Household Attribute Estimation Device and Household Attribute Estimation Method

Abstract

To provide a household attribute estimation device and a household attribute estimate method which can improve the estimation accuracy of a household attribute.SOLUTION: A controller 10C of an estimation device acquires a used power amount of each household via a communication network from a smart meter. The controller 10C comprises an attribute explanation variable creation unit 41, a power explanation variable creation unit 42, a connection processing unit 43 and an estimation processing unit 44. The attribute explanation variable creation unit 41 creates an attribute explanation variable based on data about an electricity contract of each household. The power explanation variable creation unit 42 creates a power explanation variable based on power data of each household acquired through the smart meter. The connection processing unit creates a final explanation variable by connecting the attribute explanation variable and the power explanation variable. The estimation processing unit 44 estimates a category to which each household belongs from among a plurality of categories of a household segment which is an attribute set as an objective variable, based on the final explanation variable created by the connection processing unit.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to a household attribute estimation device and a household attribute estimation method.

Background Art

[0002] In recent years, effective utilization of the power consumption of each household in a region has been studied. For example, the household size estimation device of Patent Document 1 estimates the household size of each household in a region based on the power consumption by time period measured by a power meter during the cooling operation in summer or the heating operation in winter. The estimated household size of each household is used, for example, to assume the evacuation situation in the region when a disaster occurs in the region.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] For example, in marketing activities or sales activities in a company, it may be required to grasp household attributes such as household composition or household size. It is desired to further improve the estimation accuracy of household attributes.

Means for Solving the Problems

[0005] The household attribute estimation device capable of solving the above problems has a control device that acquires the power consumption of each household via a communication network from a power meter having a communication function. The control device includes a first creation unit that creates a first explanatory variable based on at least data related to the electricity contract of each household, a second creation unit that creates a second explanatory variable based on at least the power data of each household acquired through the power meter, a combination processing unit that creates a final explanatory variable by combining the first explanatory variable created by the first creation unit and the second explanatory variable created by the second creation unit, and an estimation processing unit that estimates the category to which each household belongs from among a plurality of categories of household segments that are attributes set as target variables, based on the final explanatory variable created by the combination processing unit.

[0006] According to this configuration, a final explanatory variable is created by combining a first explanatory variable based on at least data related to the electricity contract of each household and a second explanatory variable based on at least the power data of each household acquired through the power meter. Using this final explanatory variable, the category of the household segment to which each household belongs is estimated. Each category is an attribute set as a target variable. Therefore, the estimation accuracy of household attributes can be improved.

[0007] In the above household attribute estimation device, the second creation unit may remove the power consumption of a specific device that is a contributing factor to the sharp increase in power consumption at night, which is outside the human activity time, from the power data of each household acquired through the power meter, and use the power data of each household after this removal to create the second explanatory variable.

[0008] According to this configuration, by removing the power consumption of a specific device that is a contributing factor to the sharp increase in power consumption at night, which is outside the human activity time, from the power data of each household acquired through the power meter, the estimation accuracy of household attributes can be further improved.

[0009] In the above household attribute estimation device, the specific device may be an electric water heater. As described above, it is assumed that the electric water heater boils water using inexpensive nighttime power. For this reason, there is a risk that the power consumption at midnight will increase sharply due to the influence of the electric water heater. Therefore, it is preferable to remove the power consumption of the electric water heater at midnight from the power data of each household obtained through the power meter. By doing so, it becomes possible to more accurately estimate the activity time zone of people in each household.

[0010] In the household attribute estimation device described above, the second creation unit may create a plurality of types of the second explanatory variables by calculating the average power consumption for each defined time zone division, for each of the two divisions of weekdays and holidays, and for each of the twelve divisions of months, using the power data of each household from which the power consumption of the specific device has been removed.

[0011] According to this configuration, by adding a plurality of types of second explanatory variables to the final explanatory variables, the estimation accuracy of the household attributes can be further improved. In the household attribute estimation device described above, the second creation unit may create an additional explanatory variable for detecting the difference in people's activities in the morning and evening based on the power data of each household from which the power consumption of the specific device has been removed and the holiday information, and create the second explanatory variable using the created additional explanatory variable.

[0012] According to this configuration, it becomes easier to detect the difference in people's activities in the morning and evening. Therefore, the estimation accuracy of the household attributes can be further improved. In the household attribute estimation device described above, the first creation unit may create the first explanatory variable by combining the correct data created according to a defined rule and an additional explanatory variable created based on external data related to the correct answer.

[0013] According to this configuration, the first explanatory variable is created by combining the correct data and an additional explanatory variable based on external data related to the correct answer. By using this first explanatory variable, the estimation accuracy of the household attributes can be further improved.

[0014] In the household attribute estimation device described above, the estimation processing unit may estimate the category to which each household belongs using an estimation model constructed through machine learning. According to this configuration, by using an estimation model constructed through machine learning, the estimation accuracy of household attributes can be further improved.

[0015] A household attribute estimation method capable of solving the above problems includes: obtaining the power consumption of each household from a power meter having a communication function via a communication network; creating a first explanatory variable based on at least data related to the electricity contract of each household; creating a second explanatory variable based on at least the power data of each household obtained through the power meter; creating a final explanatory variable by combining the first explanatory variable and the second explanatory variable; and estimating the category to which each household belongs from among a plurality of categories of household segments that are attributes set as target variables based on the final explanatory variable.

[0016] According to this method, a final explanatory variable is created by combining a first explanatory variable based on at least data related to the electricity contract of each household and a second explanatory variable based on at least the power data of each household obtained through the power meter. Using this final explanatory variable, the category of the household segment to which each household belongs is estimated. Each category is an attribute set as a target variable. Therefore, the estimation accuracy of household attributes can be improved.

Advantages of the Invention

[0017] According to the household attribute estimation device and the household configuration estimation method of the present invention, the estimation accuracy of household attributes can be improved.

Brief Description of the Drawings

[0018]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Mode for Carrying Out the Invention

[0019] Hereinafter, an embodiment in which a household attribute estimation device and a household attribute estimation method are embodied will be described. The household attribute estimation device estimates a household attribute based on the power consumption of a household. The household attribute is a concept including the household composition and the number of household members.

[0020] As shown in FIG. 1, the estimation device 10 is connected to a communication network 20 such as the Internet. The estimation device 10 is provided in a general power transmission and distribution business operator such as an electric power company. The estimation device 10 can communicate with the smart meter 30 via the communication network 20. The smart meter 30 is a power meter having both a power measurement function and a communication function, and is provided for each household in the area. The smart meter 30 measures the power consumption for a predetermined period (for example, 30 minutes), and transmits power data, which is data indicating the measured power consumption, to the estimation device 10 via the communication network 20. The power data also includes the location information of the smart meter 30. The estimation device 10 estimates the household attribute based on the power data of each household acquired via the communication network 20.

[0021] As shown in FIG. 2, the smart meter 30 includes a measurement device 30A, a storage device 30B, a communication device 30C, and a control device 30D. These measurement device 30A, storage device 30B, communication device 30C, and control device 30D are interconnected via a bus 30E which is a signal line.

[0022] The measurement device 30A measures the power consumption of a household, which is a consumer, every predetermined period (for example, 30 minutes). The measurement device 30A detects the voltage of the power system and the current supplied from the power system to the load. The measurement device 30A measures the power consumption of the household based on the detected voltage and current.

[0023] The memory device 30B stores the power consumption measured by the measuring device 30A according to the instructions from the control device 30D. The memory device 30B also stores the identification information unique to the smart meter 30 and the 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 such as the building where the smart meter 30 is installed.

[0024] 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 exchanges information with an external device connected to the communication network 20 according to the instructions from the control device 30D. The external device includes the estimation device 10.

[0025] The control device 30D comprehensively controls the entire smart meter 30. The control device 30D generates power data. The power data includes the power consumption 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 estimation device 10 via the communication device 30C.

[0026] As shown in FIG. 3, the estimation device 10 includes a communication device 10A, a memory device 10B, and a control device 10C. These communication device 10A, memory device 10B, and control device 10C are interconnected via a bus 10D which is a signal line.

[0027] 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 exchanges information with an external device connected to the communication network 20 according to the instructions from the control device 10C. The external device includes the smart meter 30.

[0028] The memory device 10B has, for example, a main memory unit and an auxiliary memory unit. The main memory unit stores a computer program executed by the control device 10C, data processed by the control device 10C, and the like. The main memory unit has a RAM (Random Access Memory) and a (Read Only Memory). The auxiliary memory unit is a non-volatile memory device that can read and write various programs including an OS (Operating System) and various data. The auxiliary memory unit is a flash memory, a hard disk drive (Hard Disk Drive), an SSD (Solid State Drive), or the like. The OS includes a communication interface program for transferring data to and from an external device connected via the communication device 10A. Information received via the communication device 10A is stored in the auxiliary storage device. This information includes power data sent from the smart meter 30 via the communication network 20.

[0029] The control device 10C executes various processes according to a program stored in the memory device 10B. The control device 10C has a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). The processor comprehensively controls the entire estimation device 10 through the execution of a computer program developed 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 associated with identification information unique to the smart meter 30.

[0030] The control device 10C captures various input information including the power consumption during a specific period (for example, the number of months or time zone), and estimates the household attributes based on the captured input information. <Detailed Configuration of Control Device 10C> Next, the control device 10C will be described in detail.

[0031] As shown in FIG. 4, the control device 10C includes an attribute description variable creation unit 41 which is a first data processing unit, a power description variable creation unit 42 which is a second data processing unit, a combination processing unit 43, and an estimation processing unit 44.

[0032] The attribute description variable creation unit 41 takes in, for example, electricity contract information, registration information, and open data as input information, and creates attribute description variables based on the input information thus captured.

[0033] The data acquisition target person of the electricity contract information is, for example, a contractor who can acquire the number of family members, the number of children, the age of the person himself / herself, and the age of the eldest son. The electricity contract information includes, for example, items such as the number of months of the electricity contract, prefecture, city, town, village, postal code, contract type code, payment method, and electricity consumption in the 0 to 23 months prior.

[0034] The registration information is personal information other than the electricity contract. The data acquisition target person of the registration information is also, for example, a contractor who can acquire the number of family members, the number of children, the age of the person himself / herself, and the age of the eldest son. Personal information other than the electricity contract includes information related to, for example, the number of family members, the number of children, the age of the person himself / herself, the type of gas, the size or floor plan of the house, and the household appliances or water heaters in use. These information are, for example, information registered when conducting questionnaires in campaigns or providing services other than the electricity contract. Information registered when providing services other than the electricity contract includes, for example, web service registration information. Web service registration information is data registered when using, for example, a home-oriented web service provided by an electric power company.

[0035] Open data is data for which prior permission or data format standardization has been carried out so that anyone can freely reuse it. Open data includes materials and survey data published by the government or local governments, etc. An example of open data is, for example, the results of the population census. The results of the population census include items such as the single-person household ratio, two-person household ratio, household ratio with children, and elderly household ratio as the household composition ratio by municipality. Open data is external data related to the correct answer.

[0036] The power explanatory variable creation unit 42 takes in, for example, SM data and holiday information as input information, and creates power explanatory variables based on the input information thus taken in. The data acquisition target households of the SM data are the households that receive the SM data at the data platform unit. The SM data includes the power consumption (electric power consumption) every fixed period (for example, 30 minutes).

[0037] The holiday information is external data. The holiday information includes a holiday flag indicating whether each day is a holiday or not. The combining processing unit 43 creates the final explanatory variables by combining the attribute explanatory variables created by the attribute explanatory variable creation unit 41 and the power explanatory variables created by the power explanatory variable creation unit 42.

[0038] The estimation processing unit 44 constructs an estimation model by performing machine learning using learning data. The learning data is, for example, data held by an electric power company. The algorithm of the machine learning is, for example, "XGboost (eXtreme Gradient Boosting)" or "LightGBM". Both "XGboost" and "LightGBM" are machine learning methods that combine ensemble learning by gradient boosting and decision trees which are weak learners. The estimation processing unit 44 estimates the household attributes based on the final explanatory variables created by the combining processing unit 43.

[0039] Based on the changes in the household composition or lifestyle to be captured, the household attributes are defined, for example, as seven categories (A1) to (A7) as household segments. The seven categories are the target variables.

[0040] (A1) Active single... A household of one person and the age of the contract holder is 69 years old or younger (A2) Active couple... A two-person household without children and the age of the contract holder is 69 years old or younger (A3) Infant... A household of three or more people and the eldest child is 5 years old or younger (A4) Families with 3 or more primary school students and the eldest son aged 6 to 11 (A5) Families with 3 or more junior high school students and the eldest son aged 12 to 20 (A6) Retired single person... A household of 1 person and the age of the contract holder is 70 or above (A7) Retired couple... A household of 2 people and the age of the contract holder is 70 or above The estimation processing unit 44 inputs the final explanatory variable created by the combination processing unit 43 into the estimation model, and infers the category of the household segment based on the output information that is the output from the estimation model.

[0041] <Detailed configuration of the attribute explanatory variable creation unit> Next, the detailed configuration of the attribute explanatory variable creation unit 41 will be described. As shown in FIG. 5, the attribute explanatory variable creation unit 41 includes a correct answer data creation unit 41A, an additional variable creation unit 41B, and a combination processing unit 41C.

[0042] The correct answer data creation unit 41A creates correct answer data based on the previous registration information. The correct answer data creation unit 41A first creates correct answer data according to the correct answer data creation rules. The correct answer data creation rules are, for example, as shown in Table 1 below. "N / A" indicates "not applicable".

[0043]

Table 1

[0044] Next, when the data that is not input to the estimation model is included, the correct answer data creation unit 41A deletes the data that is not input to the estimation model from the correct answer data. As a result, the final correct answer data input to the estimation model is created. The data that is not input to the estimation model is appropriately set according to the product specifications and the like. The data that is not input to the estimation model is, for example, the number of family members, the number of children, and the age of the eldest son.

[0045] The additional variable creation unit 41B creates additional variables based on the electricity contract information and the previous open data. The open data is external data. This is intended to make it easier for the estimation model to capture the correct value from the explanatory variables by assigning numerical values highly relevant to the correct answer to various categories. The additional variables are, for example, the following variables (B1) to (B5).

[0046] (B1) Ratio of single-person households divided by 5-year contract period for each prefecture, city, ward, town, and village (B2) Ratio of two-person households divided by 5-year contract period for each prefecture, city, ward, town, and village (B3) Ratio of households with young children divided by 5-year contract period for each prefecture, city, ward, town, and village (B4) Ratio of households with older children divided by 5-year contract period for each prefecture, city, ward, town, and village (B5) Ratio of elderly households divided by 5-year contract period for each prefecture, city, ward, town, and village The combining processing unit 41C creates the final processed attribute explanatory variables, which are the final attribute explanatory variables, by combining the correct answer data created by the correct answer data creation unit 41A and the additional variables created by the additional variable creation unit 41B.

[0047] <Detailed configuration of the power explanatory variable creation unit> Next, the detailed configuration of the power explanatory variable creation unit 42 will be described. The power explanatory variable creation unit 42 uses disaggregation technology to remove the influence of the night water heater from the data acquired through the smart meter 30 (hereinafter referred to as "SM data"). The power explanatory variable creation unit 42 creates explanatory variables from the SM data from which the influence of the night water heater has been removed. The SM data includes power data indicating the power consumption of each household.

[0048] As shown in FIG. 6, the power explanatory variable creation unit 42 includes a disaggregation unit 42A, an aggregated explanatory variable creation unit 42B, and an additional explanatory variable creation unit 42C. The disaggregation unit 42A uses disaggregation technology to remove the influence of the electric water heater from the SM data. The disaggregation technology refers to a technology for estimating the power consumption of each device in a household based on the SM data.

[0049] The reason for removing the influence of the night water heater from the SM data is as follows. As shown by arrow C1 in the graph of Fig. 7, for example, due to the influence of an electric water heater, the power consumption at midnight may increase sharply. Due to this, there is a risk of misidentifying the human activity time zone. Therefore, it is necessary to remove the influence of the electric water heater from the SM data. The electric water heater is, for example, a natural refrigerant heat pump water heater. A natural refrigerant heat pump water heater refers to a water heater that uses carbon dioxide as a refrigerant among electric water heaters that can boil water using the heat of the air by using heat pump technology. This electric water heater uses night power, which is cheaper in electricity bills, to boil water at night. The hot water is stored in a tank.

[0050] As shown in the graph of Fig. 8, by removing the influence C2 of the electric water heater indicated by the dashed line from the SM data, the power consumption during the time zone when people are not active can be suppressed. As a result, it becomes possible to more accurately estimate the human activity time zone in each household.

[0051] As shown in the graph of Fig. 9, the disaggregation unit 42A acquires the power consumption waveform C3 for one day (0:00 to 24:00) in the household based on the SM data. After that, as shown in the graph of Fig. 10, the disaggregation unit 42A separates the power consumption waveform C3 of the entire household into, for example, four power consumption waveforms C4 to C7. These power consumption waveforms C4 to C7 are the waveforms of the power consumption for each device in the household.

[0052] (C4) Waveform of activity power consumption depending on temperature (C5) Waveform of activity power consumption independent of temperature (C6) Waveform of basic power consumption ΔP3 (C7) Waveform of power consumption by the electric water heater at midnight The power consumption waveform C4 is the waveform of the activity power consumption depending on temperature. The power consumption waveform C4 is, for example, the waveform of the power consumption by a heating and cooling device such as an air conditioner. The method for separating the activity power consumption depending on temperature is as follows.

[0053] As shown in the graph of FIG. 11, the activity power consumption ΔP1 due to temperature is the difference between the power consumption P1 for one day (0:00 to 24:00) in summer and the average power consumption P2 in spring and autumn. The disaggregation unit 42A calculates the activity power consumption ΔP1 due to temperature using the following equation (1).

[0054] ΔP1 = P1 - P2 …(1) The power consumption waveform C5 is the waveform of the activity power consumption that is not affected by temperature. The power consumption waveform C5 is, for example, the waveform of the power consumption due to human activities such as cooking, eating, watching TV, and lighting. The method for separating the activity power consumption that is not affected by temperature is as follows.

[0055] As shown in the graph of FIG. 12, the activity power consumption ΔP2 that is not affected by temperature is the value obtained by subtracting the activity power consumption ΔP1 due to temperature from the household power consumption during the period from 7:00 to 24:00 and then further subtracting the basic power consumption ΔP3 from the subtracted value. The waveform indicated by the solid line in the graph of FIG. 12 is the waveform of the power consumption P3 obtained by subtracting the activity power consumption ΔP1 due to temperature from the household power consumption for the entire period from 7:00 to 24:00. The period from 7:00 to 24:00 is set based on the perspective of not considering the power consumption during the late-night time zone when human activities are few.

[0056] The disaggregation unit 42A calculates the activity power consumption ΔP2 that is not affected by temperature using the following equation (2). ΔP2 = P3 - ΔP3 = (P5 - ΔP1) - ΔP3 …(2) However, “P3” is the power consumption obtained by subtracting the activity power consumption ΔP1 due to temperature from the household power consumption for the entire period from 7:00 to 24:00. “ΔP3” is the basic power consumption. “P5” is the household power consumption for the entire period from 7:00 to 24:00. “ΔP1” is the activity power consumption due to temperature.

[0057] The power consumption waveform C6 is the waveform of the basic power consumption ΔP3. The basic power consumption ΔP3 refers to the minimum power consumed in a household, such as the standby power of a refrigerator. The basic power consumption ΔP3 is set to a value of a set ratio from the lower part of the activity power consumption ΔP2 that does not depend on the temperature in a day, for example. The set ratio is a value expressed as a percentage, for example, 10%. The disaggregation unit 42A calculates, for example, the value of the lower 10% of the activity power consumption ΔP2 that does not depend on the temperature in a day as the basic power consumption ΔP3.

[0058] The power consumption waveform C7 is, for example, the waveform of the power consumption by an electric water heater in the middle of the night when people are not active. The power consumption by the electric water heater in the middle of the night is the value obtained by subtracting the activity power consumption ΔP1 due to temperature, the activity power consumption ΔP2 not depending on temperature, and the basic power consumption ΔP3 from the daily power consumption in a household. The disaggregation unit 42A calculates the power consumption ΔP4 by the electric water heater in the middle of the night using the following formula (3).

[0059] Δp4 = P6 - (ΔP1 + ΔP2 + ΔP3) …(3) However, "P6" is the daily power consumption in a household. The disaggregation unit 42A extracts the part to be used by adding up the power consumption waveforms other than those of the devices that are considered unnecessary among the power consumption waveforms for each device in a household. Here, the power consumption waveform of the device considered unnecessary is the power consumption waveform C7 of the electric water heater. The disaggregation unit 42A creates SM data from which the influence of the electric water heater has been removed by adding up the power consumption waveform C4 (the waveform of the activity power consumption due to temperature), the power consumption waveform C5 (the waveform of the activity power consumption not depending on temperature), and the power consumption waveform C6 (the waveform of the basic power consumption ΔP3).

[0060] As shown in the graph of Fig. 13, assuming that the activity power consumption ΔP2 not depending on the temperature in the period T1 until 24:00 on the previous day and the activity power consumption ΔP2 not depending on the temperature in the period T2 after 7:00 on the current day are known, the disaggregation unit 42A estimates the power consumption when the electric water heater is not used in the period T3 from 24:00 to 7:00.

[0061] Here, although there is a possibility that people are active in the time zones around 24:00 and around 7:00, the possibility that people are active in the late-night time zone (for example, 2:00 to 4:00) is extremely low. Based on this perspective, the disaggregation unit 42A estimates the power consumption in the case where the electric water heater is not used.

[0062] As shown by the dashed line in the graph of FIG. 13, the disaggregation unit 42A estimates the power consumption P7 in the case where the electric water heater is not used so that the value of the power consumption after 24:00 smoothly approaches "0" in the late-night time zone. The disaggregation unit 42A estimates the power consumption P7 using, for example, a Gaussian process. The Gaussian process is a type of continuous-time stochastic process.

[0063] The aggregated explanatory variable creation unit 42B comprehensively aggregates the power consumption from the SM data with the power consumption ΔP4 by the electric water heater in the late night removed to create a power explanatory variable. Here, the power consumption for each time zone or season is considered to be closely related to the household composition. Based on this perspective, the aggregated explanatory variable creation unit 42B creates a power explanatory variable.

[0064] The aggregated explanatory variable creation unit 42B calculates the average value of the power consumption for each of the following divisions (D1) to (D3), for example, based on the SM data for 365 days with the power consumption ΔP4 by the electric water heater in the late night removed. The aggregated explanatory variable creation unit 42B creates "24×2×12" types of power explanatory variables by calculating the average of the power consumption for each "time zone × weekday / holiday × month".

[0065] (D1) 24 divisions of time zones... 0 o'clock to 23 o'clock (D2) 2 divisions of weekdays / holidays... weekdays or holidays (D3) 12 divisions of months... January to December The additional explanatory variable creation unit 42C creates additional explanatory variables focusing on morning and evening activities based on the SM data with the nighttime electric water heater power consumption ΔP4 removed and the holiday information. This is based on the fact that the power consumption due to morning and evening activities tends to vary among households. This additional explanatory variable is an explanatory variable for detecting differences in morning and evening activities. The additional explanatory variables are, for example, the following variables (E1) to (E16).

[0066] (E1) Annual power consumption (E2) Average daily standard deviation of estimated going-out time (E3) Average daily standard deviation of estimated return time (E4) Annual standard deviation of estimated going-out time (E5) Annual standard deviation of estimated return time (E6) Morning power consumption (E7) Afternoon power consumption (E8) Morning flag (E9) Afternoon at-home flag (E10) Work style type determined from morning and afternoon power consumption (E11) Average power consumption from 19:00 to 20:00 in the afternoon (E12) Average power consumption from 20:00 to 22:00 in the afternoon (E13) Ratio of the above two variables E11 and E12 (E14) Average evening peak time (E15) Standard deviation of evening peak time (E16) Standard deviation of power consumption from 0:00 to 23:00 The additional explanatory variables created by the additional explanatory variable creation unit 42C are combined with the power explanatory variables created by the aggregated explanatory variable creation unit 42B. Thereby, the final power explanatory variables are created.

[0067] <An example of the estimation result by the estimation processing unit 44> Next, an example of the estimation result of the household attributes by the estimation processing unit 44 will be described. The estimation processing unit 44 inputs the final explanatory variable created by the combination processing unit 43 into the estimation model, and infers the category of the household segment based on the output information that is the output from the estimation model. The output information from the estimation model is the probability that each household belongs to each category of the household segment. The estimation processing unit 44 estimates the category with the highest probability as the category as the attribute of each household.

[0068] As shown in FIG. 14, the probability that Household ID_1 is a single-person household is 25%, which is the highest. Therefore, the estimation processing unit 44 estimates that Household ID_1 is a single-person household (category A1). The probability that Household ID_2 is an elderly household is 45%, which is the highest. Therefore, the estimation processing unit 44 estimates that Household ID_2 is an elderly household. Note that the elderly household is a combination of the previous categories A6 (retired single) and A7 (retired couple) into one category.

[0069] The probability that Household ID_3 is a two-person household is 43%, which is the highest. Therefore, the estimation processing unit 44 estimates that Household ID_3 is a two-person household. The probability that Household ID_4 is a household with children of junior high school age or older is 55%, which is the highest. Therefore, the estimation processing unit 44 estimates that Household ID_4 is a household with children of junior high school age or older. The probability that Household ID_5 is a household with infants is 29%, which is the highest. Therefore, the estimation processing unit 44 estimates that Household ID_5 is a household with infants.

[0070] In this way, the estimation processing unit 44 estimates the category with the highest corresponding probability as the category of each household. <Estimation accuracy of the estimation processing unit 44> The estimation accuracy of the household attribute, that is, the category, by the estimation processing unit 44 is represented by, for example, recall rate, precision rate, or accuracy rate. The recall rate refers to the ratio of households determined to belong to the corresponding category among the households corresponding to each category. The precision rate refers to the ratio of households belonging to each category among the households selected from the population. The accuracy rate refers to the ratio of households for which the estimation result is correct with respect to the total number of households to be estimated.

[0071] An example of the estimation accuracy of the category by the estimation processing unit 44 is as follows. Here, as shown in FIG. 15, for example, the case of estimating the category using the SM data of 23,114 households will be considered.

[0072] As shown in FIG. 14, the recall rate of category A1 (single) is 78%, the recall rate of category A2 (two people) is 73%, the recall rate of category A3 (infant) is 76%, and the recall rate of category A4 (elementary school student) is 44%. The recall rate of category A5 (junior high school student or above) is 71%, the recall rate of category A6 (retired single) is 40%, and the recall rate of category A7 (retired couple) is 94%.

[0073] The precision rate of category A1 (single) is 81%, the precision rate of category A2 (two people) is 74%, the precision rate of category A3 (infant) is 65%, and the precision rate of category A4 (elementary school student) is 55%. The precision rate of category A5 (junior high school student or above) is 68%, the precision rate of category A6 (retired single) is 66%, and the precision rate of category A7 (retired couple) is 81%.

[0074] The total correct answer rate is 70.9%. In addition, in the present embodiment, the attribute explanatory variable creation unit 41 corresponds to a first creation unit that creates a first explanatory variable based on at least data related to the electricity contract of each household. The attribute explanatory variable corresponds to the first explanatory variable. Also, the power explanatory variable creation unit 42 corresponds to a second creation unit that creates a second explanatory variable based on at least the power data of each household acquired through the smart meter 30. The power explanatory variable corresponds to the second explanatory variable. Also, the electric water heater corresponds to a specific device that contributes to a sharp increase in power consumption during late night when people's activity time is outside.

[0075] <Effects of the Present Embodiment> Therefore, according to the present embodiment, the following effects can be obtained. (1-1) The control device 10C of the estimation device 10 has an attribute explanatory variable creation unit 41, a power explanatory variable creation unit 42, and a combination processing unit 43. The attribute explanatory variable creation unit 41 creates a first explanatory variable based on at least data related to the electricity contracts of each household. The power explanatory variable creation unit 42 creates a power explanatory variable based on at least the power data of each household obtained through the smart meter 30. The combination processing unit 43 creates a final explanatory variable by combining the attribute explanatory variable and the power explanatory variable. Using this final explanatory variable, the category of the household segment to which each household belongs is estimated. Each category is an attribute set as the target variable. Therefore, the estimation accuracy of the household attributes can be improved.

[0076] (1-2) The power explanatory variable creation unit 42 removes the power consumption of the electric water heater, which is a specific device that contributes to a sharp increase in power consumption during late night when people are outside their activity hours, from the power data of each household obtained through the smart meter 30. The power explanatory variable creation unit 42 creates a power explanatory variable using the power data of each household after the power consumption of the electric water heater at night has been removed. Thereby, the estimation accuracy of the household attributes can be further improved. Also, it becomes possible to more accurately estimate the activity time zone of people in each household.

[0077] (1-3) The power explanatory variable creation unit 42 creates multiple types of power explanatory variables by calculating the average power consumption for each of the 24 time zones, for each of the 2 weekdays and holidays, and for each of the 12 months, using the power data of each household from which the power consumption of the specific device has been removed. By taking into account multiple types of power explanatory variables in the final explanatory variable, the estimation accuracy of the household attributes can be further improved.

[0078] (1-4) The power explanatory variable creation unit 42 creates additional explanatory variables for detecting differences in people's activities in the morning and evening based on the power data of each household from which the power consumption of the electric water heater has been removed and the holiday information, and creates power explanatory variables using the created additional explanatory variables. This makes it easier to detect differences in people's activities in the morning and evening. Therefore, the estimation accuracy of household attributes can be further improved.

[0079] (1-5) The attribute explanatory variable creation unit 41 creates attribute explanatory variables by combining the correct data created according to the defined rules and additional explanatory variables created based on external data related to the correct answer. In this way, by using the generated attribute explanatory variables in which the correct data and the additional explanatory variables based on the external data related to the correct answer are combined, the estimation accuracy of household attributes can be further improved.

[0080] (1-6) The estimation processing unit 44 estimates the category to which each household belongs using an estimation model constructed through machine learning. By using an estimation model constructed through machine learning, the estimation accuracy of household attributes can be further improved.

[0081] (1-7) By dividing elderly households into "retired single" households and "retired couple" households, it is possible to capture changes such as from an "elderly couple" household to a "single elderly" household due to the death of a spouse, for example.

[0082] <Other Embodiments> Note that this embodiment may be implemented with the following modifications. · The number of categories of household segments may be appropriately changed according to specifications and the like. The number of categories may be, for example, two or three. Also, the number of categories may be eight or more.

[0083] · The estimation processing unit 44 may use an estimation model constructed through machine learning to estimate the number of household members, which is one of the household attributes, in addition to or instead of the categories (A1 to A7) of the household segments.

[0084] ·The attribute explanatory variable creation unit 41 does not have to perform a process of combining correct answer data created according to a defined rule and additional explanatory variables created based on external data related to the correct answer. The attribute explanatory variable creation unit 41 may create attribute explanatory variables using, for example, only the correct answer data.

[0085] ·The power explanatory variable creation unit 42 does not have to perform a process of creating additional explanatory variables for detecting differences in people's activities in the morning and evening based on the power data of each household from which the power consumption of the electric water heater has been removed and holiday information.

[0086] ·The power explanatory variable creation unit 42 calculates the average power consumption for each of the 24 time zones, for each of the 2 weekdays and holidays, and for each of the 12 months using the power data of each household, but the time zone division for calculating the average power consumption is not limited to 24 divisions. The time zone division for calculating the average power consumption may be more or less than 24 divisions. For example, when the smart meter 30 can measure the power usage every 30 minutes, the power explanatory variable creation unit 42 may calculate the average power consumption for each of the 48 time zones. Also, the power explanatory variable creation unit 42 may calculate the average power consumption for each of the 12 time zones. The power explanatory variable creation unit 42 calculates the average power consumption for each time zone division determined based on specifications and the like.

[0087] ·The power explanatory variable creation unit 42 does not have to perform a process of calculating the average power consumption for each of the 24 time zones, for each of the 2 weekdays and holidays, and for each of the 12 months using the power data of each household from which the power consumption of the electric water heater has been removed.

[0088] ·When there is a specific device that contributes to a sharp increase in the power consumption during late night, which is outside of people's activity hours, in addition to the electric water heater, the power explanatory variable creation unit 42 may remove the power consumption of that device from the power data of each household obtained through the smart meter 30.

[0089] ·The power explanatory variable creation unit 42 does not have to perform a process of removing the power consumption of the electric water heater, which is one of the reasons for the sharp increase in the power consumption during late night when people are outside their activity hours, from the power data of each household obtained through the smart meter 30.

Explanation of Signs

[0090] 10…Estimation device (household attribute estimation device) 20…Communication network 30…Smart meter (power meter) 10C…Control device 41…Attribute explanatory variable creation unit (first creation unit) 42…Power explanatory variable creation unit (second creation unit) 43…Combination processing unit 44…Estimation processing unit

Claims

1. A household attribute estimation device having a control device that acquires the power consumption of each household via a communication network from a power meter having a communication function, wherein the control device includes a first creation unit, a second creation unit, a combination processing unit, and an estimation processing unit, the first creation unit is configured to create a first explanatory variable based on at least data related to the electricity contract of each household, the second creation unit is configured to create a second explanatory variable based on at least the power data of each household acquired through the power meter, the combination processing unit is configured to create a final explanatory variable by combining the first explanatory variable created by the first creation unit and the second explanatory variable created by the second creation unit, the estimation processing unit is configured to estimate, based on the final explanatory variable created by the combination processing unit, the category to which each household belongs from among a plurality of categories of household segments that are attributes set as target variables, the second creation unit is configured to remove the power consumption of a specific device that contributes to a sharp increase in power consumption during late night hours when human activities are outside of normal hours from the power data of each household acquired through the power meter, and to create the second explanatory variable using the power data of each household after this removal, the second creation unit further, separates the power data of each household acquired through the power meter into power consumption due to activity depending on temperature, power consumption due to activity independent of temperature, basic power consumption which is the minimum power consumed in each household, and the power consumption of the specific device, and generates power data in which the power consumption of the specific device is removed by adding together the power consumption due to activity depending on temperature, the power consumption due to activity independent of temperature, and the basic power consumption, A household attribute estimation device configured to perform the above.

2. The household attribute estimation device according to claim 1, wherein the specific device is an electric water heater.

3. The second creation unit creates a plurality of types of the second explanatory variables by calculating the average power consumption for each defined time zone division, for each of two divisions of weekdays and holidays, and for each of twelve divisions of months, using the power data of each household from which the power consumption of the specific device has been removed. The household attribute estimation device according to claim 1 or claim 2.

4. The second creation unit creates additional explanatory variables for detecting differences in morning and evening human activities based on the power data of each household and holiday information from which the power consumption of the specific device has been removed, and creates the second explanatory variable using the created additional explanatory variables. The household attribute estimation device according to any one of claims 1 to 3.

5. The first creation unit creates the first explanatory variable by combining correct data created according to defined rules and additional explanatory variables created based on external data related to the correct answer. The household attribute estimation device according to any one of claims 1 to 4.

6. The estimation processing unit estimates the category to which each household belongs using an estimation model constructed through machine learning. The household attribute estimation device according to any one of claims 1 to 5.

7. A household attribute estimation method in which a control device estimates a household attribute, a first step in which the control device acquires the power consumption of each household from a power meter having a communication function via a communication network; a second step in which the control device creates a first explanatory variable based on at least data related to the electricity contract of each household; a third step in which the control device creates a second explanatory variable based on at least the power data of each household acquired through the power meter; a fourth step in which the control device creates a final explanatory variable by combining the first explanatory variable and the second explanatory variable; a fifth step in which the control device estimates the category to which each household belongs from among a plurality of categories of household segments that are attributes set as target variables based on the final explanatory variable, The third step includes the control device removing the power consumption of a specific device that contributes to a sharp increase in the power consumption during late night, which is outside the human activity time, from the power data of each household acquired through the power meter, and using the power data of each household after the removal to create the second explanatory variable. The third step is the control device separates the power data of each household acquired through the power meter into activity power consumption due to temperature, activity power consumption not due to the temperature, basic power consumption that is the minimum power consumed in each household, and the power consumption of the specific device. The control device generates power data in which the power consumption of the specific device is removed by adding the activity power consumption according to the temperature, the activity power consumption not according to the temperature, and the basic power consumption. A household attribute estimation method further including the above.

Images