JP7698993B2 - Household composition change detection device and household composition change detection method - Google Patents

Description

The present invention relates to a household composition change detection device and a household composition change detection method.

In recent years, effective use of the electricity consumption of each household in a region has been considered. For example, the household number estimation device in Patent Document 1 estimates the number of people in each household in a region based on the amount of electricity consumed by each time period measured by a power meter during cooling operation in summer or heating operation in winter. This estimated number of people in each household is used, for example, to estimate the evacuation situation within the region in the event of a disaster in the region.

The user attribute estimation device of Patent Document 2 estimates attributes, which are the living behavior characteristics of household members, based on the power consumption of each household, which is the user. The user attribute estimation device estimates the attributes of household members during a characteristic time period, based on the power consumption during that time period when the power consumption is characteristic according to the weather, and information related to the weather in the area to which the household belongs. The estimated attributes are used, for example, for support related to daily life.

JP 2019-36262 A Patent No. 6151202

For example, a company's marketing or sales activities may require the ability to capture changes in household composition. It would be desirable to be able to more easily detect changes in household composition.

A household composition change detection device that can solve the above problem has a control device that acquires the amount of electricity used by each household from a power meter with communication capabilities via a communication network. The control device detects changes in the household composition based on trends in electricity use that are common to changes in the household composition to be detected.

Based on the correlation between changes in household composition and changes in electricity usage, it is possible to detect changes in household composition based on characteristic changes in electricity usage. With the above configuration, it is only necessary to focus on changes in electricity usage, making it easier to detect households whose composition has changed.

In the above household composition change detection device, the change in household composition may be a birth. In this case, the control device may detect a birthing household, which is a household whose composition has changed due to a birth, based on a change in daytime electricity usage over at least six consecutive months.

Households giving birth tend to have an increased electricity usage during daytime hours for around six months around the time of the birth. For this reason, it is possible to detect households giving birth based on the continuous increase in electricity usage during daytime hours.

In the above-mentioned household composition change detection device, the control device may calculate the maximum value throughout the year of the sum of squares of the upward deviation from the median of daytime electricity usage for the same month of the previous year over at least six consecutive months as a birth score indicating the likelihood that the household is a birthing household, and detect birthing households based on the calculated birth score.

With this configuration, seasonality can be removed by focusing on the increase in electricity usage for the current year compared to the electricity usage for the same month of the previous year, i.e., the effects of seasonal fluctuations in electricity usage can be suppressed. In addition, because the timing of maternity leave, child care leave, or returning home to give birth varies from person to person, the time when changes in electricity usage appear also varies from person to person. In this regard, by using the maximum value of the sum of squares throughout the year as the birth score, it is possible to focus on the period when the change in electricity usage for each household is most noticeable. Therefore, groups extracted based on the birth score have a significantly improved accuracy in detecting birthing households compared to groups extracted randomly.

In the above household composition change detection device, the control device may be configured to reduce the birth score value of households that have attributes that are considered to be less likely to give birth based on the attribute information of each household.

According to this configuration, households having attributes that are considered to be less likely to give birth are less likely to be detected, making it possible to more effectively detect households giving birth.
In the above household composition change detection device, the change in the household composition may be an increase or decrease in the number of people in the household. In this case, the control device may detect a household in which the number of people in the household has increased or decreased based on a change in the amount of electricity used in the morning on a holiday.

In households where the number of people in the household has increased or decreased, electricity usage in the morning on holidays tends to change, regardless of the age of the person who leaves the household. For this reason, it is possible to detect households where the number of people in the household has increased or decreased based on the trend of changes in electricity usage in the morning on holidays.

In the above household composition change detection device, the control device may detect households whose number of people has increased or decreased based on a comparison of electricity usage in holiday mornings with the same month of the previous year.
With this configuration, by focusing on the year-on-year comparison of electricity usage to the same month, it is possible to eliminate seasonality, i.e., to suppress the effects of seasonal fluctuations in electricity usage, making it possible to effectively detect households whose number of people has increased or decreased.

In the above-mentioned household composition change detection device, the change in the household composition may be a decrease in the number of people in the household. In this case, the control device may detect a household in which the number of people in the household has decreased due to the departure of a member of a specific age group, based on a change in the amount of electricity used during a time period when a characteristic change in electricity use occurs depending on the age of the household member who leaves the household.

There are time periods during which electricity usage changes characteristically depending on the age of the household member who leaves the household. Therefore, based on the trend in electricity usage changes during time periods specific to the age of the household member who leaves the household, it is possible to detect households where the number of household members has decreased due to the departure of a member of a specific age.

In the above household composition change detection device, the control device may detect households where the number of household members has decreased due to the departure of children in their 20s and 30s based on changes in electricity usage on weekday nights.

Households where the number of household members has decreased due to the departure of children in their 20s and 30s tend to have a decrease in electricity usage on weekday evenings. Therefore, based on the trend of changes in electricity usage on weekday evenings, it is possible to detect households where the number of household members has decreased due to the departure of children in their 20s and 30s.

In the above household composition change detection device, the control device may detect households where the number of household members has decreased due to the departure of elderly people aged 60 or older, based on changes in electricity usage during weekday mornings.

Households where the number of household members has decreased due to the departure of elderly people aged 60 or older tend to use less electricity on weekday evenings. Therefore, based on the trend of changes in electricity usage in weekday mornings, it is possible to detect households where the number of household members has decreased due to the departure of elderly people aged 60 or older.

In the above household composition change detection device, the control device may detect households where the number of household members has decreased due to the departure of a household member of a specific age group, based on a year-on-year comparison of electricity usage during a time period when there is a characteristic change in electricity usage depending on the age of the household member who leaves the household.

With this configuration, by focusing on year-on-year comparisons of electricity usage to the same month, it is possible to remove seasonality, i.e., to reduce the effects of seasonal fluctuations in electricity usage. This makes it possible to effectively detect households where the number of household members has decreased due to the departure of household members of a certain age group.

In the above household composition change detection device, the control device may detect a household that has started telecommuting based on a trend of changes in electricity usage during the daytime on weekdays.
Households that have started working from home tend to have an increased amount of electricity usage during the day on weekdays. Therefore, it is possible to detect households that have started working from home based on the trend of changes in electricity usage during the day on weekdays.

In the above household composition change detection device, the control device may detect households that have started telecommuting based on a year-on-year comparison of daytime electricity usage on weekdays.
With this configuration, by focusing on the year-on-year comparison of electricity usage to the same month, it is possible to remove seasonality, i.e., to suppress the effects of seasonal fluctuations in electricity usage, making it possible to effectively detect households that have started working from home.

In the above household composition change detection device, the control device may detect households with members who have enrolled in elementary school based on the trend of changes in electricity usage from before noon to late afternoon in August.

In households with members who have started elementary school, electricity usage tends to increase in the period from before noon to just after noon in August. Therefore, it is possible to detect households with members who have started elementary school based on the trend in electricity usage in the period from before noon to just after noon in August.

In the above household composition change detection device, the control device may detect households with members who have entered elementary school based on the year-on-year comparison of electricity usage during the period from before noon to after noon in August to the same month of the previous year.

With this configuration, by focusing on year-on-year comparisons of electricity usage to the same month, it is possible to remove seasonality, i.e., reduce the effects of seasonal fluctuations in electricity usage. This makes it possible to effectively detect households with members who have enrolled in elementary school.

A method for detecting changes in household composition that can solve the above problem involves obtaining the amount of electricity used by each household via a communication network from an electricity meter with communication capabilities, and detecting changes in the household composition based on trends in electricity use that are common to changes in the household composition of interest.

With this method, if a correlation can be found between changes in household composition and changes in electricity usage, it is possible to detect changes in household composition based on characteristic changes in electricity usage. Since it is only necessary to focus on changes in electricity usage, it is easier to detect households whose composition has changed.

The household composition change detection device and household composition change detection method of the present invention make it easier to detect changes in household composition.

1 is a block diagram showing a first embodiment of a household configuration change detection device and a state in which a power meter is connected to a communication network. 1 is a block diagram showing the configuration of a smart meter that cooperates with a first embodiment of a household composition change detection device. FIG. 1 is a block diagram showing the configuration of a first embodiment of a household composition change detection device. FIG. 4 is an explanatory diagram showing a change in power consumption accompanying a change in household configuration according to the first embodiment. 11 is a graph showing changes in the median power consumption for the current year and the median power consumption for the same month of the previous year according to the first embodiment. 1 is a table showing the birth estimation accuracy of the first embodiment. 13 is a graph showing the relationship between the household selection rate and the childbirth household coverage rate (reproduction rate) according to the first embodiment. 13 is a table showing the birth estimation accuracy of the second embodiment. 13 is a graph showing the relationship between the household selection rate and the childbirth household coverage rate (reproduction rate) according to the second embodiment. 13 is a table showing the accuracy of estimation of independence, divorce, and bereavement in the third embodiment. 13 is a graph showing the relationship between the household selection rate and the independent/divorced/widowed household coverage rate (reproduction rate) according to the third embodiment. 13 is a table showing the accuracy of child independence estimation in the fourth embodiment. 13 is a graph showing the relationship between the household selection rate and the child independent household coverage rate (reproduction rate) according to the fourth embodiment. 13 is a table showing the elderly bereavement estimation accuracy according to the fourth embodiment. 13 is a graph showing the relationship between the household selection rate and the elderly bereaved household coverage rate (reproduction rate) according to the fourth embodiment. 13 is a table showing the estimation accuracy of the start of telecommuting according to the fifth embodiment. 13 is a graph showing the relationship between the household selection rate and the telecommuting start household coverage rate (reproduction rate) according to the fifth embodiment. 23 is a graph showing the relationship between the selection rate of households with young children in the previous year and the coverage rate (reproduction rate) of households enrolled in elementary school according to the sixth embodiment.

First Embodiment
A first embodiment of a household composition change detection device and a household composition change detection method will be described below. The household composition detection device detects a change in the household composition based on the amount of electricity used by the household.

As shown in FIG. 1, the detection device 10 is connected to a communication network 20 such as the Internet. The detection device 10 is provided at a general electricity transmission and distribution business operator such as an electric power company. The detection device 10 can communicate with a smart meter 30 via the communication network 20. The smart meter 30 is an electricity meter that has both an electricity measurement function and a communication function, and is provided at each household in the area. The smart meter 30 measures the amount of electricity used for a specified period (e.g., 30 minutes) and transmits electricity data indicating the measured amount of electricity used to the detection device 10 via the communication network 20. The electricity data also includes location information of the smart meter 30. The detection device 10 detects changes in household composition based on the electricity data of each household acquired via the communication network 20.

As shown in FIG. 2, the smart meter 30 has a measuring device 30A, a memory device 30B, a communication device 30C, and a control device 30D. The measuring device 30A, the memory device 30B, the communication device 30C, and the control device 30D are interconnected via a bus 30E, which is a signal line.

The measuring device 30A measures the amount of electricity used by a consumer household at set intervals (e.g., 30 minutes). The measuring device 30A detects the voltage of the power grid and the current supplied to the load from the power grid. The measuring device 30A measures the amount of electricity used by the household based on the detected voltage and current.

The memory device 30B stores the amount of power usage measured by the measuring device 30A according to commands from the control device 30D. The memory device 30B also stores identification information unique to the smart meter 30 and location information of the smart meter 30. This location information is information indicating the location where the smart meter 30 is installed, for example, information indicating the location of the house or the like in which the smart meter 30 is installed.

The communication device 30C is an interface between the control device 30D and the communication network 20. The communication device 30C transmits and receives information via the communication network 20. The communication device 30C transmits and receives information to and from external devices connected to the communication network 20 in accordance with commands from the control device 30D. The external devices include the detection device 10.

The control device 30D performs overall control of the smart meter 30. The control device 30D generates power data. The power data includes the amount of power usage measured by the measuring device 30A and the location information of the smart meter 30. The control device 30D transmits the power data to the detection device 10 via the communication device 30C.

As shown in FIG. 3, the detection device 10 has a communication device 10A, a storage device 10B, and a control device 10C. The communication device 10A, the storage device 10B, and the control device 10C are interconnected via a bus 10D, which is a signal line.

The communication device 10A is an interface between the control device 10C and the communication network 20. The communication device 10A transmits and receives information via the communication network 20. The communication device 10A transmits and receives information to and from external devices connected to the communication network 20 in accordance with commands from the control device 10C. The external devices include the smart meter 30.

The storage device 10B has, for example, a main storage unit and an auxiliary storage unit. The main storage unit stores the computer programs executed by the control device 10C and the data processed by the control device 10C. The main storage unit has a RAM (Random Access Memory) and a (Read Only Memory). The auxiliary storage unit is a non-volatile storage unit that stores various programs including an OS (Operating System) and various data in a readable and writable manner. The auxiliary storage unit is a flash memory, a hard disk drive, a solid state drive, an SSD, or the like. The OS includes a communication interface program that transfers data between the OS and an external device connected via the communication device 10A. The auxiliary storage unit stores information received via the communication device 10A. This information includes power data sent from the smart meter 30 via the communication network 20.

The control device 10C executes various processes according to the programs stored in the memory device 10B. The control device 10C has a processor such as a CPU (Central Processing Unit) or MPU (Micro Processing Unit). The processor performs overall control of the entire detection device 10 through the execution of computer programs deployed in the working area of the memory device 10B. The control device 10C receives power data sent from the smart meter 30 via the communication network 20 via the communication device 10A, and stores the received power data in the memory device 10B in a state in which it is associated with identification information unique to the smart meter 30.

The control device 10C detects changes in household composition based on the trend of changes in electricity usage over a specific period (e.g., months or time period). If a correlation can be found between a change in household composition of interest and changes in electricity usage, it is possible to detect the change in household composition of interest based on the characteristic change in electricity usage discovered. Changes in household composition of interest include life events such as childbirth, children becoming independent, divorce or death. Here, an example is described for detecting a childbirth household. A childbirth household is a household whose composition has changed due to the birth of a child.

As shown in Figure 4, households giving birth tend to have an increase in electricity usage during daytime hours for around six months around the time of birth. Based on this continuing trend of increasing electricity usage during daytime hours, it is possible to extract households that are more likely to be childbirth households. Incidentally, Figure 4 shows the change in electricity usage before and after the birth of a household whose number of members increased from three to four. Each horizontal square in Figure 4 represents one day, and each vertical square represents an increment of 30 minutes. The darker the square, the more electricity used.

The changes in electricity usage shown in Figure 4 can be used to gauge the living conditions of households expecting a baby. The hours of the day in the month before and the month before the month in which the household composition changed (in this case, the month in which the baby was expected) can be divided into, for example, four time periods TZ0, TZ1, TZ2, and TZ3.

Time zone TZ0: time period for getting ready in the morning Time zone TZ1: time period for the wife and child alone Time zone TZ2: time period on days when the husband returns home earlier than usual Time zone TZ3: time period after the husband returns home and the whole family is gathered together Next, the time of day in the month in which the household composition changes and the beginning of the following month are divided into, for example, three time zones TZ4, TZ5, and TZ6.

Time period TZ4: Morning preparation time period Time period TZ5: Daytime time period Time period TZ6: Time period after husband returns home In the month of household composition change and the beginning of the following month, the amount of power used in the earlier time period of the morning time period TZ4, for example, from 7:00 am to 8:00 am, increases compared to the month before and the month before the month before the month of the household composition change. From this, it is assumed that there has been a change in morning preparation. In addition, in the month of household composition change and the beginning of the following month, the amount of power used in the daytime time period TZ5 decreases sharply compared to the month before and the month before the month of the household composition change. From this, it is assumed that there is no one at home during the day. In addition, from the change in power use at night, it is assumed that the husband returns home later than usual. From the power use in these time periods TZ4, TZ5, and TZ6, it is assumed that the wife and children have returned to their parents' home.

From the middle of the month following the month in which the household composition changed, the amount of electricity used on a day increases overall, compared to the month in which the household composition changed and the beginning of the following month. In particular, there is a notable increase in electricity use during the daytime hours of TZ7 on a day. From this, it can be assumed that, for example, a mother and child have returned home, and that the air conditioner is running 24 hours a day to keep the baby in good health.

<Detection algorithm>
The control device 30D detects a birthing household according to a detection algorithm. This detection algorithm is based on a comparison between the median power usage during daytime hours over six consecutive months and the median power usage during the same month of the previous year. The control device 30D calculates a birth score that indicates the likelihood that the household is a birthing household, and detects the birthing household based on the calculated birth score. The control device 30D calculates the birth score as the maximum value throughout the year of the sum of squares of the upward deviation of the power usage during daytime hours over six consecutive months from the median power usage during the same month of the previous year.

By the way, an upswing means that the median value of electricity usage during daytime hours over a consecutive six-month period exceeds the median value of electricity usage during the same month of the previous year. Also, an upswing means the difference between the upswing median value of electricity usage during daytime hours over a consecutive six-month period and the median value of electricity usage during the same month of the previous year.

The detection algorithm is based on the following points (A1) to (A4).
(A1) Electricity usage varies greatly depending on the season, so there is a risk that changes in electricity usage before and after childbirth may be lost due to seasonal differences. In this regard, seasonality can be removed by focusing on the year-on-year comparison of electricity usage to the same month.

(A2) Except for times when electricity usage differs greatly depending on whether the mother is at home, the change in electricity usage before and after childbirth is small and may be lost in other factors. In this regard, during the daytime, electricity usage is more likely to differ greatly depending on whether the mother is at home. For this reason, by focusing only on the daytime, it is possible to make the difference in electricity usage before and after childbirth more noticeable.

(A3) Changes in electricity usage due to childbirth continue for about six months. By focusing only on one specific month, it is difficult to distinguish households that happen to use a lot of electricity. In this regard, by focusing on electricity usage over a consecutive six-month period, it is possible to appropriately capture changes in electricity usage due to childbirth.

(A4) The timing of maternity leave, childcare leave, or returning home to give birth varies from person to person. As a result, the time when changes in electricity usage appear also varies from person to person. For this reason, it is preferable to shift the six months of the year of interest by one month at a time, and focus on the period when the change in electricity usage is most noticeable.

An example of the calculation procedure for the birth score is as follows.
As shown in the graph in Figure 5, for example, in the year from November 2019 to October 2020, there are seven ways to choose six months: (B1) to (B7).

(B1) November 2019 to April 2020 (B2) December 2019 to May 2020 (B3) January 2020 to June 2020 (B4) February 2020 to July 2020 (B5) March 2020 to August 2020 (B6) April 2020 to September 2020 (B7) May 2020 to October 2020 The control device 30D shifts the starting month of the six months of interest in one year by one month, and calculates the sum of squares of the upward deviation of the current usage from the median of the same month of the previous year in each of the periods (B1) to (B7). The power usage is the power usage during the daytime, for example, from 10:00 to 13:00.

The control device 30D calculates the sum of squares Σ using the following equation (1).
Σ=S1 ² +S2 ² +S3 ² +S4 ² +S5 ² +S6 ² ...(1)
However, "S1 to S6" are values that are above the median value of the electricity usage during daytime hours for six consecutive months in the same month of the previous year.

It should be noted that when calculating the sum of squares Σ, control device 30D does not add the downward fluctuation of the amount of power usage during daytime hours for six consecutive months from the median value for the same month of the previous year.
As shown in the graph of FIG. 5, for example, in the six months from March 2020 to August 2020, the price fell below the median only in April 2020. In this case, as expressed by the following equation (2), the value of the fall from the median in April 2020 is not used in the calculation of the sum of squares Σ.

Σ=S1 ² +S3 ² +S4 ² +S5 ² +S6 ² ...(2)
However, "S1" is the upward deviation from the median in March 2020, and "S3 to S6" are the upward deviation from the median from May 2020 to August 2020.

Next, the control device 30D calculates the largest sum of squares Σ _max among the sums of squares Σ of each period (B1) to (B7) as the birth score SCR _birth as shown in the following formula (3). This is to focus on the period in which the difference from the median of the electricity usage amount for the same month of the previous year is most noticeable.

SCR _birth =Σ _max …(3)
The control device 10C determines that a household is a birthing household in descending order of the value of the birth score SCR _birth . The control device 10C sorts the households in descending order of the value of the birth score SCR _birth , and extracts a predetermined number of households from the top of the sorted order. The higher the value of the birth score SCR _birth , the higher the probability that the household is a birthing household. By extracting birthing households based on the birth score SCR _birth , it is possible to ensure the accuracy of detecting birthing households.

<Accuracy of birth estimates>
The accuracy of fertility projections can be expressed, for example, by the relationship between the household selection rate and the recall rate, or the relationship between the household selection rate and the precision rate. The household selection rate is the proportion of households selected from the population. The recall rate is the proportion of households that are determined to have given birth among all child-bearing households. The precision rate is the proportion of child-bearing households among all selected households.

The accuracy of the birth estimation is as follows. Here, for example, we consider the case of capturing 18 households out of 197 households that gave birth between November 2019 and October 2020.

As shown in the top row of the table in FIG. 6, for example, if the top 20 people with the highest birth score SCR _birth are selected from 197 people, the household selection rate is 10%. If the household selection rate is 10%, the recall rate is 33%. In other words, it is possible to guess approximately 6 households out of 18 households that gave birth. Also, if the household selection rate is 10%, the precision rate is 30%. In other words, of the 20 people selected, 6 households are actually the birth households.

As shown in the graph of FIG. 7, for example, when 20 people are randomly selected from 197 people, the household selection rate is 10% and the recall rate is 10%. From this, it can be seen that the recall rate when 20 people are selected based on the birth score SCR _birth is 3.3 times that when 20 people are randomly selected. In other words, the certainty (degree of certainty) that 20 people selected based on the birth score SCR _birth are the birth household is 3.3 times that when 20 people are randomly selected. In this way, by using the birth score SCR _birth based on changes in power usage, it is possible to improve the detection accuracy of birth households regardless of the proportion of households selected.

Effects of the First Embodiment
Therefore, according to the first embodiment, the following effects can be obtained.
(1-1) Based on the trend of changes in power usage common to households giving birth, households giving birth can be effectively detected based on changes in power usage. Specifically, households giving birth are detected based on changes in daytime power usage over at least six consecutive months. In households giving birth, power usage during daytime hours tends to increase for about six months from the time before and after the birth. For this reason, households giving birth can be detected based on the continuous trend of increasing power usage during daytime hours. In addition, since it is only necessary to focus on changes in power usage, households giving birth can be detected more easily and efficiently. Incidentally, the specific period is at least six consecutive months, but may be longer.

(1-2) The maximum value of the sum of squares of the upward deviation of daytime electricity usage from the median value of the same month of the previous year for at least six consecutive months is calculated as SCR _birth , and the birth household is detected based on the calculated birth score SCR _birth . The birth score SCR _birth is a value indicating the likelihood of being a birth household, and the higher the value of the birth score SCR _birth, the higher the probability that the household is a birth household. In this way, seasonality can be removed by focusing on the increase in electricity usage in the current year relative to the electricity usage in the same month of the previous year, that is, the influence of seasonal fluctuations in electricity usage can be suppressed. In addition, since the timing of maternity leave, childcare leave, or homecoming to give birth varies from person to person, the timing at which changes in electricity usage appear also varies from person to person. In this regard, by using the maximum value of the sum of squares throughout the year as the _birth score SCR _{birth, it is possible to focus on the period in which the change in electricity usage of each household is most noticeable. Therefore, the accuracy of detecting birth households is greatly improved in groups extracted based on the birth score SCR birth} compared to groups extracted randomly.

(1-3) It is possible to detect households where a child is about to be born as a change in household composition. The detection results can be used in corporate marketing or sales activities. For example, a change in household composition due to a birth, etc., may result in a mismatch between the property in which the person lives and the household composition. For this reason, a change in household composition due to a birth, etc., can be taken as a sign of property renovation or rehousing.

Second Embodiment
Next, a second embodiment of the household composition change detection device will be described. This embodiment basically has the same configuration as the first embodiment shown in Figures 1 to 3. Therefore, the same members and configurations as those in the first embodiment are given the same reference numerals, and detailed descriptions thereof will be omitted.

The storage device 10B stores attribute information for each household. The attribute information is associated with identification information unique to the smart meter 30 in each household and location information for the smart meter 30. The attribute information is obtained, for example, through a questionnaire survey of each household. The attributes are characteristics or properties that each household possesses, and include information such as the number of people in the household, gender, and age group.

The questionnaire survey may be conducted via the communication network 20. Furthermore, the method of acquiring the attribute information is not limited to the questionnaire survey. The attribute information may be acquired using various methods, regardless of whether or not the communication network 20 is used.

The control device 10C corrects the value of the birth score SCR _birth according to the attributes of each household. Based on the attribute information of each household, the control device 10C reduces the value of the birth score SCR _birth of households other than those having attributes that are considered to have a low probability of giving birth, i.e., households having attributes that are considered to have a high probability of giving birth. The following two attributes (C1) and (C2) can be given as attributes of households that are highly likely to give birth.

(C1) A household with two people and a woman in her 20s or 30s.
(C2) Households with three or more people and women in their 20s to 40s.
Incidentally, the attributes (C1) and (C2) are selected based on the following three viewpoints (D1) to (D3).

(D1) There is almost no possibility of giving birth to a child outside of a married couple or a household with children.
(D2) People often give birth to their first child in their 30s.
(D3) It is extremely rare for a woman to give birth after the age of 50.

The control device 10C uses the following equation (4) to correct the birth score SCR _birth of households other than those having attributes that indicate a high probability of giving birth.
SCR _{birth_c} =SCR _birth・WT=SCR _birth・(1/10000) …(4)
Here, "SCR _{birth_c} " is the birth score after correction. "SCR _birth " is the birth score before correction calculated based on the above formula (3). "WT" is a weighting value, which is set to a fixed value of, for example, "1/10000".

By using equation (4), the birth score SCR _birth of households other than those having attributes that indicate a high probability of giving birth is corrected to a value of "1/10,000".
The accuracy of the birth estimation is as follows. Here again, we consider the case of capturing 18 households out of 197 households who gave birth between November 2019 and October 2020.

As shown in the top row of the table in FIG. 8, for example, if the top 20 people with the highest birth score SCR _birth are selected from 197 people, the household selection rate is 10%. When the household selection rate is 10%, the recall rate is 44%. In other words, 8 out of 18 households that gave birth can be guessed. Also, when the household selection rate is 10%, the precision rate is 40%. In other words, of the 20 people selected, 8 are the actual birth households.

As shown in the graph of FIG. 9, for example, when 20 people are randomly selected from 197 people, the household selection rate is 10% and the recall rate is 10%. From this, it can be seen that the recall rate when 20 people are selected based on the birth score SCR _birth and household attributes is 4.4 times that when 20 people are randomly selected. In other words, the probability that 20 people selected based on the birth score SCR _birth and household attributes are the birth household is 4.4 times that when 20 people are randomly selected. In this way, by using the birth score SCR _birth and household attributes based on the change in power usage, it is possible to improve the detection accuracy of the birth household regardless of the proportion of households selected.

By the way, although the details are omitted, by appropriately narrowing down based on the attributes of each household, it is possible to narrow down the number of households that are likely to be childbirth households to 101 households. According to this narrowing down, if 20 people are selected from 197 people, the recall rate is 18%. In other words, 3.2 households out of 18 households that gave birth can be guessed. From this, it can be seen that the recall rate when 20 people are selected based on the birth score SCR _birth and the attributes of the household is 2.5 times (= 8 households / 3.2 households) higher than when 20 people are selected based only on the attributes of each household. In other words, it is possible to extract childbirth households more effectively by combining the electricity data and the attributes of each household than when childbirth households are extracted using only the attributes of each household.

<Advantages of the Second Embodiment>
Therefore, according to the second embodiment, the following effects can be obtained.
(2-1) Based on the attribute information of each household, the birth score SCR _birth of households other than those with attributes that are considered to have a high probability of giving birth is reduced. This makes it difficult to detect households with attributes that are considered to have a low probability of giving birth. This makes it possible to detect birthing households more effectively.

Third Embodiment
Next, a third embodiment of the household composition change detection device will be described. This embodiment basically has the same configuration as the first embodiment shown in Figures 1 to 3. Therefore, the same members and configurations as those in the first embodiment are given the same reference numerals, and detailed descriptions thereof will be omitted.

The detection device 10 detects households where the number of household members has decreased. Possible circumstances for a decrease in the number of household members include, for example, children becoming independent, and household members leaving or dying. As a result of searching for common characteristics when the number of household members changes (here, a decrease) in various households with different lifestyle patterns, it was found that households where the number of household members has decreased tend to have a decrease in electricity usage in the morning on holidays, regardless of the age of the person who leaves the household. This is thought to be due, for example, to the fact that many people spend holiday mornings at home. Therefore, it is possible to detect households where the number of household members is likely to have decreased, based on the trend of changes in electricity usage in holiday mornings.

<Detection algorithm>
The control device 10C detects households where the number of people in the household has decreased according to a detection algorithm. The control device 10C determines that the households have experienced one of the events of independence, divorce, and death, in order of decreasing year-on-year comparison of electricity usage during holiday mornings. The control device 10C sorts the households in order of decreasing year-on-year comparison of electricity usage during holiday mornings, and extracts a predetermined number of households from the top of the sorted order.

<Accuracy of Estimation of Decreasing Number of Households>
The estimation accuracy for households with a decrease in the number of household members (estimation accuracy for those who are independent, divorced, or widowed) is as follows. For example, let us consider the case where 18 households out of 197 households have had a decrease in the number of household members. Households with a decrease in the number of household members include independent households, divorced households, and widowed households.

As shown in the top row of the table in Figure 10, for example, if the top 22 people with the lowest year-on-year electricity usage in holiday mornings are selected from 197 people, and the numbers after the one digit are discarded, the household selection rate is 10%. When the household selection rate is 10%, the recall rate is 28%. In other words, it is possible to correctly identify 5 of the 18 households where the number of people in the household has decreased. Also, when the household selection rate is 10%, the precision rate is 23%. In other words, of the 22 people selected, 5 households have actually experienced a decrease in the number of people in the household.

As shown in the graph in Figure 11, for example, if 22 people are randomly selected from 197 people, the household selection rate is 10% and the recall rate is 10%. This shows that the recall rate when 22 people are selected based on electricity data, i.e., changes in electricity usage, is 2.8 times higher than when 22 people are selected randomly. Furthermore, the probability that the 22 people selected based on changes in electricity usage are households with a decrease in the number of people is 2.8 times higher (5 households / 1.8 households) than when 22 people are selected randomly. In this way, by using electricity data, households with a decrease in the number of people can be effectively and efficiently detected regardless of the proportion of households selected.

<Advantages of the Third Embodiment>
Therefore, according to the third embodiment, the following effects can be obtained.
(3-1) Based on the trend of changes in electricity usage common to households with a decrease in the number of household members, households with a decrease in the number of household members can be effectively detected based on changes in electricity usage. Households with a decrease in the number of household members tend to have a decrease in electricity usage in the mornings of holidays, regardless of the age of the person who leaves the household. For this reason, it is possible to detect households with a decrease in the number of household members based on the trend of changes in electricity usage in the mornings of holidays. In addition, since it is necessary to focus only on changes in electricity usage, households where a child is being born can be detected more easily and efficiently.

(3-2) Detect households with a decrease in the number of people in the household based on the year-on-year change in electricity usage in holiday mornings. The smaller the year-on-year change in electricity usage in holiday mornings for a household, the higher the probability that the number of people in the household has decreased. By focusing on the year-on-year change in electricity usage, seasonality can be removed, i.e., the effects of seasonal fluctuations in electricity usage can be reduced. This makes it possible to effectively detect households with a decrease in the number of people in the household.

(3-3) It is possible to detect households where the number of people in the household has decreased as a change in household composition. This detection result can be used in corporate marketing or sales activities. For example, a change in the number of people in a household due to independence, divorce, or death of a loved one may result in a mismatch between the property in which the person lives and the household composition. For this reason, a decrease in the number of people in a household can be taken as a sign of property renovation or rehousing.

The third embodiment may be modified as follows.
That is, the detection device 10 detects households with an increase in the number of household members based on the same perspective as when detecting households with a decrease in the number of household members. Households with an increase in the number of household members tend to use more electricity in the mornings of holidays. Therefore, it is possible to detect households with an increase in the number of household members based on the trend of changes in electricity usage in the mornings of holidays. Since it is only necessary to focus on changes in the amount of electricity used, it is possible to detect households where children are giving birth more easily and efficiently.

<Fourth embodiment>
Next, a fourth embodiment of the household composition change detection device will be described. This embodiment basically has the same configuration as the first embodiment shown in Figures 1 to 3. Therefore, the same members and configurations as those in the first embodiment are given the same reference numerals, and detailed descriptions thereof will be omitted.

As in the third embodiment, the detection device 10 detects households with a decrease in the number of people in the household based on changes in power usage. As mentioned above, in households with a decrease in the number of people in the household, power usage tends to decrease in the mornings on holidays, regardless of the age of the person who leaves the household. However, there are time periods specific to the age group in which characteristic changes in power usage occur only for the age group of people who have left the household.

For example, when a child-generation household member in their 20s or 30s leaves the household, electricity usage on weekday evenings tends to decrease. This trend is thought to be due to the following. For example, it is assumed that the parent generation in their 20s or 30s who leave the household are older and tend to go to bed earlier than their children. For this reason, it is thought that the decrease in electricity usage following the departure of a household member in their 20s or 30s is more noticeable.

In addition, when an elderly household member aged 60 or over leaves the household, electricity usage during weekday mornings tends to decrease. This trend is thought to be due to the following. For example, it is assumed that the rate of people other than elderly people being at home during weekday mornings is low. For this reason, it is thought that the decrease in electricity usage following the departure of an elderly household member is more noticeable.

Therefore, based on the trend of changes in electricity usage on weekday evenings, it is possible to detect households from which members in their 20s and 30s are likely to have left. Also, based on the trend of changes in electricity usage on weekday mornings, it is possible to detect households from which members of the elderly generation are likely to have left.

The storage device 10B stores attribute information for each household. The attribute information is associated with identification information unique to the smart meter 30 in each household and location information for the smart meter 30. The attribute information is obtained, for example, through a questionnaire survey of each household. The attributes are characteristics or properties that each household possesses, and include information such as the number of people in the household, gender, and age group.

<Detection algorithm>
The control device 10C detects households where the number of household members has decreased due to the departure of a household member of a specific age group according to a detection algorithm. In other words, the control device 10C selectively detects households where a household member of a specific age group has left. The control device 10C detects households where a person of a specific age group has left based on changes in power usage in the following two time periods (E1) and (E2).

(E1) A time period in which characteristic changes in electricity usage are seen across all age groups. In this case, it is the morning of a holiday.
(E2) Time periods specific to different generations in which there are characteristic changes in electricity usage depending on the generation. For the younger generation in their 20s and 30s, it is the evening on weekdays, and for the elderly generation aged 60 or older, it is the morning on weekdays.

The control device 10C calculates the dropout probability for each age group using logistic regression, which is an algorithm for classification.
The control device 10C calculates the child generation departure probability P _child based on the following equation (5).

P _child =F _child・inv_logit(α ₀ −α ₁・Y ₁ −α ₂・Y ₂ )…(5)
Here, "F _child " is the value of a flag indicating that the household has children in their 20s to 30s. "logit()" is the logit function, and "inv_logit" is the inverse function of the logit function. "α ₀ ", "α ₁ ", and "α ₂ " are weights that are determined through learning of power data using logistic regression. "Y ₁ " is the year-on-year change in power usage in the morning on a holiday. "Y ₂ " is the year-on-year change in power usage in the evening on a weekday.

The control device 10C calculates the dropout probability P _senior of the elderly generation based on the following equation (6).
P _senior =F _senior・inv_logit(α ₀ −α ₁・Y ₁ −α ₂・Y ₃ )…(6)
Here, "F _senior " is the value of a flag indicating that there is an elderly person aged 60 or older in the household. "logit()" is the logit function, and "inv_logit" is the inverse function of the logit function. "α ₀ ", "α ₁ ", and "α ₂ " are weights that are determined through learning of power data using logistic regression. "Y ₁ " is the year-on-year change in power usage during the morning on a holiday. "Y ₃ " is the year-on-year change in power usage during the morning on a weekday.

The control device 10C sets the value of the flag F _child based on the attribute information of each household stored in the storage device 10B. When a household has a child generation in their 20s to 30s, the control device 10C sets the value of the flag F _child to "1." When a household does not have a child generation in their 20s to 30s, the control device 10C sets the value of the flag F _child to "0."

The control device 10C sets the value of the flag F _senior based on the attribute information of each household stored in the storage device 10B. When the household contains an elderly person aged 60 or older, the control device 10C sets the value of the flag F _senior to "1." When the household contains no elderly person aged 60 or older, the control device 10C sets the value of the flag F _senior to "0."

The control device 10C calculates the weighted sum of the year-on-year comparison of the amount of electricity used compared to the same month of the previous year, and obtains the logit, which is a value obtained by logit-transforming the calculated weighted sum, as the result of the logistic regression. The control device 10C performs an inverse logit transformation on the logit obtained as the result of the logistic regression to calculate the dropout probability P _child of the child generation or the dropout probability P _senior of the elderly generation. The dropout probability P _child of the child generation or the dropout probability P _senior of the elderly generation is a value between "0" and "1". Incidentally, the logit is a value expressed as the natural logarithm of the odds of a probability p. The odds are the ratio (=p/1-p) of the probability of an event occurring with a probability p between the probability of the event occurring and the probability of the event not occurring.

<Accuracy of Estimation of Decreasing Household Size: Independence of Children>
The estimation accuracy for households whose children have become independent (child independence estimation accuracy) is as follows. For example, let us consider the case where 12 households whose children have become independent out of 197 households are captured.

As shown in the top row of the table in FIG. 12, for example, if the top 20 people with the highest values of the offspring dropout probability P _child are selected from 197 people, the household selection rate will be 10%. When the household selection rate is 10%, the recall rate will be 42%. In other words, it is possible to correctly guess 5 households out of 12 households whose offspring have become independent. Also, when the household selection rate is 10%, the precision rate will be 25%. In other words, of the 20 people selected, 5 households are truly independent households.

As shown in the graph of FIG. 13, for example, when 20 people are randomly selected from 197 people, the household selection rate is 10% and the recall rate is 10%. From this, it can be seen that the recall rate when 20 people are selected according to the child generation's departure probability P _child based on the power data and household attributes is 4.2 times that when 20 people are randomly selected. In other words, the probability that 20 people selected according to the child generation's departure probability P _child are independent households is 4.2 times that when 20 people are randomly selected. In this way, by using the departure probability P _child based on the power data and household attributes, it is possible to improve the detection accuracy of independent households regardless of the proportion of households to be selected.

By the way, although the details are omitted, by appropriately narrowing down the number of households based on the attributes of each household, it is possible to narrow down the number of households that are highly likely to be independent households, that is, households in which the child generation in their 20s to 30s are present as household members, to 127 households. Based on this narrowing down, if 20 households are selected from 197 households (effectively, the narrowed down 127 households), that is, if the household selection rate is 10%, the recall rate is 16%. It is possible to guess 1.9 households out of 12 households in which the child generation has become independent. From this, it can be seen that the recall rate when 20 people are selected according to the child generation's withdrawal probability P _child is 2.6 times (= 5 households / 1.9 households) higher than when 20 people are selected based only on the attributes of each household. In other words, it is possible to extract independent households more effectively by using a combination of electricity data and the attributes of each household than by using only the attributes of each household.

<Estimation accuracy of the decrease in household size: death of elderly people>
The estimation accuracy for households where an elderly member has passed away (elderly bereavement estimation accuracy) is as follows. For example, let us consider the case where 4 households out of 197 households where an elderly member has left the household are captured.

As shown in the top row of the table in FIG. 14, for example, if the top 20 people with the highest value of the elderly generation's defection probability P _senior are selected from 197 people, the household selection rate will be 10%. When the household selection rate is 10%, the recall rate will be 75%. In other words, it is possible to correctly identify three out of four households from which the elderly generation has defected. Also, when the household selection rate is 10%, the precision rate will be 15%. In other words, of the 20 people selected, three households are actually households from which the elderly generation has defected.

As shown in the graph of FIG. 15, for example, when 20 people are randomly selected from 197 people, the household selection rate is 10% and the recall rate is 10%. From this, it can be seen that the recall rate when 20 people are selected according to the elderly generation's departure probability P _senior based on the power data and household attributes is 7.5 times that when 20 people are selected randomly. In other words, the probability that 20 people selected according to the elderly generation's departure probability P _senior are independent households is 7.5 times that when 20 people are selected randomly. In this way, by using the elderly generation's departure probability P _senior based on the power data and household attributes, it is possible to improve the detection accuracy of households from which the elderly generation has left, regardless of the proportion of households to be selected.

By the way, although the details are omitted, by appropriately narrowing down the number of households based on the attributes of each household, it is possible to narrow down the number of households that are highly likely to be households from which the elderly generation has left, that is, households with members in their 60s or older, to 52 households. Based on this narrowing down, if 20 households are selected from 197 households (effectively, the narrowed down 52 households), that is, if the household selection rate is 10%, the recall rate is 38%. It is possible to guess 1.5 households out of 4 households from which the elderly generation has left. From this, it can be seen that the recall rate when 20 people are selected according to the elderly generation's departure probability P _senior is 2.0 times (= 3 households / 1.5 households) higher than when 20 people are selected based only on the attributes of each household. In other words, it is possible to extract households from which the elderly generation has left more effectively by using a combination of electricity data and the attributes of each household than by using only the attributes of each household.

<Advantages of the Fourth Embodiment>
(4-1) By finding a common trend in change in electricity usage for each age group of household members who have left the household, it is possible to effectively and efficiently detect households in which a member of a specific age group has left the household based on changes in electricity usage. For example, there may be a time period in which a characteristic change in electricity usage is observed depending on the age group of household members who leave the household. In this case, it is possible to detect households in which the number of household members has decreased due to the departure of a member of a specific age group based on the trend in change in electricity usage during a time period in which a characteristic change in electricity usage is observed only for a specific age group. In addition, since it is necessary to focus only on changes in electricity usage, it is possible to more easily and efficiently detect households in which the number of household members has decreased due to the departure of a member of a specific age group.

(4-2) Households where the number of household members has decreased due to the departure of children in their 20s and 30s tend to have a decrease in electricity usage on weekday evenings. Therefore, based on the trend of changes in electricity usage on weekday evenings, it is possible to detect households where the number of household members has decreased due to the departure of children in their 20s and 30s.

(4-3) Households where the number of household members has decreased due to the departure of elderly people aged 60 or older tend to have a decrease in electricity usage on weekday evenings. Therefore, it is possible to detect households where the number of household members has decreased due to the departure of elderly people aged 60 or older, based on the trend of changes in electricity usage in weekday mornings.

(4-4) Based on the year-on-year comparison of electricity usage on weekday evenings, detect households in which the number of household members has decreased due to the departure of children in their 20s and 30s. The smaller the year-on-year comparison of electricity usage on weekday evenings, the higher the probability that the household is a household in which children in their 20s and 30s have left. Also, based on the year-on-year comparison of electricity usage on weekday mornings, detect households in which the number of household members has decreased due to the departure of elderly people in their 60s or older. The smaller the year-on-year comparison of electricity usage on weekday mornings, the higher the probability that the household is a household in which elderly people in their 60s or older have left. In this way, by focusing on the year-on-year comparison of electricity usage, seasonality can be removed, that is, the effects of seasonal fluctuations in electricity usage can be suppressed. This makes it possible to effectively detect households in which the number of household members has decreased.

(4-5) By appropriately narrowing down the search based on the attribute information of each household before the household composition changed, it is possible to narrow down the households with household members of a specific generation. In other words, based on the attribute information of each household, it is possible to easily exclude households from which household members other than those of the desired age group have left. This makes it possible to more effectively and efficiently detect households from which household members of a specific generation have left. For example, by narrowing down the search to households with children in their 20s and 30s, it is possible to efficiently detect households from which the children have left. In addition, by narrowing down the search to households with elderly people aged 60 or older, it is possible to efficiently detect households from which the elderly generation has left.

(4-6) It is possible to detect households where a member of a specific age has left as a change in household composition. This detection result can be used in corporate marketing or sales activities. For example, the departure of a member of a specific age from a household may result in a mismatch between the property in which the person lives and the household composition. For this reason, the departure of a member of a specific age from a household can be taken as a sign of property renovation or rehousing.

Fifth embodiment
Next, a fifth embodiment of the household composition change detection device will be described. This embodiment basically has the same configuration as the first embodiment shown in Figures 1 to 3. Therefore, the same members and configurations as those in the first embodiment are given the same reference numerals, and detailed descriptions thereof will be omitted.

The detection device 10 detects households whose working style has changed, in this case, households who have started telecommuting. As a result of searching for common characteristics among households who have started telecommuting among various households with different lifestyle patterns, households who have started telecommuting tend to use more electricity during the day on weekdays. This is thought to be due to the use of lighting fixtures, air conditioners, electronic devices, etc. at home when telecommuting begins. In particular, characteristic changes in electricity usage often occur from April to May. This is because many companies set one accounting period as one year from April to March of the following year, and it is thought that changes in working style are often made around the beginning of the accounting period. Therefore, it is possible to detect households that are likely to have started telecommuting based on the trend of changes in electricity usage during the day on weekdays from April to May. Incidentally, daytime refers to the period from 9:00 to 17:00, for example.

<Detection algorithm>
The control device 10C detects households that have started telecommuting according to a detection algorithm. The control device 10C determines that the households have started telecommuting in descending order of the value of the year-on-year change in power usage during the period from April to May from 9:00 to 17:00 on weekdays. The control device 10C sorts the households in descending order of the value of the year-on-year change in power usage during the period from April to May from 9:00 to 17:00 on weekdays, and extracts a predetermined number of households from the top of the sorted order.

<Estimated accuracy of telecommuting start date>
The estimation accuracy of households that started telecommuting (estimation accuracy of telecommuting start) is as follows. Here, we consider the case where 29 households out of 197 households were captured that started telecommuting.

As shown in the top row of the table in Figure 16, for example, if the top 20 people with the highest year-on-year change in electricity usage during the period from April to May and from 9:00 to 17:00 on weekdays are selected from 197 people, the household selection rate will be 10%. When the household selection rate is 10%, the recall rate will be 21%. In other words, it is possible to correctly identify 6 of the 29 households that have started telecommuting. Also, when the household selection rate is 10%, the precision rate will be 30%. In other words, of the 20 people selected, 6 households actually started telecommuting.

As shown in the graph in Figure 17, for example, if 20 people are randomly selected from 197 people, the household selection rate is 10% and the recall rate is 10%. This shows that the recall rate when 20 people are selected based on electricity data, i.e., changes in electricity usage, is 2.1 times that of when 20 people are selected randomly. In addition, the probability that the 20 people selected based on changes in electricity usage are households that have started telecommuting is 2.0 times (6 households / 2.9 households) that of when 20 people are selected randomly. In this way, by using electricity data, households that have started telecommuting can be effectively and efficiently detected regardless of the proportion of households selected.

<Advantages of the Fifth Embodiment>
(5-1) By finding characteristics or trends in changes in electricity usage common to households that have started telecommuting, households that have started telecommuting can be effectively and efficiently detected based on changes in electricity usage. Households that have started telecommuting tend to have increased electricity usage from 9:00 to 17:00 on weekdays from April to May. Therefore, it is possible to detect households that have started telecommuting based on the trend in changes in electricity usage from 9:00 to 17:00 on weekdays from April to May. In addition, since it is only necessary to focus on changes in electricity usage, it is easier and more efficient to detect households that have given birth.

(5-2) Detect households that have started telecommuting based on the year-on-year comparison of electricity usage during the period from April to May from 9:00 to 17:00 on weekdays. For example, the greater the year-on-year comparison of electricity usage during the period from April to May from 9:00 to 17:00 on weekdays, the higher the probability that the household has started telecommuting. By focusing on year-on-year comparison of electricity usage, seasonality can be removed, i.e., the effects of seasonal fluctuations in electricity usage can be reduced. This makes it possible to effectively detect households that have started telecommuting.

(5-3) It is possible to detect households that have started telecommuting as a change in household composition. This detection result can be used in corporate marketing or sales activities. For example, the start of telecommuting may result in a mismatch between the property in which the person lives and the household composition or lifestyle. For this reason, the start of telecommuting can be seen as a sign of property renovation or rehousing.

The fifth embodiment may be modified as follows.
That is, the detection device 10 detects households that have started telecommuting based on the trend of changes in electricity usage during the daytime on weekdays. Households that have started telecommuting tend to increase their electricity usage between 9:00 and 17:00 on weekdays. Therefore, it is possible to detect households that have started telecommuting based on the trend of changes in electricity usage between 9:00 and 17:00 on weekdays. Since it is only necessary to focus on changes in electricity usage, it is possible to detect households that have started to give birth more easily and efficiently.

Sixth embodiment
Next, a sixth embodiment of the household composition change detection device will be described. This embodiment basically has the same configuration as the first embodiment shown in Figures 1 to 3. Therefore, the same members and configurations as those in the first embodiment are given the same reference numerals, and detailed descriptions thereof will be omitted.

The detection device 10 detects households with a new elementary school member. An elementary school household is one in which there is a member who has entered elementary school. As a result of searching for common characteristics of elementary school households, it was found that electricity usage in elementary school households tends to increase from before noon to just after noon in August. This is thought to be because, for example, nursery schools do not generally have summer vacations, whereas elementary schools do. Therefore, based on the trend in changes in electricity usage from before noon to just after noon in August, it is possible to detect households with a high probability of having a member who has entered elementary school. Incidentally, the period from before noon to just after noon refers to the period from 10:30 to 1:30 p.m., for example.

<Detection algorithm>
The control device 10C detects households entering elementary school according to a detection algorithm. The control device 10C determines that the households are elementary school enrolled in order of the largest value of the power usage amount in the period from 10:30 to 13:30 in August compared to the same month of the previous year. The control device 10C sorts the households in order of the largest value of the power usage amount in the period from 10:30 to 13:30 in August compared to the same month of the previous year, and extracts a predetermined number of households from the top of the sorted order.

<Accuracy of elementary school enrollment estimates>
The estimation accuracy of households with children entering elementary school is as follows. Here, we consider the case where 18 households who entered elementary school out of 63 households that had children in the previous year were captured. Incidentally, households that had children in the previous year were extracted based on attribute information.

As shown in the graph in Figure 18, for example, if we select the top six households with the highest year-on-year change in electricity usage between 10:30 and 13:30 in August from 63 households that had young children in the previous year, the household selection rate is 10% and the recall rate is 22%. In other words, we can guess four of the 18 households that enrolled in elementary school.

In contrast, if six people are selected randomly from the 63 households that had young children in the previous year, the household selection rate is 10% and the recall rate is 10%. This shows that the recall rate when six people are selected based on electricity data is 2.2 times higher than when six people are selected randomly. Furthermore, the probability that the 20 people selected based on changes in electricity usage are elementary school enrollment households is 2.2 times higher (4 households / 1.8 households) than when six people are selected randomly. In this way, by using electricity data, elementary school enrollment households can be detected effectively and efficiently, regardless of the proportion of households selected.

<Advantages of the Sixth Embodiment>
(6-1) By finding characteristics or trends in changes in electricity usage common to households entering elementary school, households entering elementary school can be effectively and efficiently detected based on changes in electricity usage. Households entering elementary school tend to have an increase in electricity usage from before noon to after noon in August. Therefore, it is possible to detect households with members who have entered elementary school based on the trend in changes in electricity usage from before noon to after noon in August. In addition, because it is necessary to focus only on changes in electricity usage, households entering elementary school can be detected more easily and efficiently.

(6-2) Detect the start of elementary school based on the year-on-year comparison of electricity usage from before noon to after noon in August to the same month of the previous year. The higher the year-on-year comparison of electricity usage from before noon to after noon in August to the same month of the previous year, the higher the probability that the household is an elementary school starter household. By focusing on year-on-year comparison of electricity usage, seasonality can be removed, that is, the effects of seasonal fluctuations in electricity usage can be reduced. This makes it possible to effectively detect households starting elementary school.

(6-3) Households that are confirmed to have young children are extracted in advance based on the attribute information of each household. By focusing on the change in electricity usage of the extracted households from before noon to after noon in August, it is possible to more efficiently detect households that are about to start elementary school.

(6-4) It is possible to detect households where a child is entering elementary school as a change in household composition. The detection results can be used in corporate marketing or sales activities. For example, changes in household composition due to children entering elementary school may result in a mismatch between the property in which the person lives and the household composition. For this reason, changes in household composition due to children entering elementary school can be seen as a sign of property renovation or rehousing.

10...Detection device (household composition change detection device)
10C: Control device 20: Communication network 30: Smart meter

Claims (5)

A household composition change detection device having a control device that acquires the amount of electricity used by each household from an electricity meter having a communication function via a communication network,
The control device is configured to detect a change in the household composition based on a trend of change in the amount of electricity used that is common to the change in the household composition to be captured,
The change in household composition is a birth,
The control device is configured to detect a birth household, which is a household whose household composition has changed due to a birth, based on a change in daytime electricity usage for at least six consecutive months;
The control device is further configured to calculate the maximum value throughout the year of the sum of squares of the upward deviation of daytime electricity usage from the median for the same month of the previous year over at least six consecutive months as a birth score indicating the likelihood that the household is a birthing household, and to detect birthing households based on the calculated birth score, in the household composition change detection device. Further comprising a storage device configured to store attribute information of each household;
The household composition change detection device of claim 1 , wherein the control device is configured to reduce the value of the birth score of households having attributes that are considered to be less likely to give birth based on the attribute information of each household stored in the storage device. A household composition change detection device having a control device that acquires the amount of electricity used by each household from a power meter having a communication function via a communication network , and a storage device configured to store attribute information of each household ,
The control device is configured to detect a change in the household composition based on a trend of change in the amount of electricity used that is common to the change in the household composition to be captured,
The change in household composition is a decrease in the number of people in the household,
The control device is configured to detect a household in which the number of household members has decreased due to the departure of a household member of a specific age group, based on attribute information of each household stored in the storage device and a change in electricity usage during a time period that is set as a time period in which a characteristic change in electricity usage occurs depending on the age of the household member who leaves the household,
The control device is a household composition change detection device configured to detect a household in which the number of household members has decreased due to the departure of children in their 20s and 30s based on a change in electricity usage on weekday nights . A household composition change detection device having a control device that acquires the amount of electricity used by each household from a power meter having a communication function via a communication network , and a storage device configured to store attribute information of each household ,
The control device is configured to detect a change in the household composition based on a trend of change in the amount of electricity used that is common to the change in the household composition to be captured,
The change in household composition is a decrease in the number of people in the household,
The control device is configured to detect a household in which the number of household members has decreased due to the departure of a household member of a specific age group, based on attribute information of each household stored in the storage device and a change in electricity usage during a time period that is set as a time period in which a characteristic change in electricity usage occurs depending on the age of the household member who leaves the household,
The control device is a household composition change detection device configured to detect households where the number of household members has decreased due to the departure of elderly people aged 60 or older, based on changes in electricity usage during weekday mornings . A household composition change detection method in which a control device acquires the amount of electricity used by each household from a power meter having a communication function via a communication network, and detects a change in the household composition based on a trend in the amount of electricity used that is common to changes in the household composition to be captured,
The change in household composition is a birth,
The control device detects a birth household, which is a household whose household composition has changed due to a birth, based on a change in daytime electricity usage for at least a consecutive six months;
The method for detecting changes in household composition further includes the control device calculating the maximum value throughout the year of the sum of squares of the upward deviation of daytime electricity usage from the median for the same month of the previous year over at least six consecutive months as a birth score indicating the likelihood that the household is a birthing household, and detecting the birthing household based on the calculated birth score .

Images