Monitoring system

Abstract

To improve accuracy of detection of fluctuation of patterns of wake-up / sleep and presence / absence based on data of power consumption to detect fluctuation of life at an early stage from various viewpoints.SOLUTION: A watching system 1 detects an abnormality of a life situation of a resident in a residence and comprises: a data acquisition unit 11 which acquires power consumption data 2 for every unit time from the outside to be recorded in a power consumption DB 12; an AI prediction unit 13 which estimates the life situation for every unit time of the resident by using a learning model 14 on the basis of the data of power consumption; an AI alert processing unit 15 which determines whether or not the life situation of the resident is in an abnormal state on the basis of an estimation result to output an alert when it is determined that the life situation is in the abnormal state; and a spreadsheet alert processing unit 17 which acquires the data of the power consumption from the power consumption DB 12, determines whether or not the life situation of the resident is in the abnormal state by spreadsheet software, and outputs the alert when it is determined that the life situation is in the abnormal state.SELECTED DRAWING: Figure 1

Description

Technical Field

[0005] , ,

[0006]

[0001] The present invention relates to a technique for monitoring single elderly people and the like, and particularly to a technique effective for application to a monitoring system that performs monitoring based on power consumption.

Background Art

[0002] In Japan, the aging population is progressing, and one in about three Japanese nationals is an elderly person aged 65 or older. Among them, the number of single elderly people is also increasing. Along with this, the number of elderly people who die alone, so-called "lonely deaths," is also increasing, becoming a social problem.

[0003] To prevent the lonely deaths of single elderly people, for example, regular rounds and monitoring by relatives, neighboring residents, medical and nursing staff, etc. are effective. However, since it is realistically impossible to directly monitor 24 hours a day, 365 days a year, as a technical mechanism to complement or replace this, so-called monitoring services and systems that remotely or indirectly monitor and detect changes in the lives of single elderly people, discover abnormalities early, and report and notify them are also being considered.

[0004] Among them, as a method that does not impose a burden on the single elderly person themselves, such as wearing some device, for example, in Japanese Patent Application Laid-Open No. 2007-183890 (Patent Document 1), at a server, data obtained by patternizing the daily power usage status of a user and data on the daily power consumption received from a client terminal installed near the user's home are compared to extract a difference. When the difference exceeds a threshold value, an abnormality is detected and notified to a predetermined destination. [[ID=二十一]] [[ID=二十二]]

[0005] [[ID=二十三]] [[ID=二十四]]In addition, Japanese Patent Application Laid-Open No. 2019-46292 (Patent Document 2) describes a mechanism in which the electricity consumption every predetermined time is compared with the life rhythm pattern analyzed from the transition of the past electricity consumption of each monitoring target person to determine whether an abnormality has occurred in the monitoring target person, and if an abnormality has occurred, it is notified. [[ID=二十五]] [[ID=二十六]]

[0006] [[ID=二十七]] Furthermore, Japanese Patent Publication No. 6830298 (Patent Document 3) describes a mechanism for acquiring usage data indicating the usage status of lifelines in a residence, inputting the usage data into a model trained with past lifeline usage data and data indicating the presence or absence of residents as training datasets, calculating a predicted value for presence or absence of residents, and detecting the frailty status of residents (a state in which physical and cognitive functions have declined due to aging) based on the predicted value and usage data. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2007-183890 [Patent Document 2] Japanese Patent Publication No. 2019-46292 [Patent Document 3] Patent No. 6830298 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] Conventional technology allows for the realization of monitoring services and systems that detect and notify of changes in the lifestyle patterns of elderly people living alone, such as their waking / sleeping times and whether they are at home or not, based on data on the electricity usage of their residences.

[0009] However, conventional technologies still have room for improvement in their accuracy in understanding lifestyle patterns such as wake / sleep and presence / absence based on electricity usage. Furthermore, there is a need for monitoring from various perspectives to enable early detection of lonely deaths and lifestyle changes, including from the standpoint of preventing heatstroke, which has been increasing in recent years. In addition, in the case of systems that use AI (machine learning) for detection, there is a need to know the reasons behind the judgment results.

[0010] Therefore, the object of the present invention is to provide a monitoring system that improves the accuracy of detecting changes in wake / sleep and presence / absence patterns based on electricity usage data, and enables early detection of lonely deaths and lifestyle changes, including the risk of heatstroke, from various perspectives. Another object of the present invention is to provide a monitoring system that enables inference of the reasons for the judgment when a lifestyle change is detected.

[0011] The aforementioned and other objectives and novel features of the present invention will become apparent from the description herein and the accompanying drawings. [Means for solving the problem]

[0012] A brief overview of some of the representative inventions disclosed in this application is as follows:

[0013] A typical embodiment of the present invention is a monitoring system for detecting abnormalities in the living conditions of residents in a residence, comprising: a data acquisition unit that acquires data on the amount of electricity used in the residence at predetermined intervals from an external source and records it in an electricity usage data recording unit; an AI prediction unit that acquires the electricity usage data from the electricity usage data recording unit and estimates the living conditions of the residents at predetermined intervals based on the data using a learning model previously generated by machine learning; an AI alert processing unit that determines whether the living conditions of the residents are in an abnormal state based on the estimation results by the AI ​​prediction unit and outputs an alert if it is determined that they are in an abnormal state; and a spreadsheet alert processing unit that acquires the electricity usage data from the electricity usage data recording unit and determines whether the living conditions of the residents are in an abnormal state using spreadsheet software based on the data and outputs an alert if it is determined that they are in an abnormal state. [Effects of the Invention]

[0014] The effects obtained by some of the representative inventions disclosed in this application can be briefly explained as follows.

[0015] That is, according to a representative embodiment of the present invention, it is possible to improve the accuracy of detecting the modulation of waking / sleeping and presence / absence patterns based on power consumption data, and to detect early from various viewpoints about sudden death alone including the risk of heatstroke and the modulation of life. In addition, it becomes possible to infer the reason for judgment when detecting the modulation of life.

Brief Description of the Drawings

[0016] [Figure 1] It is a diagram showing an outline of a configuration example of a monitoring system which is an embodiment of the present invention. [Figure 2] It is a diagram showing an outline of the detection content of an abnormal state by AI in an embodiment of the present invention. [Figure 3] It is a diagram showing an outline of the detection content of an abnormal state by spreadsheet software in an embodiment of the present invention. [Figure 4] It is a flowchart showing an outline of an example of an alert processing flow by an AI alert processing unit in an embodiment of the present invention. [Figure 5] It is a diagram showing an example of representing the result of performing cluster analysis on the estimated data of a subject in one day in an embodiment of the present invention on a scatter diagram. [Figure 6] It is a flowchart showing an outline of an example of another alert processing flow by an AI alert processing unit in an embodiment of the present invention. [Figure 7] It is a flowchart showing an outline of an example of another alert processing flow by an AI alert processing unit in an embodiment of the present invention. [Figure 8] It is a flowchart showing an outline of an example of an alert processing flow by a spreadsheet alert processing unit in an embodiment of the present invention. [Figure 9] It is a flowchart showing an outline of an example of another alert processing flow by a spreadsheet alert processing unit in an embodiment of the present invention. [Figure 10]This is a flowchart outlining an example of the flow of other alert processing by the spreadsheet alert processing unit in one embodiment of the present invention.

Embodiment for Carrying Out the Invention

[0017] Hereinafter, embodiments of the present invention will be described in detail based on the drawings. In all the drawings for explaining the embodiments, the same parts are generally denoted by the same reference numerals, and repeated explanations thereof are omitted. On the other hand, for the parts explained with reference numerals in a certain figure, they will not be shown again in the explanations of other figures, but may be referred to with the same reference numerals.

[0018] <Overview> Regarding the power consumption of each household, conventionally, a meter reader visually inspected the meter once a month to measure the total monthly consumption. However, in recent years, the so-called smart meters have been spreading nationwide. For example, the power consumption every 30 minutes can be automatically inspected remotely, and users can refer to it at any time via a web service.

[0019] A monitoring system, which is one embodiment of the present invention, uses data such as power consumption measured by a smart meter to indirectly judge the usage status such as the waking / sleeping and presence / absence of residents (mainly single elderly people), and detect abnormal states, so as to discover early the risk of a resident dying alone, suffering from heatstroke, or having a disrupted life. In this embodiment, by providing a function of judgment by AI (machine learning) and a function of judgment based on rules implemented using the macro function of spreadsheet software such as Microsoft Excel (registered trademark), the accuracy of detection and judgment is improved, and the reason for the judgment by the black-boxed AI can be inferred to a certain extent.

[0020] Figure 2 is a diagram illustrating the detection of abnormal conditions by AI in one embodiment of the present invention. In this embodiment, based on time-series data of each resident's electricity usage measured by a smart meter, the AI ​​estimates three items as shown on the left side of the figure: whether the resident is at home or not (present / absent), whether they are sleeping or not (sleeping / not sleeping), and whether the air conditioner is running or not (air conditioner ON / OFF). If the values ​​of these estimation results meet predetermined conditions, it is considered an abnormal condition (or there is a possibility of one), and three types of alerts are output as shown on the right side of the figure. Specifically, an alert indicating an abnormal value is output based on the estimation results of present / absent, sleep / not sleeping, and air conditioner ON / OFF. An alert for lifestyle disturbance (sleep) is also output based on the estimation result of sleep / not sleeping. Furthermore, an alert for heatstroke is output based on the estimation results of present / absent and air conditioner ON / OFF.

[0021] Figure 3 is a diagram illustrating the detection of abnormal conditions using spreadsheet software in one embodiment of the present invention. In this embodiment, as shown on the left side of the figure, time-series data of each resident's electricity usage measured by a smart meter is used as is, and based on the electricity usage, it is estimated whether the resident is in an active state where electricity is being actively used during the day, or in an inactive state such as sleeping or being out (active / inactive).

[0022] This estimation is performed, for example, by converting time-series data of electricity consumption per unit time into data of activity (1) or inactivity (0) based on whether it is greater than or greater than a predetermined baseline value, and then digitizing the data. This makes it easy to process the electricity consumption data using spreadsheet software. The baseline value can be adjusted as needed according to the general trends of electricity consumption increases and decreases, such as monthly or seasonally, to further improve the estimation accuracy.

[0023] Then, a rule-based determination is made as to whether these estimated values ​​meet predetermined conditions. If the conditions are met, it is considered an abnormal state (or potentially abnormal state), and three types of alerts are output, as shown on the right side of the figure. Specifically, an alert is output indicating an abnormal value based on raw power usage data. In addition, alerts indicating an abnormal value and lifestyle disturbances (sleep) are output based on activity / inactivity estimation results. By comparing the output status of these alerts with the output status of alerts based on AI estimation results shown in Figure 2, it is sometimes possible to infer the reason why alerts based on AI estimation results (especially abnormal values ​​and lifestyle disturbances (sleep)) were output. The details of the alert output processing will be described later.

[0024] <System Configuration> Figure 1 is a diagram illustrating an example configuration of a monitoring system, which is one embodiment of the present invention. The monitoring system 1 is composed of a server system, such as a server device or a virtual server built on a cloud computing service, or a computer system such as a PC (Personal Computer). A CPU (Central Processing Unit), not shown in the figure, executes middleware such as an OS (Operating System), DBMS (Database Management System), and web server programs, which are loaded into memory from a storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), and the software that runs on it, thereby realizing various functions related to detecting the safety of residents and changes in their daily lives.

[0025] The monitoring system 1 includes, for example, a data acquisition unit 11, an AI prediction unit 13, an AI alert processing unit 15, and a spreadsheet alert processing unit 17, all implemented as software. It also has a power consumption database (DB) 12 implemented using a database or file table. In the diagram, each function is shown as being implemented on a single device or machine, but the functions may be distributed and implemented across multiple devices or machines.

[0026] The data acquisition unit 11 has the function of acquiring electricity usage data 2 and other data 3 pertaining to the residents of each residence being monitored, and recording them in the electricity usage DB 12. Data may be acquired automatically on a regular basis in cooperation with an external system that serves as a data source, or data may be acquired manually by an operator or other person. As mentioned above, the electricity usage data 2 is data measured by a smart meter and can be acquired, for example, from the system or website of a power company. Other data 3 may include, for example, temperature data referenced to detect the risk of heatstroke, or medical record information to understand the health status of each resident.

[0027] The AI ​​prediction unit 13 has the function of estimating the three items shown in Figure 2 above—presence / absence, sleep / not sleep, and air conditioner ON / OFF—for each resident based on data recorded in DB 12, such as electricity usage, using AI (machine learning). The AI ​​algorithm and engine to be applied are not particularly limited, but in this embodiment, LightGBM (Light Gradient Boosting Machine), an open-source supervised machine learning algorithm based on the decision tree algorithm, will be used.

[0028] Prior to performing estimation using AI, a training data dataset is generated based on electricity usage data obtained from multiple elderly people living alone as subjects, and actual data for each of the target variables: presence / absence, sleep / non-sleep, and air conditioner ON / OFF. A learning model 14 is then generated by performing machine learning based on this dataset. As a preprocessing step for generating the training data dataset, explanatory features are extracted from the electricity usage data, etc. In this embodiment, five items are extracted as features: electricity usage, electricity usage shift, day of the week, time TE (Target Encoding: qualitative data converted to numerical values), and outside temperature.

[0029] The "Power Consumption" item is the cumulative value of power consumption (W) measured every 30 minutes (0.5 hours) by a smart meter. The "Power Consumption Shift" item is the power consumption (W) for the most recent multiple samples (9 samples = 4.5 hours in this embodiment). Since the LightGBM algorithm does not have the concept of "time," this is a feature to understand the transition of power consumption from 4.5 hours ago as a single continuity. The "Day of the Week" item is a numerical representation of each day of the week from Sunday to Saturday, assigned a value from 0 to 6. The "Time TE" item is the time value replaced with the average value of all subjects for the target variable in the training data. The "Outside Temperature" item is outside temperature (°C) data obtained from the Japan Meteorological Agency website, etc.

[0030] By using the features described above, we generated a learning model 14 based on approximately two years' worth of data from about 20 subjects. When we performed estimations for each item—present / absent, sleep / not sleep, and air conditioner ON / OFF—we achieved a high estimation accuracy of approximately 95% for each item.

[0031] The AI ​​alert processing unit 15 has the function of determining whether it is necessary to output alerts for abnormal values, lifestyle disturbances (sleep), and heatstroke, as shown in Figure 2 above, based on the estimation results of each item (presence / absence, sleep / non-sleep, air conditioner ON / OFF) estimated by the AI ​​prediction unit 13, and outputting the corresponding AI alert 16 if it is determined that output is necessary.

[0032] The spreadsheet alert processing unit 17 uses the macro function of a spreadsheet program to estimate activity / inactivity as shown in Figure 3 based on data recorded in DB 12, such as power consumption. It also determines whether it is necessary to output alerts for abnormal values ​​and lifestyle disturbances (sleep) based on the estimated power consumption and activity / inactivity results, and outputs the corresponding spreadsheet alert 18 if it determines that output is necessary. In this embodiment, the macro function of a spreadsheet program is used, but this is not the only configuration, and it may also be implemented as a software program without using a spreadsheet program.

[0033] Note that for both the AI alert 16 and the spreadsheet alert 18, the output destination of the alert can be, for example, an emergency contact or a caregiver that is preset for a single elderly person or the like who is the target person. The target person himself or herself may also be included as the output destination. There are no particular limitations on the output format, method, etc. For example, a message may be push-notified to the mobile terminal or the like of the user who is the output destination of the alert, or an email may be sent. An operator or other operation staff who has confirmed the alert may make a phone call to check on the well-being of the target person.

[0034] <Flow of AI alert processing> FIG. 4 is a flowchart showing an overview of an example of the flow of alert processing by the AI alert processing unit 15 in one embodiment of the present invention. FIG. 4 shows an example of the case where an AI alert 16 indicating that there is an abnormal value is output based on each of the estimation results of presence / absence, sleep / non-sleep, and air conditioner ON / OFF shown in FIG. 2 described above.

[0035] For the target person, first, the estimated data for one day of three types of items, namely presence / absence, sleep / non-sleep, and air conditioner ON / OFF, estimated by the AI prediction unit 13 is acquired (S01). In the present embodiment, a total of 144 data (144 vectors) of 24 hours (48 pieces) every 30 minutes are acquired for each of the three types of items. Then, for these data (vectors), for example, by performing cluster analysis using a known k-means method (S02), dimensionality reduction is performed to two components (two dimensions).

[0036] FIG. 5 is a diagram showing an example of representing the result of performing cluster analysis on the estimated data of a target person for one day in one embodiment of the present invention on a scatter diagram. The origin of the scatter diagram is a group of estimated data determined to be normal, and the distance from the origin of each estimated data plotted on the scatter diagram represents the degree of abnormality of each estimated data. In the present embodiment, estimated data with a distance greater than an arbitrarily set threshold (for example, 5.649 clusters in the example of FIG. 5) is determined to be an abnormal value group.

[0037] In other words, returning to Figure 4, the distance between the estimated data obtained from the cluster analysis in step S02 and the group of estimated values ​​judged to be normal (the origin of the scatter plot) is calculated (S03), and it is determined whether or not there is any estimated data whose distance is above a predetermined threshold (S04). If there is no estimated data above the predetermined threshold (No in step S04), the alert processing related to the anomaly is terminated. On the other hand, if there is estimated data above the predetermined threshold (Yes in step S04), an AI alert 16 related to the anomaly is output (S05), and the alert processing is terminated.

[0038] Figure 6 is a flowchart outlining an example of another alert processing flow by the AI ​​alert processing unit 15 in one embodiment of the present invention. Figure 6 shows an example in which an AI alert 16 related to lifestyle disturbances (sleep) is output based on the sleep / non-sleep estimation results shown in Figure 2 above.

[0039] First, estimated data for one day of sleep / non-sleep items is obtained for the subject (S11). In this embodiment, a total of 48 data points (48 vectors) are obtained for each type of item, with data collected every 30 minutes for 24 hours. Then, these data (vectors) are subjected to dimensionality reduction to two components (2 dimensions) by performing cluster analysis using, for example, the k-means method (S12), similar to the example in Figure 5 above.

[0040] Then, for the estimated data obtained from the cluster analysis in step S12, the distance from the estimated value group judged to be normal (the origin of the scatter plot) is calculated (S13), and it is determined whether or not there is any estimated data whose distance is above a predetermined threshold (S14). Note that the predetermined threshold here may be different from the predetermined threshold in step S04 of Figure 4. If there is no estimated data above the predetermined threshold (No in step S14), the alert processing related to lifestyle disturbance (sleep) is terminated. On the other hand, if there is estimated data above the predetermined threshold (Yes in step S14), an AI alert 16 related to lifestyle disturbance (sleep) is output (S15), and the alert processing is terminated.

[0041] Figure 7 is a flowchart outlining an example of another alert processing flow by the AI ​​alert processing unit 15 in one embodiment of the present invention. Figure 7 shows an example of outputting an AI alert related to heatstroke based on the estimated results of presence / absence and air conditioner ON / OFF, as shown in Figure 2 above.

[0042] For the target individuals, first, estimated data for alert time periods for each item, such as presence / absence and air conditioner ON / OFF, is obtained by the AI ​​prediction unit 13 (S21). Here, the alert time period is arbitrarily set in advance as a time period that is generally assumed to be prone to heatstroke, such as 13:00 to 16:00.

[0043] Subsequently, a loop process is performed on each estimated data point obtained in step S21 (data every 30 minutes during the alert period) in sequence. In the loop process, it is first determined whether the air conditioner is OFF or OFF for the target estimated data point (S22). If the air conditioner is ON (No in step S22), the alert process related to heatstroke is terminated.

[0044] If the air conditioner remains in the OFF state (Yes in step S22), the system then determines whether the target estimated data is in a "present" state (S23). If it is not in a "present" state (No in step S23), the present time counter, which indicates the duration of the present state, is reset (S24), and the system proceeds to processing the next estimated data. On the other hand, if it is in a "present" state (Yes in step S23), the data's unit time (30 minutes in this embodiment) is added to the present time counter (S25).

[0045] Then, it is determined whether the time spent in the room has been longer than a predetermined time (for example, 2 hours in this embodiment) (S26). If the time spent in the room has not yet reached the predetermined time (No in step S25), the process moves on to the next estimated data. On the other hand, if the time spent in the room has been longer than the predetermined time (Yes in step S26), it is further determined whether the outside temperature is higher than a predetermined temperature (for example, 35°C in this embodiment) (S27). If the outside temperature is lower than the predetermined temperature (No in step S27), the process moves on to the next estimated data. On the other hand, if the outside temperature is higher than the predetermined temperature (Yes in step S27), an AI alert 16 related to heatstroke is output (S28), and the alert processing ends.

[0046] In this embodiment, the AI ​​alert 16 related to heatstroke is output when three conditions are met: the air conditioner is constantly OFF during the alert period, the presence of the air conditioner during the alert period lasts for a set time or longer, and the outside temperature is higher than the set temperature. The above conditions are judged based on estimated data every 30 minutes, and if the conditions are met and the AI ​​alert 16 is output even once, it is treated as if the AI ​​alert 16 related to heatstroke was output for that day, even if the conditions are no longer met afterward.

[0047] <Spreadsheet alert processing flow> Figure 8 is a flowchart outlining an example of the alert processing flow by the spreadsheet alert processing unit 17 in one embodiment of the present invention. Figure 8 shows an example in which a spreadsheet alert 18 indicating an abnormal value is output based on the power usage data (cumulative value measured every 30 minutes by a smart meter) as shown in Figure 3 above. As mentioned above, this processing can be executed, for example, by opening the power usage data into cells in spreadsheet software and using a macro function or the like.

[0048] First, data on the electricity usage of the subject over a certain period in the past is obtained from the electricity usage data DB12, and the total electricity usage on a daily basis (daily cumulative electricity usage) is obtained (S31). In this embodiment, for example, electricity usage data for one month (30 days) is obtained by accumulating it on a daily basis. Then, the minimum value among the obtained daily cumulative electricity usage (minimum daily cumulative electricity usage) is calculated (S32), and the calculated minimum daily cumulative electricity usage is further multiplied by a predetermined coefficient (S32). This coefficient can be set in advance to a value of 1 or more, and in this embodiment, for example, the default value is set to 1.05.

[0049] Then, it is determined (S33) whether there have been a predetermined number of consecutive days (for example, 2 days in this embodiment) where the total power consumption for the day (cumulative power consumption for the day) is lower than the minimum daily cumulative power consumption multiplied by a coefficient. If there have been no consecutive days (No in step S33), the alert processing related to power consumption is terminated. On the other hand, if there have been two or more consecutive days (Yes in step S34), a spreadsheet alert 18 related to the abnormal value of power consumption is output (S34), and the alert processing is terminated.

[0050] In step S32, by multiplying the minimum daily cumulative power consumption by a predetermined coefficient to increase the value, a spreadsheet alert 18 can be output earlier when a situation of low power consumption occurs.

[0051] Figure 9 is a flowchart outlining an example of another alert processing flow by the spreadsheet alert processing unit 17 in one embodiment of the present invention. Figure 9 shows an example in which a spreadsheet alert 18 indicating an abnormal value is output based on the activity / inactivity estimation results shown in Figure 3 above. This processing can also be executed, for example, by expanding the power usage data into cells in a spreadsheet program and using a macro function or the like.

[0052] First, the electricity usage data (96 entries) for the most recent two days (48 hours) for the subject is retrieved from the electricity usage data DB12 (S41). Then, for each electricity usage data entry, a value of 1 (active) or 0 (inactive) is set depending on whether it is greater than a pre-set baseline value, and the data is converted into digital data (S42). In the example in Figure 9, the data is converted into digital data based on the baseline at this time, but it is also possible to store the pre-digitized data in the electricity usage data DB12 and retrieve it from there.

[0053] Subsequently, the system determines whether all 96 data points of power consumption for the past two days (estimated activity / inactivity results), which have been digitized, are 1 (active) or 0 (inactive) (S43). If the results are not all 1 or all 0 (No in step S44), the alert processing for the abnormal values ​​in the activity / inactivity estimation results is terminated. On the other hand, if all are 1 or all are 0 (Yes in step S44), a spreadsheet alert 18 regarding the abnormal values ​​in the activity / inactivity estimation results is output (S44), and the alert processing is terminated.

[0054] In addition to cases where the system has been continuously inactive for the past two days, even if the system has been continuously active, a spreadsheet alert 18 can be generated in cases such as forgetting to turn off electrical appliances, or, for example, if the resident becomes unable to move for some reason after using an electrical appliance.

[0055] Figure 10 is a flowchart outlining an example of the flow of other alert processing by the spreadsheet alert processing unit 17 in one embodiment of the present invention. Figure 10 shows an example in which a spreadsheet alert 18 related to lifestyle changes (sleep) is output based on the activity / inactivity estimation results shown in Figure 3 above. This processing can also be executed, for example, by expanding the power consumption data into cells in a spreadsheet program and using a macro function or the like.

[0056] First, electricity usage data for the most recent two days for each subject is obtained from the electricity usage data DB12 (S51), then digitized using a baseline, and a 1 (active) / 0 (inactive) value is estimated (S52). The above processing is the same as steps S41 and S42 in Figure 9 above.

[0057] Subsequently, the average wake-up time over a certain period is calculated (S53), and the wake-up time for the current day is also calculated (S54). The period for which the average wake-up time is calculated can be set to any value in advance, but in this embodiment, for example, the average wake-up time over the most recent two months is calculated. Alternatively, instead of the average wake-up time, the most frequent wake-up time over a certain period may be calculated.

[0058] As a method for calculating the wake-up time on any given day, for example, 96 estimated activity / inactivity data points (in 30-minute increments) for the current day and the previous day are referenced chronologically starting from the beginning (24:00 on the previous day). If the inactive state continues for a predetermined time (for example, 1.5 hours in this embodiment), it is determined that the person is in a sleep state. Subsequently, if the person transitions to an active state and remains active for a predetermined short time (for example, 30 minutes in this embodiment), it is determined that the person is awake, and the time of the transition to the active state is taken as the wake-up time. If the active state does not continue for a predetermined short time (for example, 30 minutes in this embodiment), it is treated as a brief awakening, such as going to the toilet at night, and the person is not considered to have transitioned to an awake state but rather as still in a sleep state. This improves the accuracy of the wake-up time estimation.

[0059] Subsequently, the difference between the average wake-up time calculated in step S53 and the wake-up time for the day calculated in step S54 is calculated (S55), and it is determined whether the absolute value of the difference is greater than or equal to a predetermined set value (S56). This set value can be pre-set as a value estimated to be the difference between the subject's wake-up time for the day and the average wake-up time, but in this embodiment, for example, the default value is set to 5 hours.

[0060] Then, if the absolute value of the difference calculated in step S55 is not greater than or equal to a predetermined setting value (No in step S56), the alert processing related to lifestyle changes (sleep) is terminated. On the other hand, if the absolute value of the difference is greater than or equal to a predetermined setting value, that is, if the wake-up time on the day calculated in step S54 deviates from the average wake-up time calculated in step S53 by a predetermined setting value or more (Yes in step S56), a spreadsheet alert 18 related to lifestyle changes (sleep) is output (S57), and the alert processing is terminated.

[0061] <Conclusion> As described above, according to the monitoring system 1, which is one embodiment of the present invention, AI can estimate the presence / absence, sleep / non-sleep status, and air conditioner ON / OFF status based on electricity usage data measured by a smart meter. Based on these various estimation results, it is possible to detect abnormal values, lifestyle changes (sleep), and the risk of heatstroke at an early stage. Furthermore, by considering electricity usage shift as a feature when making estimations with AI, it is possible to recognize the electricity usage data as continuous time-series data and improve the accuracy of the estimation results.

[0062] Furthermore, by using spreadsheet software to detect abnormal values ​​in power consumption and lifestyle changes (sleep) in conjunction with AI-based estimation, it is sometimes possible to infer the reasons for the detection and avoid a black box in decision-making. In addition, by digitizing power consumption data using a baseline, it becomes possible to easily process it using spreadsheet software.

[0063] The present inventors have described the invention in detail based on embodiments above, but it goes without saying that the present invention is not limited to the above embodiments and can be modified in various ways without departing from its essence. Furthermore, the above embodiments are described in detail for the purpose of explaining the present invention in an easy-to-understand manner and are not necessarily limited to those having all the configurations described. In addition, it is possible to add, delete, or replace some of the configurations of the above embodiments with other configurations.

[0064] Furthermore, each of the above configurations, functions, processing units, and processing means may be implemented in hardware, in whole or in part, for example, by designing them as integrated circuits. Alternatively, each of the above configurations, functions, and means may be implemented in software by having the processor interpret and execute programs that implement each function. Information such as programs, tables, and files that implement each function can be stored in memory, hard disks, SSDs, or other recording devices, or in recording media such as IC cards, SD cards, or DVDs.

[0065] Furthermore, in the diagrams above, the control lines and information lines shown are those deemed necessary for explanation and do not necessarily represent all control lines and information lines that would be present in the actual implementation. In reality, it can be assumed that almost all components are interconnected. [Industrial applicability]

[0066] This invention can be used in monitoring systems that perform monitoring based on electricity usage. [Explanation of Symbols]

[0067] 1…Monitoring system, 2…Electricity usage data, 3…Other data 11...Data acquisition unit, 12...Power usage data database, 13...AI alert processing unit, 14...Learning model, 15...AI alert processing unit, 16...AI alert, 17...Spreadsheet alert processing unit, 18...Spreadsheet alert

Claims

1. A monitoring system that detects abnormalities in the living conditions of residents in a residence, A data acquisition unit acquires data on the electricity usage of the residence at predetermined intervals from an external source and records it in the electricity usage data recording unit, An AI prediction unit acquires electricity usage data from the electricity usage recording unit and estimates the living conditions of the residents per unit of time using a learning model previously generated by machine learning based on the data. Based on the estimation results from the AI ​​prediction unit, the AI ​​alert processing unit determines whether the resident's living situation is abnormal and outputs an alert if it determines that the situation is abnormal. The system includes a spreadsheet alert processing unit that acquires electricity usage data from the electricity usage recording unit, determines whether the resident's living situation is abnormal based on the data using spreadsheet software, and outputs an alert if it determines that the situation is abnormal. The aforementioned spreadsheet alert processing unit digitizes the electricity usage data of the resident per unit time over a predetermined period based on a predetermined reference value to estimate the active / inactive state, estimates the resident's daily wake-up time from the transition status of the active / inactive state, and outputs an alert for a lifestyle disturbance related to sleep if the absolute value of the difference between the average wake-up time or most frequent wake-up time over the predetermined period and the wake-up time on the current day is greater than or equal to a predetermined value.

2. In the monitoring system described in claim 1, The AI ​​prediction unit estimates the presence / absence, sleep / non-sleep, and ON / OFF status of the resident for each unit of time over a predetermined period, performs cluster analysis on the estimated data, and outputs an alert indicating that there are abnormal values ​​if any data is greater than a predetermined threshold from the normal value.

3. In the monitoring system described in claim 1, The AI ​​prediction unit estimates the sleep / non-sleep state of the resident for each unit of time over a predetermined period, performs cluster analysis on the estimated data, and outputs an alert for lifestyle disturbances related to sleep if there is data where the distance from the normal value is greater than a predetermined threshold, thus providing a monitoring system.

4. In the monitoring system described in claim 1, The AI ​​prediction unit estimates the presence / absence of the resident at each unit of time during a predetermined period, as well as the ON / OFF status of the air conditioner, and outputs an alert indicating a risk of heatstroke based on the estimated data and outside temperature data, thus forming a monitoring system.

5. In the monitoring system described in claim 1, The aforementioned spreadsheet alert processing unit is a monitoring system that outputs an alert indicating an abnormal value if the resident's daily electricity usage on a given day is less than the value obtained by multiplying the minimum daily electricity usage of the resident over a predetermined period by a predetermined coefficient.

6. In the monitoring system described in claim 1, The aforementioned spreadsheet alert processing unit is a monitoring system that digitizes the electricity usage data of the resident per unit time over a predetermined period based on predetermined reference values ​​to estimate the active / inactive state, and outputs an alert indicating that there is an abnormal value if all data is in the same state.

7. In the monitoring system described in claim 1, A monitoring system in which, as features used when generating the learning model used by the AI ​​prediction unit by machine learning, the transition status of the electricity usage data of the residence per unit time over a predetermined period of time is included.

8. A monitoring system for detecting abnormalities in the living conditions of residents in a residence, A data acquisition unit acquires data on the electricity usage of the residence at predetermined intervals from an external source and records it in the electricity usage data recording unit, An AI prediction unit acquires electricity usage data from the electricity usage recording unit and estimates the living conditions of the residents per unit of time using a learning model previously generated by machine learning based on the data. Based on the estimation results from the AI ​​prediction unit, the AI ​​alert processing unit determines whether the resident's living situation is abnormal and outputs an alert if it determines that the situation is abnormal. The system includes a spreadsheet alert processing unit that acquires electricity usage data from the electricity usage recording unit, determines whether the resident's living situation is abnormal based on the data using spreadsheet software, and outputs an alert if it determines that the situation is abnormal. The aforementioned spreadsheet alert processing unit is a monitoring system that outputs an alert indicating an abnormal value if the resident's daily electricity usage on a given day is less than the value obtained by multiplying the minimum daily electricity usage of the resident over a predetermined period by a predetermined coefficient.

9. A monitoring system for detecting abnormalities in the living conditions of residents in a residence, A data acquisition unit acquires data on the electricity usage of the residence at predetermined intervals from an external source and records it in the electricity usage data recording unit, An AI prediction unit acquires electricity usage data from the electricity usage recording unit and estimates the living conditions of the residents per unit of time using a learning model previously generated by machine learning based on the data. Based on the estimation results from the AI ​​prediction unit, the AI ​​alert processing unit determines whether the resident's living situation is abnormal and outputs an alert if it determines that the situation is abnormal. The system includes an alert processing unit that acquires electricity usage data from the electricity usage recording unit, determines whether the resident's living situation is abnormal based on the data using a rule-based system, and outputs an alert if it determines that the situation is abnormal. The alert processing unit digitizes the electricity usage data of the resident per unit time over a predetermined period based on a predetermined reference value to estimate the active / inactive state, estimates the resident's daily wake-up time from the transition status of the active / inactive state, and outputs an alert for a lifestyle disturbance related to sleep if the absolute value of the difference between the average wake-up time or most frequent wake-up time over the predetermined period and the wake-up time on the current day is greater than or equal to a predetermined value.

Images