JP7462259B2 - MEASUREMENT APPARATUS, SYSTEM, MEASUREMENT METHOD, AND PROGRAM - Google Patents

Description

This disclosure generally relates to a measurement device, a system, a measurement method, and a program. More specifically, this disclosure relates to a measurement device that measures resource usage, a system including the same, a measurement method that measures resource usage, and a program for executing the same.

Patent Document 1 discloses an energy management system that predicts the living activities (conditions) of residents (users) of a residence (facility). In Patent Document 1, the energy management system determines the status of users from the power usage status in the facility.

JP 2012-174030 A

The energy management system described in Patent Document 1 determines the user's condition by comparing the amount of electricity used, which is the usage status of power (resources), with a predetermined threshold value, which has the problem that there are limitations to the accuracy of the determination and the variety of items that can be determined.

The present disclosure has been made in consideration of the above points, and aims to provide a measurement device, system, measurement method, and program that can easily improve the accuracy of judgment and the variety of items that can be judged.

A measurement device according to an aspect of the present disclosure includes a measurement unit, a judgment unit, a re-learning unit, an acquisition unit, and an output unit. The measurement unit measures the usage status of a resource. The judgment unit judges the predetermined item based on the measurement result of the measurement unit using a trained model that has been machine-learned to output the predetermined item using the usage status of the resource and the correctness of the predetermined item as learning data. The re-learning unit re-learns the trained model using data in which the judgment result of the judgment unit is corrected according to the measurement result of the measurement unit and the judgment result of the judgment unit. The acquisition unit acquires a second trained model, which is different from a first trained model that is the trained model, from an external learning device. The output unit outputs the measurement result of the measurement unit to the learning device. The judgment unit adopts either the first trained model or the second trained model acquired by the acquisition unit. The output unit does not output time information related to the time measured by the measurement unit among the measurement results of the measurement unit to the learning device. The time information includes information that reminds the user of what time a specific action was performed. The time information is used for re-learning the first trained model in the re-learning unit.

A system according to an aspect of the present disclosure includes the above-described measurement device and a learning device, which generates the trained model by machine learning using the resource usage status and the accuracy of the predetermined items as the learning data, and is used by the determination unit of the measurement device.

A measurement method according to one aspect of the present disclosure is a measurement method executed by one or more processors. The measurement method includes a measurement step, a determination step, a re-learning step, an acquisition step, and an output step. The measurement step is a step of measuring the usage status of a resource. The determination step is a step of determining the predetermined item based on the measurement result of the measurement step using a trained model that has been machine-learned to output the predetermined item using the usage status of the resource and the correctness of the predetermined item as learning data. The re-learning step is a step of re-learning the trained model using data in which the judgment result of the judgment step is corrected according to the measurement result of the measurement step and the judgment result of the judgment step. The acquisition step acquires a second trained model, which is different from the first trained model that is the trained model, from an external learning device. The output step outputs the measurement result of the measurement step to the learning device. The determination step employs either the first trained model or the second trained model acquired in the acquisition step. In the output step, among the measurement results in the measurement step, time information related to the time measured in the measurement step is not output to the learning device. The time information includes information that reminds a user of what time a specific action was performed. The time information is used for re-learning the first trained model in the re-learning step.

A program according to one aspect of the present disclosure causes one or more processors to execute the above measurement method.

The present disclosure has the advantage of making it easier to improve the accuracy of assessment and the variety of items that can be assessed.

FIG. 1 is a block diagram showing the configuration of a system including a measurement device and a learning device according to an embodiment of the present disclosure. FIG. 2 is a schematic diagram showing the overall configuration including the above system. FIG. 3 is a flowchart showing the operation of the measuring device. FIG. 4 is a flowchart showing the operation of the learning device. FIG. 5 is a block diagram showing a configuration of a system including a measurement device and a learning device according to a modified example of an embodiment of the present disclosure.

(1) Overview A measurement device 1 (see FIG. 1) of this embodiment is a device for measuring resource usage. In this embodiment, the measurement device 1 measures resource usage in a target facility F1 (see FIG. 1).

In this disclosure, "resources" refers to resources used (consumed) in facility F1, such as electricity (electrical energy), gas, and tap water. The "resources" consumed in facility F1 include not only resources consumed in part of facility F1, such as electric locks, automatic doors, electric shutters, or escalators, but also resources consumed by equipment 3 (see FIG. 2) used in facility F1. When the resource is electricity, the resource also includes electricity supplied from the power grid and electricity output from distributed power sources (e.g., solar power generation equipment, power storage equipment, or wind power generation equipment) installed in facility F1.

In this disclosure, "facility" includes non-residential facilities such as stores, offices, factories, buildings, schools, welfare facilities, or hospitals, as well as residential facilities such as detached houses, apartment buildings, or individual units of detached houses or apartment buildings. Non-residential facilities also include theaters, movie theaters, public halls, amusement parks, complexes, department stores, hotels, inns, kindergartens, libraries, museums, art galleries, underground shopping malls, stations, and airports. Furthermore, in this disclosure, "facility" includes not only buildings (structures), but also outdoor facilities such as baseball stadiums, gardens, parking lots, grounds, and parks. In this embodiment, a detached house will be described as an example of facility F1.

In this disclosure, "equipment" refers to various equipment (including stationary or portable facilities, devices, systems, etc.) that is used in facility F1 and uses (consumes) resources. In this embodiment, equipment 3 is, as shown in FIG. 2, lighting equipment 31, air conditioning equipment 32, electrical outlets 33, toilets 34, showers 35, baths 36, and water heaters 37. In other words, equipment 3 is not limited to electrical equipment that consumes electricity, but may include equipment that consumes water or gas. Furthermore, equipment 3 is not limited to equipment used by users, but includes sensors that measure the environmental conditions of facility F1 (e.g., temperature sensors, humidity sensors, etc.), as well as wall switches, for example.

Electrical equipment may include facility equipment or home appliances. Examples of facility equipment include packaged air conditioners (air conditioning equipment), lighting equipment (including base lights and spot lights), power storage equipment, kitchen equipment (including induction heaters and dishwashers), entrance and exit management equipment, copy machines, and facsimiles. In addition, facility equipment also includes, for example, hot water supply equipment (including EcoCute (registered trademark)), electric shutters, ventilation fans, 24-hour ventilation systems, and other residential equipment. Examples of home appliances include television sets, lighting equipment (including ceiling lights), and recorders and players (including DVD recorders with HDDs and external HDDs). In addition, home appliances also include, for example, washing machines, refrigerators, air conditioners, air purifiers, personal computers, smart speakers, and computer game consoles.

The measuring device 1 is also a device for determining a predetermined item based on the measured resource usage. In this disclosure, the "predetermined item" is an item estimated from a feature included in the resource usage, and may include, for example, the type of equipment 3 used in the facility F1, or the state of the user (whether or not the user has performed a specific action). In other words, the measuring device 1 does not directly determine the predetermined item using a sensor or the like, but indirectly determines the predetermined item based on the measured resource usage. In this embodiment, since the facility F1 is a detached house, the user is a resident of the detached house.

As shown in FIG. 1, the measurement device 1 includes a measurement unit 11, a determination unit 13, and a relearning unit 15.

The measurement unit 11 measures the usage of resources. For example, if the resource is electricity, the measurement unit 11 measures the amount of power (electricity usage) in the entire facility F1, or the amount of power (electricity usage) of each device 3 used in the facility F1, etc.

The determination unit 13 uses the trained model M1 that has been machine-learned to output a predetermined item using the resource usage status as learning data to determine a predetermined item based on the measurement result of the measurement unit 11. The "trained model" in this disclosure is a program that, when the measurement result (resource usage status) of the measurement unit 11 is input, determines a predetermined item from the feature amount included in the input resource usage status and outputs the determination result, and is a model in which machine learning using the training data has been completed. Also, the "training data" in this disclosure is a data set that combines input information (in this embodiment, the resource usage status) input to the model and a label (in this embodiment, the correctness or incorrectness of the predetermined item) assigned to the input information, and is so-called teacher data. In other words, in this embodiment, the trained model M1 is a model in which machine learning by supervised learning has been completed. In other words, the determination unit 13 inputs the measurement result of the measurement unit 11 into the trained model M1 to determine a predetermined item from the feature amount included in the input resource usage status. In this embodiment, the judgment unit 13 makes judgments using a trained model M1 generated by an external learning device 2 separate from the measurement device 1, rather than a trained model M1 generated by the measurement device 1 itself.

The re-learning unit 15 re-learns the trained model M1 using the measurement results of the measurement unit 11. In other words, the re-learning unit 15 re-learns and updates the trained model M1 by machine learning using the measurement results of the measurement unit 11 (i.e., resource usage status) as learning data.

As described above, in this embodiment, the measurement device 1 uses the trained model M1 generated by machine learning to determine a specific item from the features included in the resource usage. Therefore, this embodiment has the advantage that it is easier to improve the accuracy of the determination and the variety of items that can be determined, compared to when the determination is made using a simple determination method such as comparing the resource usage amount with a threshold value.

In addition, in this embodiment, the measurement device 1 can re-learn the trained model M1 on the measurement device 1, which has the advantage that it is easy to construct a trained model M1 specific to the environment in which the measurement device 1 is installed (here, facility F1).

(2) Details A system 100 including a measuring device 1 and a learning device 2 according to this embodiment will be described in detail below with reference to Fig. 1 and Fig. 2. The learning device 2 uses the resource usage status as learning data to generate a trained model M1 used by the determination unit 13 of the measuring device 1 through machine learning. The trained model M1 generated by the learning device 2 is output to the measuring device 1.

In this disclosure, "generating a trained model" may include generating a trained model M1 when no trained model M1 previously existed, as well as generating (updating) a trained model M1 by retraining an existing trained model M1.

In this embodiment, as shown in FIG. 1, the learning device 2 is configured to be able to communicate individually with one or more (here, multiple) measuring devices 1 via a network N1 such as the Internet. The one or more measuring devices 1 are provided in one or more facilities F1 that are different from each other. The learning device 2 outputs a trained model M1 to each measuring device 1. In other words, the trained model M1 acquired by each measuring device 1 is the same. The following description focuses on one measuring device 1 out of the one or more measuring devices 1.

1, the measurement device 1 includes a measurement unit 11, a communication unit 12, a determination unit 13, a memory unit 14, and a relearning unit 15. In this embodiment, the measurement device 1 is configured as a device in which the measurement unit 11, the communication unit 12, the determination unit 13, the memory unit 14, and the relearning unit 15 are housed in a single housing.

In this embodiment, the measuring device 1 is at least partially configured with a microcontroller having one or more processors and memory. In other words, at least a part of the measuring device 1 is realized in a computer system having one or more processors and memory, and the computer system functions as at least a part of the measuring device 1 by the one or more processors executing a program stored in the memory. Here, the program is pre-recorded in the memory, but it may also be provided via a telecommunications line such as the Internet, or recorded on a non-transitory recording medium such as a memory card.

As shown in FIG. 2, the measuring device 1 is arranged in the cabinet of the distribution board 5. The distribution board 5 includes a main breaker 51 electrically connected to the system power supply 6 and a plurality of branch breakers 52 electrically connected to the secondary side of the main breaker 51 in the cabinet. The measuring device 1 is electrically connected to the current sensors 41 and 42. The current sensor 41 is provided on the primary side of the main breaker 51 and measures the current flowing through the main line. The plurality of current sensors 42 are provided corresponding to the plurality of branch breakers 52. The plurality of current sensors 42 measure the current flowing through the corresponding plurality of branch circuits. The "branch circuit" here means each circuit branched off from the main line by the plurality of branch breakers 52. The branch circuit includes wiring connected to the branch breaker 52, lighting fixtures 31, air conditioning equipment 32, outlets 33, wall switches, etc. In the facility F1, which is a detached house as in this embodiment, such a branch circuit is provided for each room, such as the living room, bedroom, entrance, toilet, children's room, or kitchen, and for each type of device 3, such as a lighting fixture 31 or an air conditioner 32. One branch circuit may include one device 3 or multiple devices 3.

The measurement unit 11 measures the resource usage. The measurement unit 11 is the entity that executes the measurement step ST1 (see FIG. 3) described below. In this embodiment, the measurement unit 11 measures the resource usage, including the electricity usage, the tap water usage, and the gas usage. The measurement results of the measurement unit 11 are stored in the memory unit 14.

The measurement unit 11 uses the output of the current sensors 41, 42 to measure at least one of the instantaneous power and the amount of power (electricity usage) for the main line and each of the multiple branch circuits. The measurement unit 11 also uses the output of the current sensors 41, 42 to measure the characteristics of the current flowing through the main line and each of the multiple branch circuits. Examples of the characteristics of the current include the frequency or phase of the current. The "frequency of the current" here may include harmonic components of the current. In this way, the measurement unit 11 measures the electricity usage.

The measurement unit 11 also measures the amount of tap water used in the facility F1 using the output of the water meter 43 electrically connected to the measurement device 1. Specifically, the measurement unit 11 measures the amount of tap water used in the facility F1 by counting the pulses emitted by a pulse transmitter in the water meter 43. In this embodiment, the measurement unit 11 measures the amount of tap water used in the toilet 34, shower 35, bath 36, washroom, etc. In this way, the measurement unit 11 measures the tap water usage.

The measurement unit 11 also measures the amount of gas used in the facility F1 using the output of the gas meter 44 that is electrically connected to the measurement device 1. Specifically, the measurement unit 11 measures the amount of gas used in the facility F1 by counting the pulses emitted by a pulse generator in the gas meter 44. In this embodiment, the measurement unit 11 measures the amount of gas used in the water heater 37, etc. In this way, the measurement unit 11 measures the gas usage status.

The communication unit 12 transmits and receives signals to and from the external learning device 2 directly or indirectly via a network or a repeater (router, etc.) by an appropriate communication method such as wired communication or wireless communication. In this embodiment, the communication unit 12 is configured to communicate with the learning device 2 via a network N1 such as the Internet. The communication unit 12 has a function as an acquisition unit 121 and a function as an output unit 122.

The acquisition unit 121 acquires the trained model M1 by receiving a signal including the trained model M1 from the learning device 2. The acquisition unit 121 is the entity that executes the acquisition step ST2 (see FIG. 3 ) described below. If the acquisition unit 121 has not acquired the trained model M1 in the past, it acquires the trained model M1 itself from the learning device 2.

On the other hand, if the acquisition unit 121 has previously acquired the trained model M1 and the learning device 2 has re-learned the trained model M1, the learning device 2 outputs the parameters of the trained model M1 after re-learning to the measurement device 1. In this case, the acquisition unit 121 does not acquire the trained model M1 itself, but only acquires the parameters (e.g., weighting coefficients) of the trained model M1 after re-learning. Then, the determination unit 13 can use the parameters acquired by the acquisition unit 121 to update the existing trained model M1 to the trained model M1 after re-learning. In this aspect, the amount of communication can be reduced compared to the case of transmitting the trained model M1 itself.

In other words, the acquisition unit 121 acquires a second trained model M12 (see FIG. 5), which is the same as or different from the first trained model M11 (see FIG. 5), which is the trained model M1, from the external learning device 2. Here, the first trained model M11 is an existing trained model M1 possessed by the measurement device 1. Also, the second trained model M12 is a trained model M1 acquired from the learning device 2. As described later, the measurement device 1 and the learning device 2 each independently re-learn the trained model M1, so that the first trained model M11 and the second trained model M12 are basically different from each other. Of course, the first trained model M11 and the second trained model M12 may be the same.

The output unit 122 outputs the measurement results of the measurement unit 11 to the learning device 2 via the network N1. The output unit 122 is the entity that executes the output step ST4 (see FIG. 3) described below. In this embodiment, the measurement results of the measurement unit 11 that are output to the learning device 2 may include the measurement results of electricity usage, the measurement results of tap water usage, and the measurement results of gas usage.

In this embodiment, the output unit 122 does not output the measurement results of the measurement unit 11 directly to the learning device 2, but also outputs the judgment results of the judgment unit 13, which have received the measurement results of the measurement unit 11 as input, to the learning device 2. In other words, the output unit 122 outputs learning data including the measurement results of the measurement unit 11 and the judgment results of the judgment unit 13, which serve as labels, to the learning device 2.

Here, in this embodiment, the output unit 122 does not output to the learning device 2, among the measurement results of the measurement unit 11, time information (i.e., time stamp) related to the time measured by the measurement unit 11. For example, when the output unit 122 outputs the measurement results of the usage status of electricity as a resource to the learning device 2, it outputs the measured instantaneous power and power amount data to the learning device 2, but does not output the time stamp to the learning device 2. Here, the time information can be private information that evokes the user's daily life, such as the time at which the user performed a specific action. In this embodiment, by outputting the measurement results of the measurement unit 11, excluding the time information, to the learning device 2, it is possible to avoid exposing the user's private information to the public.

The determination unit 13 determines the predetermined items using the trained model M1. The determination unit 13 is the entity that executes the determination step ST3 (see FIG. 3) described below. In this embodiment, the predetermined items determined by the determination unit 13 (i.e., the trained model M1) include at least one of device information related to the device 3 that consumes resources and user information related to the user.

The device information may include, for example, the type of device 3 or the operating status of the device 3. For example, the determination unit 13 determines the type of one or more devices 3 used in the facility F1 based on the measurement results of the measurement unit 11. The "type" here may include the type (product number) of the device 3, etc. Also, for example, the determination unit 13 determines whether the device 3 is operating or not, and if it is operating, the operating mode or operating time of the device 3, etc., based on the measurement results of the measurement unit 11.

As already mentioned, in this embodiment, since the facility F1 is a detached house, the user is a resident of the detached house. Therefore, the user information may include, for example, whether or not the user is at home, whether or not the user is watching television, whether or not the user is cooking, whether or not the user is taking a bath, whether or not the user is using the toilet, or whether or not the user is sleeping. In this way, the user information may include whether or not the user is performing a particular action.

In this embodiment, the determination unit 13 adopts either the first trained model M11 or the second trained model M12 acquired by the acquisition unit 121. For example, the determination unit 13 basically adopts the first trained model M11, but when the trained model M1 is significantly updated in the learning device 2, it may adopt the second trained model M12 acquired from the learning device 2. The decision as to whether to adopt the first trained model M11 or the second trained model M12 may be determined according to a preset algorithm or according to the user's judgment. In the latter case, the user's judgment may be sought via the information terminal 7 described below.

The storage unit 14 includes, for example, an electrically rewritable non-volatile memory such as an EEPROM (Electrically Erasable Programmable Read-Only Memory) or a volatile memory such as a RAM (Random Access Memory). The storage unit 14 stores the measurement results of the measurement unit 11 and the judgment results of the judgment unit 13. In other words, the storage unit 14 stores learning data used for relearning in the relearning unit 15 (or learning data output to the learning device 2). The learning data may be erased after being used for relearning in the relearning unit 15 (or output to the learning device 2).

The relearning unit 15 uses the measurement results (i.e., resource usage status) of the measurement unit 11 as learning data to relearn and update the trained model M1 through machine learning. The relearning unit 15 is the entity that executes the relearning step ST5 described below. In this embodiment, the trained model M1 is a model that uses a neural network, as described below. Therefore, the relearning unit 15 updates the parameters (e.g., weighting coefficients) of the neural network by relearning the trained model M1.

In this embodiment, the above-mentioned time information (time stamp) is used by the re-learning unit 15 to re-learn the trained model M1. In other words, since the above-mentioned time information may be private information reminiscent of the user's daily life, the output unit 122 removes the time information from the measurement results of the measurement unit 11 and outputs it to the learning device 2 to prevent it from being publicly exposed. On the other hand, by including the above-mentioned time information in the learning data, the re-learning unit 15 is able to re-learn so as to reflect the user's private information in the trained model M1, and as a result, it is easier to optimize the trained model M1 to suit the user.

In addition, in this embodiment, the re-learning unit 15 does not use all of the measurement results of the measurement unit 11, but rather uses data that has been thinned out from the measurement results of the measurement unit 11 according to thinning conditions to re-learn the trained model M1. Examples of thinning conditions are listed below. The re-learning unit 15 may thin out data according to all of the thinning conditions listed below, or may thin out data according to only some of the thinning conditions. In addition, it is preferable for the re-learning unit 15 to thin out data so that the number of data for each condition is uniform, so that the number of data that meets a specific condition does not become significantly higher or lower.

The thinning conditions may include, for example, conditions related to input from the user. For example, the re-learning unit 15 thins out data from the measurement results of the measurement unit 11 that is prohibited from being used as learning data due to input from the user, and re-learns the trained model M1. In this embodiment, the measurement device 1 accepts input from the user by receiving information input to the information terminal 7 (see FIG. 2) by user operation via the communication unit 12 via the network N1. The information terminal 7 is a terminal owned or used by the user, such as a smartphone, tablet terminal, or personal computer. In addition, the measurement device 1 may accept input from the user via various display devices and input devices, such as a built-in touch panel display.

In addition, the thinning conditions may include, as an example, conditions regarding the correctness of the judgment result of the judgment unit 13. For example, the re-learning unit 15 thins out data for which the judgment result of the judgment unit 13 is correct from among the measurement results of the measurement unit 11, and re-learns the learned model M1. In other words, the re-learning unit 15 re-learns the learned model M1 using data for which the judgment result of the judgment unit 13 is incorrect from among the measurement results of the measurement unit 11. More specifically, when the predetermined item is whether or not the lighting device 31 is turned on, it is assumed that the lighting device 31 is not actually turned on, but the judgment unit 13 judges that the lighting device 31 is turned on. In this case, the re-learning unit 15 re-learns the learned model M1 using data obtained by correcting the judgment result of the judgment unit 13 according to the judgment of the correctness of the judgment result of the judgment unit 13 as learning data. In this aspect, since re-learning can be performed to correct the judgment result of the learned model M1, improvement in the judgment accuracy of the learned model M1 can be expected.

The judgment of whether the judgment result of the judgment unit 13 is correct or not can be realized, for example, by presenting the judgment result of the judgment unit 13 to a user and having the user evaluate the judgment result of the judgment unit 13. Specifically, the measurement device 1 outputs the judgment result of the judgment unit 13 to the information terminal 7 via the network N1. The user checks the judgment result of the judgment unit 13 by looking at a display attached to the information terminal 7 and inputs an evaluation of whether the judgment result is correct or not. The information terminal 7 then sends a response signal including the input evaluation back to the measurement device 1 via the network N1.

The thinning conditions may include, for example, conditions related to time. For example, the re-learning unit 15 thins out data measured during a specific time period set in advance from among the measurement results of the measurement unit 11, and re-learns the trained model M1. The specific time period can be set, for example, by the user inputting it on the information terminal 7.

Here, when the thinning conditions include a time-related condition, the time-related condition may be set uniformly regardless of the type of device 3 and the day of the week, but is not limited to this. In other words, multiple time-related conditions may be set. For example, the time-related condition may be set to a different time period for each day of the week. Specifically, it may be set so that data from the daytime hours is thinned out on weekdays from Monday to Friday, and so that data from the early morning hours is thinned out on holidays such as Saturday and Sunday. Also, for example, the time-related condition may be set to a different time period for each type of device 3. Specifically, the time periods for thinning out data for lighting fixtures 31 and bathroom 36 may be set to be different from each other.

In this embodiment, the re-learning unit 15 re-learns the trained model M1 during a specific period that is specified in advance. The specific period is, for example, a time period during which the processing load of the measuring device 1 is relatively low, such as late at night. In this embodiment, the trained model M1 can be re-learned during a period when the measuring device 1 has spare processing capacity, making it possible to use a device with a relatively low processing capacity as the measuring device 1.

(2.2) Learning Device As shown in FIG. 1, the learning device 2 includes a communication unit 21, a model generation unit 22, and a storage unit 23.

In this embodiment, at least a portion of the learning device 2 is configured with a microcontroller having one or more processors and memory. In other words, at least a portion of the learning device 2 is realized in a computer system having one or more processors and memory, and the computer system functions as at least a portion of the learning device 2 by the one or more processors executing a program stored in the memory. Here, the program is pre-recorded in the memory, but it may also be provided via a telecommunications line such as the Internet, or recorded on a non-transitory recording medium such as a memory card.

The communication unit 21 transmits and receives signals to and from the measurement device 1 directly or indirectly via a network or a repeater (router, etc.) by an appropriate communication method such as wired communication or wireless communication. In this embodiment, the communication unit 21 is configured to communicate with the measurement device 1 via a network N1 such as the Internet. The communication unit 21 outputs (transmits) the trained model M1 generated by the model generation unit 22 to the measurement device 1, and acquires (receives) the measurement results (training data) of the measurement unit 11 output from the measurement device 1.

The model generation unit 22 generates the trained model M1 used in the measurement device 1 by machine learning using the resource usage status as learning data. Specifically, the model generation unit 22 generates a linear model such as an SVM (Support Vector Machine), a model using a neural network, or a model by deep learning using a multi-layer neural network as the trained model M1. In this embodiment, the model generation unit 22 generates a model using a neural network as the trained model M1. The neural network may include, for example, a CNN (Convolutional Neural Network) or a BNN (Bayesian Neural Network). In this embodiment, the model generation unit 22 generates the trained model M1 by supervised learning, as already described.

When no previously trained model M1 exists, the model generation unit 22 uses a separately prepared resource usage status as learning data to generate a trained model M1 through machine learning, rather than using the measurement results of the measurement unit 11 acquired from the measurement device 1 as learning data. Also, when a trained model M1 already exists, the model generation unit 22 re-trains and updates the trained model M1 through machine learning, using the measurement results of the measurement unit 11 acquired from the measurement device 1 (i.e., the resource usage status) as learning data.

In this embodiment, as shown in FIG. 1, the learning device 2 acquires the measurement results of the measurement unit 11 from each of the multiple measurement devices 1. Therefore, in this embodiment, the learning device 2 can acquire a large amount of diverse learning data compared to the case where the measurement results of the measurement unit 11 are acquired from only one measurement device 1, and as a result, it is expected that the judgment accuracy of the trained model M1 after re-learning will be improved.

The memory unit 23 includes, for example, an electrically rewritable non-volatile memory such as an EEPROM, or a volatile memory such as a RAM. The memory unit 23 stores the measurement results of the measurement unit 11 obtained from the measurement device 1. In other words, the memory unit 23 stores the learning data obtained from the measurement device 1. The learning data may be erased after re-learning the trained model M1.

(3) Operation An example of the operation of the measurement device 1 will be described below with reference to Fig. 3, and an example of the operation of the learning device 2 will be described with reference to Fig. 4. In the example of the operation of the measurement device 1, it is assumed that the measurement device 1 has not yet acquired the trained model M1 from the learning device 2. In addition, in the example of the operation of the learning device 2, it is assumed that the learning device 2 has not yet generated the trained model M1.

(3.1) Measuring Device In the measuring device 1, first, the acquiring unit 121 acquires the trained model M1 output from the learning device 2 (S1). After this, the determining unit 13 becomes able to determine a predetermined item using the trained model M1 acquired from the learning device 2. Process S1 corresponds to the acquiring step ST2.

After acquiring the trained model M1, the measurement unit 11 periodically measures resource usage (S2). The judgment unit 13 then inputs the measurement results of the measurement unit 11 into the trained model M1 and judges predetermined items based on the measurement results of the measurement unit 11 (S3). Process S2 corresponds to measurement step ST1. Process S3 corresponds to judgment step ST3. The measurement results of the measurement unit 11 and the judgment results of the judgment unit 13 linked to these are stored in the memory unit 14 as learning data (S4). Processes S2 to S4 are repeated until a predetermined number of learning data are accumulated in the memory unit 14 (S5: No).

When a predetermined number of learning data are accumulated in the memory unit 14 (S5: Yes), the output unit 122 outputs the accumulated learning data (measurement results of the measurement unit 11) to the learning device 2 (S6). The re-learning unit 15 uses the accumulated learning data to re-learn the trained model M1 (S7). The learning data accumulated in the memory unit 14 is then erased. In other words, from process S7 onwards, new learning data is accumulated in the memory unit 14. Process S6 corresponds to output step ST4. Process S7 corresponds to re-learning step ST5.

After that, the learning device 2 re-learns the trained model M1, and acquires the parameters of the trained model M1 after re-learning (i.e., the second trained model M12) from the learning device 2 (S8: Yes). Process S8: Yes corresponds to acquisition step ST2. Then, the determination unit 13 determines whether to adopt the existing trained model M1 (i.e., the first trained model M11) or the second trained model M12 according to a predefined algorithm. When the first trained model M11 is adopted (S9: Yes), the determination unit 13 does not execute anything in particular. On the other hand, when the second trained model M12 is adopted (S9: No), the determination unit 13 uses the parameters acquired from the learning device 2 to update the trained model M1 (the first trained model M11) to the second trained model M12 (S10). Note that until the parameters of the second trained model M12 are acquired from the learning device 2, the determination unit 13 judges the predetermined items using the existing trained model M1 (S8: No).

(3.2) Learning Device In the learning device 2, first, the model generation unit 22 generates a trained model M1 by using training data prepared in advance (S11). Then, the learning device 2 outputs the trained model M1 generated by the model generation unit 22 to the measurement device 1 (S12).

After the trained model M1 is generated, the communication unit 21 periodically or irregularly acquires the measurement results (i.e., training data) of the measurement unit 11 output from the measurement device 1 (S13). The training data acquired by the communication unit 21 is stored in the memory unit 23 (S14). Processes S13 and S14 are repeated until a predetermined number of training data are accumulated in the memory unit 23 (S15: No).

When a predetermined number of learning data have been accumulated in the memory unit 23 (S15: Yes), the learning device 2 uses the accumulated learning data to re-learn the trained model M1 (S16). The learning device 2 then outputs the parameters of the trained model M1 after re-learning to the measurement device 1 (S17). Thereafter, steps S13 to S17 are repeated. Here, the learning data accumulated in the memory unit 23 is erased. In other words, from step S17 onwards, new learning data is accumulated in the memory unit 23.

As described above, in this embodiment, the measurement device 1 uses the trained model M1 generated by machine learning to determine a specific item from features included in the resource usage. In other words, in this embodiment, the determination unit 13 uses the trained model M1 to determine a specific item using latent features that are difficult for humans to notice, rather than determining the specific item using a threshold value defined by humans. Therefore, this embodiment has the advantage that it is easier to improve the determination accuracy and the variety of items that can be determined, compared to when a determination is made using a simple determination method such as comparing the resource usage amount with a threshold value.

In addition, in this embodiment, the measurement device 1 can re-learn the trained model M1 on the measurement device 1, which has the advantage that it is easy to construct a trained model M1 specific to the environment in which the measurement device 1 is installed (facility F1 in this case). In other words, the trained model M1 generated (updated) by the learning device 2 is a generic model that does not take into account the environment in which an arbitrary measurement device 1 is installed. For this reason, if the trained model M1 acquired from the learning device 2 is used, it may not be possible to make a judgment that corresponds to the environment in which the measurement device 1 is installed.

On the other hand, in this embodiment, the measurement device 1 can re-learn the trained model M1 using the measurement results of the measurement unit 11, that is, information specific to the environment in which the measurement device 1 is installed, as learning data. Therefore, in this embodiment, it is expected that it will be possible to make judgments according to the environment in which the measurement device 1 is installed.

(4) Modifications The above-described embodiment is merely one of various embodiments of the present disclosure. The above-described embodiment can be modified in various ways depending on the design, etc., as long as the object of the present disclosure can be achieved. In addition, functions similar to those of the measurement device 1 may be embodied in a (computer) program, a non-transitory recording medium on which a program is recorded, or the like, in addition to a measurement method.

The measurement method according to one embodiment includes a measurement step ST1, a determination step ST3, and a re-learning step ST5. The measurement step ST1 is a step of measuring resource usage. The determination step ST3 is a step of determining a predetermined item based on the measurement result of the measurement step ST1 using a trained model M1 that has been machine-learned to output a predetermined item as learning data for the resource usage. The re-learning step ST5 is a step of re-learning the trained model M1 using the measurement result of the measurement step ST1. The (computer) program according to one embodiment causes one or more processors to execute the above measurement method.

Below, we list some variations of the above-mentioned embodiment. The variations described below can be applied in appropriate combinations.

Each of the measuring devices 1 in the present disclosure includes a computer system. The computer system is mainly composed of a processor and a memory as hardware. The function of the measuring device 1 in the present disclosure is realized by the processor executing a program recorded in the memory of the computer system. The program may be pre-recorded in the memory of the computer system, provided through an electric communication line, or recorded and provided in a non-transitory recording medium such as a memory card, an optical disk, or a hard disk drive that can be read by the computer system. The processor of the computer system is composed of one or more electronic circuits including a semiconductor integrated circuit (IC) or a large-scale integrated circuit (LSI). The integrated circuits such as IC or LSI referred to here are called differently depending on the degree of integration, and include integrated circuits called system LSI, VLSI (Very Large Scale Integration), or ULSI (Ultra Large Scale Integration). Furthermore, a field-programmable gate array (FPGA) that is programmed after the manufacture of the LSI, or a logic device that can reconfigure the connection relationship inside the LSI or reconfigure the circuit partition inside the LSI, can also be used as a processor. The multiple electronic circuits may be integrated into one chip, or may be distributed across multiple chips. The multiple chips may be integrated into one device, or may be distributed across multiple devices. The computer system referred to here includes a microcontroller having one or more processors and one or more memories. Therefore, the microcontroller is also composed of one or more electronic circuits including a semiconductor integrated circuit or a large-scale integrated circuit.

In the above embodiment, the determination unit 13 employs either the first trained model M11 or the second trained model M12, but this is not limited to the above embodiment. For example, the determination unit 13 may employ both the first trained model M11 and the second trained model M12.

That is, as shown in FIG. 5, the device may include a first judgment unit 131 and a second judgment unit 132. The first judgment unit 131 is the same as the judgment unit 13 in the above-mentioned embodiment, and judges a predetermined item based on the measurement results of the measurement unit 11 using a first trained model M11. The second judgment unit 132 is a judgment unit separate from the first judgment unit 131, which is the judgment unit 13, and judges a predetermined item based on the measurement results of the measurement unit 11 using a second trained model M12. And the second trained model M12 is a model separate from the first trained model M11.

In this aspect, the measurement device 1 can determine a specific item using the latest trained model (second trained model M12) acquired from the learning device 2. Also, in this aspect, the measurement device 1 has the first trained model M11 that is included in the measurement device 1 itself. Therefore, in this aspect, even if the determination accuracy of the second trained model M12 is deteriorated compared to the first trained model M11, it is possible to determine a specific item using the first trained model M11.

In the above embodiment, when the measurement device 1 does not previously have a trained model M1, the measurement device 1 makes a judgment using the trained model M1 acquired from the external learning device 2, but this is not limited to the above embodiment. For example, depending on the processing capacity, the measurement device 1 may generate the trained model M1 by itself using the measurement results (i.e., resource usage status) of the measurement unit 11 as learning data, and make a judgment using the generated trained model M1.

In the above-described embodiment, the re-learning unit 15 may re-learn the trained model M1 regardless of a specific period. For example, the re-learning unit 15 may start re-learning the trained model M1 from the point in time when a predetermined number of pieces of training data have been accumulated in the memory unit 14.

In the above embodiment, the measurement unit 11 measures electricity, water, and gas as resources, but this is not limited to the above. For example, the measurement unit 11 may measure at least one of the resources of electricity, water, and gas.

In the above-described embodiment, the accuracy of the judgment result of the judgment unit 13 may be evaluated not only by the user's input, but also, for example, by another system different from the measurement device 1.

In the above-described embodiment, the trained model M1 may be generated not only by supervised learning, but also by unsupervised learning. Alternatively, the trained model M1 may be generated by reinforcement learning.

In the above embodiment, the output unit 122 outputs data from the measurement results of the measurement unit 11 that has been thinned out according to the thinning conditions to the learning device 2, but it may also output all of the measurement results of the measurement unit 11 to the learning device 2.

In the above-described embodiment, the measuring device 1 does not have to be housed inside the distribution board 5. For example, the measuring device 1 may be installed by being attached to an inner wall or an outer wall of the facility F1 outside the distribution board 5. In other words, the measuring device 1 may have a structure that allows it to be attached to a wall, separate from the distribution board 5. As an example, the measuring device 1 is attached to an inner wall or an outer wall of the facility F1 by screwing the housing to the inner wall or the outer wall of the facility F1.

In the above-described embodiment, the measurement unit 11 may measure the value of the voltage applied to the device 3 or a characteristic quantity of the voltage as the usage status of electricity as a resource. An example of a characteristic quantity of the voltage is the frequency or phase of the voltage. The "frequency of the voltage" here may include harmonic components of the voltage.

In the above-described embodiment, when re-learning of the trained model M1 is performed in the learning device 2, the acquisition unit 121 may acquire the trained model M1 itself after re-learning, rather than acquiring only the parameters of the trained model M1 after re-learning.

(summary)
As described above, the measurement device (1) according to the first aspect includes a measurement unit (11), a judgment unit (13), and a re-learning unit (15). The measurement unit (11) measures resource usage. The judgment unit (13) judges a predetermined item based on the measurement result of the measurement unit (11) using a trained model (M1) that has been machine-learned to output a predetermined item from the resource usage as learning data. The re-learning unit (15) re-learns the trained model (M1) using the measurement result of the measurement unit (11).

This aspect has the advantage that it is easy to improve the accuracy of judgment and the variety of items that can be judged. In addition, this aspect has the advantage that the measurement device (1) can easily build a trained model (M1) that is specific to the environment in which the measurement device (1) is installed.

The measurement device (1) according to the second aspect is the first aspect, and further includes an acquisition unit (121). The acquisition unit (121) acquires a second trained model (M12) that is the same as or different from the first trained model (M11), which is the trained model (M1), from an external learning device (2). The determination unit (13) adopts either the first trained model (M11) or the second trained model (M12) acquired by the acquisition unit (121).

This aspect has the advantage of improving the freedom of selection of the trained model (M1).

The measurement device (1) according to the third aspect further includes a second judgment unit (132) in the second aspect. The second judgment unit (132) is a judgment unit separate from the first judgment unit (131) which is the judgment unit (13), and judges a predetermined item based on the measurement results of the measurement unit (11) using a second trained model (M12). The second trained model (M12) is a model separate from the first trained model (M11).

This aspect has the advantage that the latest trained model (second trained model (M12)) obtained from the learning device (2) can be used.

In the measurement device (1) according to the fourth aspect, in any of the first to third aspects, the re-learning unit (15) re-learns the trained model (M1) using data that has been thinned out according to the thinning conditions from among the measurement results of the measurement unit (11).

According to this embodiment, data not necessary for re-learning can be thinned out, which has the advantage that it is easier to optimize the trained model (M1) compared to using all of the measurement results of the measurement unit (11) for re-learning.

In the measurement device (1) according to the fifth aspect, in the fourth aspect, the thinning conditions include conditions related to input from the user.

According to this aspect, the re-learning can be limited to the data of the measurement results of the measurement unit (11) that the user wants to reflect in the trained model (M1), which has the advantage of making it easier to optimize the trained model (M1).

In the measuring device (1) according to the sixth aspect, in the fourth or fifth aspect, the thinning conditions include conditions regarding the correctness of the judgment result of the judgment unit (13).

According to this aspect, re-learning can be limited to data that was actually correct or incorrectly determined by the determination unit (13) from among the measurement results of the measurement unit (11), which has the advantage of making it easier to optimize the trained model (M1).

The measurement device (1) according to the seventh aspect, in any one of the first to sixth aspects, further includes an output unit (122) that outputs the measurement results of the measurement unit (11) to an external learning device (2).

This aspect has the advantage that the measurement results of the measurement unit (11) can be used to re-learn the trained model (M1) in the learning device (2).

In the measurement device (1) according to the eighth aspect, in the seventh aspect, the output unit (122) does not output to the learning device (2) time information related to the time measured by the measurement unit (11) among the measurement results of the measurement unit (11).

This has the advantage that private information that may be reminiscent of a user's actions does not need to be publicly exposed.

In the measurement device (1) according to the ninth aspect, in the eighth aspect, the time information is used for re-learning the trained model (M1) in the re-learning unit (15).

This aspect has the advantage that it is easier to optimize the trained model (M1) to suit the user.

In the measurement device (1) according to the tenth aspect, in any of the first to ninth aspects, the re-learning unit (15) re-learns the trained model (M1) for a specific period of time that is specified in advance.

This aspect has the advantage that the trained model (M1) can be retrained during periods when there is spare processing load, making it possible to employ a device with relatively low processing power.

In the measurement device (1) according to the eleventh aspect, in any of the first to tenth aspects, the predetermined items include at least one of device information about the device (3) that consumes resources and user information about the user.

This aspect has the advantage that it becomes easier to grasp at least one of the state of the device (3) and the state of the user.

A system (100) according to a twelfth aspect includes a measuring device (1) according to any one of the first to eleventh aspects and a learning device (2). The learning device (2) uses the resource usage status as learning data to generate a trained model (M1) by machine learning, which is used in a determination unit (13) of the measuring device (1) according to any one of the first to eleventh aspects.

This aspect has the advantage that it is easy to improve the accuracy of judgment and the variety of items that can be judged. In addition, this aspect has the advantage that the measurement device (1) can easily build a trained model (M1) that is specific to the environment in which the measurement device (1) is installed.

The measurement method according to the thirteenth aspect includes a measurement step (ST1), a determination step (ST3), and a re-learning step (ST5). The measurement step (ST1) is a step of measuring resource usage. The determination step (ST3) is a step of determining a predetermined item based on the measurement result of the measurement step (ST1) using a trained model (M1) that has been machine-learned to output a predetermined item as learning data for the resource usage. The re-learning step (ST5) is a step of re-learning the trained model (M1) using the measurement result of the measurement step (ST1).

This aspect has the advantage that it is easy to improve the accuracy of judgment and the variety of items that can be judged. In addition, this aspect has the advantage that it is easy for the entity executing the measurement method to construct a trained model (M1) that is specific to the environment in which the entity executing the measurement method is installed.

The program according to the fourteenth aspect causes one or more processors to execute the measurement method according to the thirteenth aspect.

This aspect has the advantage that it is easy to improve the accuracy of judgment and the variety of items that can be judged. In addition, this aspect has the advantage that it is easy for the entity executing the measurement method to construct a trained model (M1) that is specific to the environment in which the entity executing the measurement method is installed.

The configurations according to the second to eleventh aspects are not essential to the measuring device (1) and may be omitted as appropriate.

REFERENCE SIGNS LIST 100 System 1 Measurement device 11 Measurement unit 121 Acquisition unit 122 Output unit 13 Judgment unit 131 First judgment unit 132 Second judgment unit 15 Re-learning unit 2 Learning device 3 Equipment M1 Trained model M11 First trained model M12 Second trained model ST1 Measurement step ST3 Judgment step ST5 Re-learning step

Claims (10)

A measurement unit that measures resource usage;
a determination unit that determines the predetermined item based on the measurement result of the measurement unit by using a trained model that has been machine-learned to output the predetermined item using the resource usage status and the accuracy of the predetermined item as learning data;
a re-learning unit that re-learns the trained model using data obtained by correcting the judgment result of the judgment unit in accordance with a judgment of the accuracy of the measurement result of the measurement unit and the accuracy of the judgment result of the judgment unit;
An acquisition unit that acquires a second trained model different from the first trained model, which is the trained model, from an external learning device;
an output unit that outputs a measurement result of the measurement unit to the learning device,
The determination unit adopts either the first trained model or the second trained model acquired by the acquisition unit,
the output unit does not output, to the learning device, time information related to the time measured by the measurement unit, among the measurement results of the measurement unit;
The time information is
Contains information that recalls when a user performed a particular action,
Used for relearning the first trained model in the relearning unit,
Measuring equipment. The apparatus further includes a second determination unit, which is a determination unit different from the first determination unit, and which determines the predetermined item based on the measurement result of the measurement unit by using the second trained model.
The measurement device according to claim 1. The re-learning unit re-learns the trained model using data culled according to a culling condition from among the measurement results of the measurement unit.
The measuring device according to claim 1 or 2. the thinning condition includes a user condition regarding an input from a user,
the user condition includes a condition that data prohibited from being used as the learning data by an input from the user is thinned out.
The measuring device according to claim 3. the thinning condition includes a correctness condition regarding correctness of the determination result of the determination unit,
the correctness condition includes a condition that data determined by the determination unit as a correct result is thinned out.
The measuring device according to claim 3 or 4. The re-learning unit re-learns the trained model for a predetermined specific period.
The measuring device according to any one of claims 1 to 5. the predetermined items include at least one of device information related to a device consuming the resource and user information related to a user,
The user information includes whether the user is performing a specific action.
The measuring device according to any one of claims 1 to 6. A measuring device according to any one of claims 1 to 7,
A learning device that generates the trained model used in the determination unit of the measurement device according to any one of claims 1 to 7 by machine learning using the resource usage status and the correctness or incorrectness of the predetermined items as the learning data.
system. 1. A method of measurement executed by one or more processors, comprising:
a measurement step of measuring resource usage;
a determination step of determining the predetermined item based on the measurement result of the measurement step by using a trained model that has been machine-learned to output the predetermined item using the resource usage status and the correctness or incorrectness of the predetermined item as learning data;
a re-learning step of re-learning the trained model using data obtained by correcting the judgment result of the judgment step in accordance with a judgment of the correctness of the measurement result of the measurement step and the judgment result of the judgment step ;
An acquisition step of acquiring a second trained model different from the first trained model, which is the trained model, from an external learning device;
an output step of outputting a measurement result of the measurement step to the learning device,
In the determination step, either the first trained model or the second trained model acquired in the acquisition step is adopted;
In the output step, time information related to the time measured in the measurement step is not output to the learning device among the measurement results of the measurement step,
The time information is
Contains information that recalls when a user performed a particular action,
Used for re-learning the first trained model in the re-learning step.
Measurement method. The one or more processors,
10. A measuring method according to claim 9,
program.

Images