forms a node N4-04 and a node N4-05 by differentiating the node N4-03 into two edges under the condition of "whether or not B is less than yxkWh". When B is less than yxkWh or is a missing value, it is differentiated into node N4-04, and when B is more than yxkWh, it is differentiated into node N4-05. In node N4-04, the composition ratio of households with elderly people is about bi%, and the composition ratio of households without elderly people is about bh%. In node N4-05, the composition ratio of households with elderly people is about bk%, and the composition ratio of households without elderly people is about bj%. The learning unit 156 ends learning with node N4-04 as the final node.
In the example shown in FIG. 10, the objective variable is "presence or absence of the elderly", the explanatory variables are "whether or not it was newly established before 1979", and "whether or not B is smaller than yxkWh". Is it? " In this case, the explanatory variables do not include any or a combination of the first electrical information, the second electrical information, and the third electrical information.

Here, the case of differentiating the node N4-05 will be further described.
The learning unit 156 generates a condition of "whether or not the value obtained by subtracting A by B is smaller than yz". The learning unit 156 forms a node N4-08 and a node N4-09 by differentiating into two edges under the condition of "whether or not the value obtained by subtracting A by B is smaller than yz". When the value obtained by subtracting A by B is smaller than yz, it is differentiated into node N4-08, and when the value obtained by subtracting A by B is greater than or equal to yz, it is differentiated into node N4-09. At node N4-08, the composition ratio of households with elderly people is bm%, and the composition ratio of households without elderly people is bl%. At node N4-09, the composition ratio of households with elderly people is bo%, and the composition ratio of households without elderly people is bn%.
According to the node N4-09, in addition to the large B, the difference in the amount of A-B used is also large, so that it is possible to profile if the home rate is high and the heating is heavily used.

(Operation of learning device)
FIG. 11 is a flowchart showing an example of the operation of the learning device according to the embodiment.
(Step S101) The acquisition unit 152 of the learning device 100 acquires time-series data of the electricity usage of each of the plurality of households. The acquisition unit 152 stores the acquired time-series data of the electricity usage of each of the plurality of households in the electricity usage time-series data 1104 of the storage unit 110.
(Step S102) The acquisition unit 152 of the learning device 100 acquires the air temperature information during the period in which the time-series data of the amount of electricity used is acquired. The acquisition unit 152 stores the acquired air temperature information in the air temperature information 1106 of the storage unit 110.
(Step S103) The acquisition unit 152 of the learning device 100 acquires the attribute information of the home and the attribute information of the building from which the time-series data of the amount of electricity used has been obtained. The acquisition unit 152 stores the acquired home attribute information and building attribute information in the attribute information 1108 of the storage unit 110.
(Step S104) The calculation unit 154 of the learning device 100 acquires the time-series data of the electricity usage for each of the plurality of households from the electricity usage time-series data 1104 of the storage unit 110. In addition, the calculation unit 154 calculates the temperature information during the period in which the time-series data of the amount of electricity used is acquired for each of the plurality of households. Then, the calculation unit 154 obtains the relationship between the time-series data of the amount of electricity used and the temperature information for each of the plurality of households. The calculation unit 154 calculates the first electric information, the second electric information, and the third electric information based on the relationship between the time series data of the amount of electricity used and the temperature information. The calculation unit 154 outputs the first electric information, the second electric information, and the third electric information to the learning unit 156.
(Step S105) The learning unit 156 of the learning device 100 acquires the first electric information, the second electric information, and the third electric information output by the calculation unit 154. Further, the learning unit 156 has attribute information associated with the time-series data of the amount of electricity used obtained from the attribute information 1108 stored in the storage unit 110, the first electric information, the second electric information, and the third electric information. To get. The learning unit 156 performs machine learning using the acquired attribute information as teacher data based on the first electric information, the second electric information, and the third electric information.

(Step S106) The learning unit 156 of the learning device 100 associates the attribute information with a variable (explanatory variable) used for explaining the attribute information obtained by learning. The learning unit 156 associates the attribute information with the explanatory variables obtained by machine learning and outputs the information to the output unit 158.
(Step S107) The output unit 158 of the learning device 100 stores the objective variable and the explanatory variable output by the learning unit 156 in the attribute information estimation table 1110 of the storage unit 110.
Here, the effect of the learning device 100 according to the embodiment will be described.
FIG. 12 is a diagram showing an example of the relationship between the ratio of all customers and the cumulative ratio with gas floor heating. According to FIG. 12, when the ratio of all customers is 20%, the cumulative ratio with gas floor heating is 20% in the case of random sampling, whereas the cumulative ratio with gas floor heating is 20% in the case of model extraction. The cumulative ratio with heating is 70%. That is, in the case of model extraction, the capture probability increases 3.5 times as compared with random sampling.

In the above-described embodiment, the electricity usage data server 200 stores information indicating the past electricity usage, and the learning device 100 acquires time-series data of the household electricity usage from the electricity usage data server 200. The case has been described, but it is not limited to this example. For example, the learning device 100 acquires information indicating the amount of electricity used from a system capable of outputting the amount of electricity used to the outside, such as a smart meter or HEMS, at a predetermined cycle, and information indicating the acquired amount of electricity used. May be stored in association with the acquisition time and the location information of the building in which the system capable of outputting electricity to the outside is installed. With this configuration, the learning device 100 obtains the attribute information and the variables used for explaining the attribute information from the electricity usage data server 200 without acquiring the time-series data of the household electricity usage. Can be associated and memorized.
In the above-described embodiment, the case where the learning device 100 includes the storage unit 110 in which the attribute information estimation table 1110 is stored has been described, but the present invention is not limited to this example. For example, the attribute information estimation table 1110 may be stored in the cloud. In this case, the learning device 100 stores the attribute information and the variables used for explaining the attribute information in association with each other in the attribute information estimation table 1110 on the cloud.

According to the learning device 100 according to the present embodiment, the time-series data of the electricity usage of each of the plurality of households, the temperature information during the period when the time-series data was acquired, and the household from which the time-series data was obtained. Get the attribute information of and the attribute information of the building. The learning device 100 uses time-series data and temperature information for each of the plurality of households based on the time-series data of the electricity usage of each of the plurality of households and the temperature information during the period in which the time-series data was acquired. Seeking a relationship with. The learning device 100 calculates the first electric information, the second electric information, and the third electric information from the relationship between the time-series data and the temperature information, and the attribute information of the house or the building from which the time-series data is obtained is calculated. Using attribute information as teacher data, machine learning is performed based on the first electrical information, the second electrical information, and the third electrical information. The learning device 100 stores the attribute information in association with the variable used for explaining the attribute information obtained by learning.
With this configuration, when the time-series data of the unknown electricity usage and the temperature information of the period when the time-series data was acquired are given, the time of the unknown electricity usage It is possible to estimate the attribute information of the house and the attribute information of the building from which the series data was obtained. Specifically, when determining whether or not a home or building has certain attribute information, the relationship between the time-series data of the electricity usage of the household whose attribute information is unknown and the temperature information is obtained, and the obtained attribute information is obtained. Calculates the first electric information, the second electric information, and the third electric information from the relationship between the time-series data of the unknown electricity usage and the temperature information. Then, it is differentiated according to the variables used for explaining the attribute information stored in association with a certain attribute information, and it is determined whether or not it corresponds to the final node. If it corresponds to the final node, it is determined that the home or building has the attribute information, and if it does not correspond, it is determined that the home or building does not have the attribute information.
By being able to estimate the home attribute information and the building attribute information, it is possible to take an approach according to the estimation result of the home attribute information and the building estimation information. Specifically, when attribute information with poor heat insulation is obtained, it is possible to recommend remodeling to a building with good heat insulation.

(Second embodiment)
(Attribute estimation device)
The attribute estimation device according to the embodiment will be described. The attribute estimation device 300 according to the embodiment stores the attribute information in association with the variable used for explaining the attribute information. The attribute estimation device 300 acquires time-series data of the amount of electricity used in a household whose attribute information is unknown, and temperature information during the period in which the time-series data was acquired.
The attribute estimation device 300 estimates the attribute information of the home and the attribute information of the building from which the acquired time-series data of the amount of electricity used is obtained. Specifically, the attribute estimation device 300 obtains the relationship between the time-series data of the electricity usage of a household whose attribute information is unknown and the temperature information, and the obtained attribute information is the time-series data of the unknown electricity usage and the temperature. From the relationship with the information, the first electric information, the second electric information, and the third electric information are calculated. The attribute estimation device 300 selects the attribute information for which it is desired to determine whether or not the home or building has it. The attribute estimation device 300 differentiates according to the variables used for explaining the attribute information stored in association with the selected attribute information, and determines whether or not it corresponds to the final node. The attribute estimation device 300 determines that the home or building has the certain attribute information when it corresponds to the final node, and determines that the home or building does not have the certain attribute information when it does not correspond. To do. The attribute estimation device 300 outputs the attribute information determined to be possessed by the home or building. As described above, the attribute estimation device 300 is connected to the communication network 50 in the learning system 1 shown in FIG.

(Configuration of attribute estimation device)
FIG. 13 is a diagram showing an example of the hardware configuration of the attribute estimation device according to the embodiment.
The attribute estimation device 300 includes a CPU 302, a ROM 304, a RAM 306, a communication I / F 308, a storage unit 310, an operation unit 312, a display unit 314, and a bus 350.
The CPU 302 reads the program 3102 or the like that controls the operation of the attribute estimation device 300 from the storage unit 310, expands it into the RAM 306, and executes it.
The communication I / F 308 is composed of communication modules such as an NFC module and a wireless LAN module. Further, the communication I / F 308 may be configured by USB. The communication I / F 308 communicates with an external device by NFC or wireless LAN.
The storage unit 310 is composed of a non-volatile memory such as a flash memory, an HDD, an SSD, and an SD card. The storage unit 310 stores the program 3102 executed by the CPU 302 and the attribute information estimation table 3110. The attribute information estimation table 1110 described above can be applied to the attribute information estimation table 3110.
The operation unit 312 is an input device that accepts user operations, and includes a pointing device such as a touch panel, buttons, dials, touch sensors, touch pads, and the like.
The display unit 314 is configured by, for example, a liquid crystal display or the like, and displays the attribute information of a home or a building from which time-series data of the electricity consumption of the home whose attribute information is unknown is obtained.

(Functional configuration of attribute estimation device)
FIG. 14 is a functional block diagram of the attribute estimation device according to the embodiment.
The attribute estimation device 300 functions as an acquisition unit 352, a calculation unit 354, an estimation unit 356, and an output unit 358 by executing the program 3102 developed on the RAM 306 from the storage unit 310 by the CPU 302.

(Each functional configuration of the attribute estimation device)
Each functional configuration of the attribute estimation device 300 will be described in detail with reference to FIGS. 13 and 14. In the following description of each function of the attribute estimation device 300, the relationship with the main hardware for realizing each functional block of the attribute estimation device 300 shown in FIG. 14 will also be described.
The acquisition unit 352 is realized by a command from the CPU 302 and a communication I / F 308. The acquisition unit 352 acquires time-series data of household electricity usage for which attribute information is unknown and temperature information during the period in which the time-series data was acquired, and the acquired time-series data and the time-series data are obtained. The temperature information of the acquired period is output to the calculation unit 354.
The calculation unit 354 is realized by an instruction from the CPU 302. The calculation unit 354 acquires the time-series data of the amount of electricity used for which the attribute information output by the acquisition unit 352 is unknown and the temperature information during the period when the time-series data is acquired, and the acquired attribute information is unknown. Based on the time-series data and the temperature information during the period in which the time-series data was acquired, the relationship between the time-series data whose attribute information is unknown and the temperature information is obtained. The calculation unit 354 calculates the first electric information, the second electric information, and the third electric information from the relationship between the time-series data of the electric consumption whose attribute information is unknown and the temperature information. The calculation unit 354 outputs the first electric information, the second electric information, and the third electric information obtained by the calculation to the estimation unit 356.

The estimation unit 356 is realized by an instruction from the CPU 302. The estimation unit 356 acquires the attribute information estimation table 1110 in which the attribute information and the variables used for explaining the attribute information are associated with each other from the learning device 100 via the communication I / F 308. The estimation unit 356 stores the acquired attribute information estimation table 1110 in the attribute information estimation table 3110 of the storage unit 310.
The estimation unit 356 acquires the first electric information, the second electric information, and the third electric information output by the calculation unit 354. The estimation unit 356 selects from the attribute information estimation table 3110 of the storage unit 310 the attribute information for determining whether or not the house or building from which the time-series data of the amount of electricity consumption whose attribute information is unknown has is acquired. The estimation unit 356 applies and applies the first electrical information, the second electrical information, and the third electrical information to the variables used for explaining the attribute information stored in association with the selected attribute information. Analyze by differentiating with the variables obtained.
As a result of differentiation, the estimation unit 356 determines that the node has attribute information when it corresponds to the final node, and determines that it does not have attribute information when it does not correspond. The estimation unit 356 outputs the estimation result including the attribute information estimated to have by the analysis to the output unit 358.
The output unit 358 is realized by a command from the CPU 302. The output unit 358 acquires the estimation result output by the estimation unit 356, and outputs the acquired estimation result to the display unit 314.

(Operation of attribute estimation device)
FIG. 15 is a flowchart showing an example of the operation of the attribute estimation device according to the embodiment. In FIG. 15, the attribute estimation device 300 acquires the attribute information estimation table 1110 in which the attribute information and the variables used for explaining the attribute information are associated with each other from the learning device 100, and the acquired attribute information estimation table 1110 is obtained from the attribute information. The operation after being stored in the estimation table 3110 is shown.
(Step S201) The acquisition unit 352 of the attribute estimation device 300 acquires time-series data of the amount of electricity used in a household whose attribute information is unknown. The acquisition unit 352 outputs the acquired time-series data of the amount of electricity used to the calculation unit 354.
(Step S202) The acquisition unit 352 of the attribute estimation device 300 acquires the air temperature information during the period in which the time-series data of the electricity usage is acquired. The acquisition unit 352 outputs the temperature information during the period in which the acquired time-series data of the electricity usage is acquired to the calculation unit 354.
(Step S203) The calculation unit 354 of the attribute estimation device 300 has the time-series data of the electricity usage for which the attribute information output by the acquisition unit 352 is unknown, and the temperature during the period in which the time-series data of the electricity usage is acquired. The attribute information is unknown based on the time-series data of the electricity usage for which the information is acquired and the acquired attribute information is unknown, and the temperature information during the period when the time-series data of the electricity usage is acquired. Find the relationship between time-series data of electricity usage and temperature information. The calculation unit 354 calculates the first electric information, the second electric information, and the third electric information from the relationship between the time-series data of the amount of electricity used and the temperature information for which the obtained attribute information is unknown. The calculation unit 354 outputs the first electric information, the second electric information, and the third electric information obtained by the calculation to the estimation unit 356.
(Step S204) The estimation unit 356 of the attribute estimation device 300 acquires the first electrical information, the second electrical information, and the third electrical information output by the calculation unit 354. The estimation unit 356 selects from the attribute information estimation table 3110 of the storage unit 310 the attribute information for determining whether or not the house or building from which the time-series data of the amount of electricity consumption whose attribute information is unknown has is acquired. The estimation unit 356 applies and applies the first electrical information, the second electrical information, and the third electrical information to the variables used for explaining the attribute information stored in association with the selected attribute information. The variables obtained by this are analyzed by differentiation. The estimation unit 356 outputs the estimation result including the attribute information estimated to have by the analysis to the output unit 358.
(Step S205) The output unit 358 of the attribute estimation device 300 acquires the estimation result output by the estimation unit 356, and outputs the acquired estimation result to the display unit 314.

According to the attribute estimation device 300 according to the embodiment, the attribute information estimation table 1110 acquired from the learning device 100 is stored in the attribute information estimation table 3110 of the storage unit 310. The attribute estimation device 300 acquires the time-series data of the electricity usage for which the attribute information is unknown and the temperature information for the period during which the time-series data was acquired, and the time-series of the electricity usage for which the attribute information is unknown. Find the relationship between data and temperature information. The calculation unit 354 calculates the first electric information, the second electric information, and the third electric information from the relationship between the time-series data of the electric consumption whose attribute information is unknown and the temperature information. The attribute estimation device 300 selects the attribute information for determining whether or not the home or building from which the time-series data of the amount of electricity used whose attribute information is unknown has it. The estimation unit 356 applies and applies the first electrical information, the second electrical information, and the third electrical information to the variables used for explaining the attribute information stored in association with the selected attribute information. By differentiating with the variables obtained by the above, the attribute information of the house or building from which the time-series data of electricity usage was obtained is estimated.
By being able to estimate the attribute information of the house and the attribute information of the building, it is possible to take an approach according to the estimation result of the attribute information of the house and the estimated information of the building. Specifically, when attribute information with poor heat insulation is obtained, it is possible to recommend remodeling to a building with good heat insulation.

(Third Embodiment)
The attribute estimation system according to the embodiment will be described. The attribute estimation system according to the embodiment includes a learning device 100 according to the first embodiment and an attribute estimation device 300 according to the second embodiment.
As described above, the learning device 100 according to the embodiment acquires each of the time-series data of the electricity usage of each of the plurality of households and the plurality of electricity usages included in the acquired time-series data of the electricity usage. The temperature information of the period and the attribute information of the house or building from which the acquired time series data was obtained are acquired. The learning device 100 calculates the first electric information, the second electric information, and the third electric information from the relationship between the acquired time series data and the acquired temperature information. The learning device 100 obtains variables used for explaining the attribute information by machine learning using the attribute information as teacher data based on the first electric information, the second electric information, and the third electric information. The learning device 100 associates and stores the attribute information and the variables used for explaining the attribute information.

Further, as described above, the attribute estimation device 300 according to the embodiment acquires the attribute information estimation table 1110 in which the attribute information and the variables used for explaining the attribute information are associated with each other from the learning device 100, and the acquired attribute information. The estimation table 1110 is stored in the attribute information estimation table 3110 of the storage unit 310.
The attribute estimation device 300 acquires time-series data of the amount of electricity used in a household whose attribute information is unknown, and temperature information during the period in which the time-series data was acquired. The attribute estimation device 300 obtains the relationship between the time-series data of the electricity usage of a household whose attribute information is unknown and the temperature information, and from the relationship between the time-series data of the electricity usage whose attribute information is unknown and the temperature information. , Calculates the first electrical information, the second electrical information, and the third electrical information. The attribute estimation device 300 selects the attribute information for which it is desired to determine whether or not it has the attribute information, and calculates the variable used for explaining the attribute information stored in association with the selected attribute information. The information, the second electric information, and the third electric information are applied, and according to the variables obtained by applying the information, it is differentiated and it is determined whether or not it corresponds to the final node. The attribute estimation device 300 determines that the home or building has the certain attribute information when it corresponds to the final node, and determines that the home or building does not have the certain attribute information when it does not correspond. To do. The attribute estimation device 300 outputs the attribute information determined to be possessed by the home or building.

(Operation of attribute estimation system)
FIG. 16 is a diagram showing the operation of the attribute estimation system according to the embodiment.
(Step S301) The acquisition unit 152 of the learning device 100 acquires time-series data of the electricity usage of each of the plurality of households from the electricity usage data server 200. In addition, the acquisition unit 152 acquires the temperature information during the period in which the time series data is acquired by accessing a site that provides meteorological information such as the Japan Meteorological Agency. In addition, the acquisition unit 152 acquires the attribute information of the household from which the information indicating the amount of electricity used has been obtained.
(Step S302) The calculation unit 154 of the learning device 100 obtains the relationship between the time series data and the temperature information for each of the plurality of homes. The calculation unit 154 calculates the first electric information, the second electric information, and the third electric information based on the relationship between the time series data of the amount of electricity used and the temperature information. The calculation unit 154 outputs the first electric information, the second electric information, and the third electric information to the learning unit 156.
(Step S303) The learning unit 156 of the learning device 100 bases the first electric information, the second electric information, and the third electric information output by the calculation unit 154 on the first electric information and the second electric information. Machine learning is performed using the attribute information associated with the time-series data obtained from the information and the third electrical information as teacher data. As a result of machine learning, machine-learned data such as the attribute information estimation table 1110 is obtained.

(Step S304) The acquisition unit 352 of the attribute estimation device 300 acquires the time-series data of the electricity consumption of the household whose attribute information is unknown and the temperature information during the period in which the time-series data is acquired.
(Step S305) The calculation unit 354 of the attribute estimation device 300 has the time-series data of the electricity usage for which the attribute information output by the acquisition unit 352 is unknown, and the temperature during the period when the time-series data of the electricity usage is acquired. The attribute information is unknown based on the time-series data of the electricity usage for which the information is acquired and the acquired attribute information is unknown, and the temperature information during the period when the time-series data of the electricity usage is acquired. Find the relationship between time-series data of electricity usage and temperature information. The calculation unit 354 calculates the first electric information, the second electric information, and the third electric information from the relationship between the time series data of the amount of electricity used and the temperature information.
(Step S306) The estimation unit 356 of the attribute estimation device 300 acquires the first electrical information, the second electrical information, and the third electrical information output by the calculation unit 354, and the attribute information estimation table of the storage unit 310. From 3110, the attribute information for determining whether or not the home or building from which the time-series data for which the attribute information is unknown has is selected is selected. The estimation unit 356 applies and applies the first electrical information, the second electrical information, and the third electrical information to the variables used for explaining the attribute information stored in association with the selected attribute information. The variables obtained by this are analyzed by differentiation. The estimation unit 356 outputs the estimation result including the attribute information estimated to have by the analysis to the output unit 358. When the output unit 358 acquires the attribute information output by the estimation unit 356, the output unit 358 displays the attribute information on the display unit 114. In the example shown in FIG. 11, as attribute information, gas floor heating possession, presence / absence of infants, presence / absence of elderly people, and years of use of air conditioner are displayed.

In the above-described embodiment, the case where the attribute estimation device 300 acquires the attribute information estimation table 1110 from the learning device 100 has been described, but the present invention is not limited to this example. For example, the attribute estimation device 300 transmits to the learning device 100 time-series data of the amount of electricity used in a household whose attribute information is unknown, temperature information during the period in which the time-series data is acquired, and attribute information to be determined. Then, the first electric information, the second electric information, the third electric information, and the variables used to explain the attribute information associated with the attribute information to be determined may be inquired. Then, in response to an inquiry from the attribute estimation device 300, the learning device 100 receives the time-series data of the electricity usage of the household whose attribute information is unknown and the temperature information during the period in which the time-series data of the electricity usage is acquired. The variable used to explain the attribute information associated with the attribute information to be determined by calculating the first electrical information, the second electrical information, and the third electrical information from the obtained relationship is calculated. , The attribute estimation device 300 acquires the first electric information, the second electric information, the third electric information, and the variables used for explaining the attribute information associated with the attribute information to be determined, which are acquired from the attribute information estimation table 1110. You may send it to.
The attribute estimation device 300 acquires the first electric information, the second electric information, the third electric information, and the variables used for explaining the attribute information associated with the attribute information to be determined, which are transmitted by the learning device 100. The attribute estimation device 300 applies the acquired first electric information, the second electric information, and the third electric information to the variables used for explaining the acquired attribute information, and according to the variables obtained by applying the acquired attribute information. It differentiates and determines whether it corresponds to the final node. The attribute estimation device 300 determines that the home or building has the certain attribute information when it corresponds to the final node, and determines that the home or building does not have the certain attribute information when it does not correspond. To do. The attribute estimation device 300 outputs the attribute information determined to be possessed by the home or building.
In the above-described embodiment, the case where the learning device 100 and the attribute estimation device 300 are different devices has been described, but the present invention is not limited to this example. For example, the learning device 100 and the attribute estimation device 300 may be realized by one device.

According to the attribute estimation system according to the embodiment, the time-series data of the electricity usage of each of the plurality of households, the temperature information during the period when the time-series data of the electricity usage is acquired, and the time of the electricity usage. Acquire the attribute information of the household from which the series data was obtained. The attribute estimation system uses electricity for each of the plurality of households based on the time-series data of the electricity usage of each of the plurality of households and the temperature information during the period when the time-series data of the electricity usage is acquired. Find the relationship between time-series data of quantity and temperature information. The attribute estimation system calculates the first electric information, the second electric information, and the third electric information from the relationship between the time-series data of the amount of electricity used and the temperature information, and the attributes of the household from which the time-series data is obtained. Using the information as teacher data, machine learning is performed based on the first electrical information, the second electrical information, and the third electrical information. The attribute estimation system stores the attribute information in association with the variables used for explaining the attribute information obtained by learning.

Further, according to the attribute estimation system, time-series data of the amount of electricity used whose attribute information is unknown and temperature information during the period when the time-series data was acquired are acquired, and the acquired attribute information is unknown. The relationship between the time-series data of the quantity and the temperature information during the period in which the time-series data was acquired is obtained. The attribute estimation system calculates the first electric information, the second electric information, and the third electric information from the relationship between the time series data of the amount of electricity used whose attribute information is unknown and the temperature information. The attribute estimation device 300 selects the attribute information for determining whether or not the household has the time-series data of the amount of electricity used for which the attribute information is unknown. The estimation unit 356 applies and applies the first electrical information, the second electrical information, and the third electrical information to the variables used for explaining the attribute information stored in association with the selected attribute information. By differentiating with the variables obtained by the above, the attribute information of the household from which the time-series data of electricity usage was obtained is estimated. Since the quality and performance of a pre-owned house can be estimated from the attribute information of the home living in the pre-owned house, the quality and performance of the pre-owned house can be determined by acquiring the attribute information of the home. I can grasp it. It can also be applied not only to the quality and performance of used houses but also to the evaluation of the quality and performance of buildings such as houses.

Although the embodiments and modifications thereof have been described above, these embodiments and modifications thereof are presented as examples and are not intended to limit the scope of the invention. These embodiments and variations thereof can be implemented in various other embodiments, and various omissions, replacements, changes, and combinations can be made without departing from the gist of the invention. These embodiments and variations thereof are included in the scope and gist of the invention, and at the same time, are included in the scope of the invention described in the claims and the equivalent scope thereof.

The learning device and the attribute estimation device described above may be realized by a computer. In that case, a program for realizing the function of each functional block is recorded on a computer-readable recording medium. The program recorded on the recording medium may be read by the computer system and executed by the CPU. The term "computer system" as used herein includes hardware such as an OS (Operating System) and peripheral devices.
Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, or a CD-ROM. Further, the "computer-readable recording medium" includes a storage device such as a hard disk built in a computer system.

Further, the "computer-readable recording medium" may include a medium that dynamically holds the program for a short period of time. What dynamically holds the program for a short period of time is, for example, a communication line when the program is transmitted via a network such as the Internet or a communication line such as a telephone line.
Further, the "computer-readable recording medium" may include a medium that holds a program for a certain period of time, such as a volatile memory inside a computer system serving as a server or a client. Further, the above program may be for realizing a part of the above-mentioned functions. Further, the above-mentioned program may be realized by combining the above-mentioned functions with a program already recorded in the computer system. Further, the above program may be realized by using a programmable logic device. The programmable logic device is, for example, an FPGA (Field Programmable Gate Array).

The learning device and the attribute estimation device described above have a computer inside. The process of each process of the learning device and the attribute estimation device described above is stored in a computer-readable recording medium in the form of a program, and the computer reads and executes this program to perform the process. It is said.
Here, the computer-readable recording medium refers to a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Further, this computer program may be distributed to a computer via a communication line, and the computer receiving the distribution may execute the program.
Further, the above program may be for realizing a part of the above-mentioned functions.
Further, it may be a so-called difference file (difference program) that can realize the above-mentioned function in combination with a program already recorded in the computer system.

1 ... Learning system, 50 ... Communication network, 100 ... Learning device, 102 ... CPU, 104 ... ROM, 106 ... RAM, 108 ... Communication I / F, 110 ... Storage unit, 112 ... Operation unit, 114 ... Display unit, 150 ... Bus, 152 ... Acquisition unit, 154 ... Calculation unit, 156 ... Learning unit, 158 ... Output unit, 200 ... Electric usage data server, 300 ... Attribute estimation device, 302 ... CPU, 304 ... ROM, 306 ... RAM, 308 ... communication I / F, 310 ... storage unit, 312 ... operation unit, 314 ... display unit, 350 ... bus, 352 ... acquisition unit, 354 ... calculation unit, 356 ... estimation unit, 358 ... output unit, 1102 ... program, 1104 ... Electric consumption time series data, 1106 ... Temperature information, 1108 ... Attribute information, 1110 ... Attribute information estimation table, 3102 ... Program, 3104 ... Attribute information estimation table

Claims (11)

Acquisition of time-series data of household electricity usage and temperature information during the period when each of the plurality of electricity usages included in the time-series data was acquired and attribute information of the household from which the time-series data was acquired. Department and
Based on the relationship between the time-series data acquired by the acquisition unit and the temperature information, the first electricity information indicating the amount of electricity used at the representative temperature in winter and the amount of electricity used at the representative temperature in summer are obtained. A calculation unit that calculates the second electric information to be shown and the third electric information to show the minimum value of the amount of electricity used.
Based on the first electrical information, the second electrical information, and the third electrical information, a learning unit that performs machine learning using the attribute information of the household as teacher data.
A learning device equipped with. The learning device according to claim 1, wherein the acquisition unit acquires attribute information of a building in which a consumer constituting the home resides. The learning device according to claim 1 or 2, wherein the learning unit performs machine learning using any or a combination of the first electrical information, the second electrical information, and the third electrical information as explanatory variables. .. The learning device according to any one of claims 1 to 3, wherein the learning device performs decision tree learning. According to any one of claims 1 to 4, each of the plurality of electricity consumptions included in the time-series data is acquired every period of one week or more and one month or less. The learning device described. The learning device according to claim 4, wherein the air temperature information is an average value of the air temperature during a period in which each of the plurality of electricity consumptions included in the time series data is acquired. An acquisition unit that acquires time-series data of household electricity usage for which attribute information is unknown and temperature information during the period in which each of the plurality of electricity usages included in the time-series data was acquired.
Based on the relationship between the time-series data acquired by the acquisition unit and the temperature information, the first electricity information indicating the amount of electricity used at the representative temperature in winter and the amount of electricity used at the representative temperature in summer are obtained. An arithmetic unit that calculates the second electrical information to be shown and the third electrical information to show the minimum value,
When the representative temperature of winter is obtained from the relationship between the time-series data including multiple information indicating the household electricity usage and the temperature information during the period when each of the plurality of electricity usages included in the time-series data is acquired. Based on the amount of electricity used, the amount of electricity used at the representative temperature in summer, and the minimum value of the amount of electricity used, the attribute information of the household from which the time-series data was obtained was used as teacher data for machine learning. Using the fruit, the attribute information of the household from which the time series data acquired by the acquisition unit has been obtained is estimated based on the first electrical information, the second electrical information, and the third electrical information. Attribute estimation device including an estimation unit. A step of acquiring the time-series data of the electricity usage of the household and the temperature information of the period in which each of the plurality of electricity usages included in the time-series data was acquired and the attribute information of the household from which the time-series data was acquired. When,
Based on the relationship between the time-series data acquired in the acquisition step and the temperature information, the first electricity information indicating the electricity consumption at the representative temperature in winter and the electricity consumption at the representative temperature in summer And the step of calculating the second electric information indicating the minimum value of the electric consumption and the third electric information indicating the minimum value of the electric consumption,
Based on said first electrical information and the second electric information and the third electrical information, the attribute information of the home as teacher data, the steps of machine learning
The a learning process by the learning apparatus is performed. A step of acquiring time-series data of household electricity usage for which attribute information is unknown and temperature information for a period in which each of a plurality of electricity usages included in the time-series data was acquired, and
Based on the relationship between the time-series data acquired in the acquisition step and the temperature information, the first electricity information indicating the electricity consumption at the representative temperature in winter and the electricity consumption at the representative temperature in summer Steps to calculate the second electrical information indicating the minimum value and the third electrical information indicating the minimum value,
When the representative temperature of winter is obtained from the relationship between the time-series data including multiple information indicating the household electricity usage and the temperature information during the period when each of the plurality of electricity usages included in the time-series data is acquired. and electricity consumption, and electricity consumption when the representative summer temperatures, based on the minimum value of the electricity consumption, the result obtained by the machine learning the attribute information of the home was obtained time series data as teacher data use, estimates the first on the basis of the electric information and the second electric information and the third electrical information, the attribute information of the home the acquired time-series data is obtained in the step of obtaining An attribute estimation method performed by an attribute estimation device that has steps to perform. For computers equipped with learning devices
A step of acquiring the time-series data of the electricity usage of the household and the temperature information of the period in which each of the plurality of electricity usages included in the time-series data was acquired and the attribute information of the household from which the time-series data was acquired. When,
Based on the relationship between the time-series data acquired in the acquisition step and the temperature information, the first electricity information indicating the electricity consumption at the representative temperature in winter and the electricity consumption at the representative temperature in summer And the step of calculating the second electric information indicating the minimum value of the electric consumption and the third electric information indicating the minimum value of the electric consumption,
Based on said first electrical information and the second electric information and the third electrical information, the attribute information of the home as teacher data, the steps of machine learning
A program that runs. For computers equipped with attribute estimators
A step of acquiring time-series data of household electricity usage for which attribute information is unknown and temperature information for a period in which each of a plurality of electricity usages included in the time-series data was acquired, and
Based on the relationship between the time-series data acquired in the acquisition step and the temperature information, the first electricity information indicating the electricity consumption at the representative temperature in winter and the electricity consumption at the representative temperature in summer Steps to calculate the second electrical information indicating the minimum value and the third electrical information indicating the minimum value,
When the representative temperature of winter is obtained from the relationship between the time-series data including multiple information indicating the household electricity usage and the temperature information during the period when each of the plurality of electricity usages included in the time-series data is acquired. and electricity consumption, and electricity consumption when the representative summer temperatures, based on the minimum value of the electricity consumption, the result obtained by the machine learning the attribute information of the home was obtained time series data as teacher data use, estimates the first on the basis of the electric information and the second electric information and the third electrical information, the attribute information of the home the acquired time-series data is obtained in the step of obtaining A program that executes the steps to be performed.

Images