JP7331041B2 - Action prediction support device, action prediction support method, and action prediction system - Google Patents

Description

The present invention relates to a behavior prediction support device, a behavior prediction support method, and a behavior prediction system.

In recent years, IoT (Internet of Things) devices are becoming popular. In addition, with the spread of IoT devices, techniques have been proposed for performing machine learning using sensor information obtained from these IoT devices and predicting human behavior. For example, Patent Literature 1 discloses a system that provides services using sensor information collected from IoT devices.

JP 2017-11525 A

By the way, the conventional method of performing behavior prediction by performing machine learning using information acquired from sensors such as IoT devices requires a certain amount of large memory capacity and high processing power. For this reason, it was not realistic to equip an IoT device (IoT equipment) with a relatively low processing capability and a relatively small memory capacity with a machine learning function to perform behavior prediction.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a behavior prediction support device, a behavior prediction support method, and a behavior prediction system capable of simplifying calculation processing for behavior prediction. do.

A first aspect of the present invention is a behavior prediction support device for supporting behavior prediction of a person or an animal, comprising a receiving unit for receiving sensor information from a sensor, and behavior information creation for creating behavior information based on the sensor information. a first prediction model generation unit that generates a first prediction model by machine learning using a plurality of the behavior information created in the past as teacher data; A second prediction that associates a prediction processing unit that performs behavior prediction with respect to behavior information, a behavior prediction result of the prediction processing unit, and an action condition based on the past behavior information when the behavior prediction result is obtained. a second prediction model generation unit that generates a model; a device identification unit that identifies an operation target device corresponding to the behavior prediction result included in the second prediction model; and a control device that controls the identified operation target device. and a transmission unit that transmits the second prediction model to the behavior prediction support device.

A second aspect of the present invention includes an operation target device having a communication function, and the behavior prediction support apparatus described above, wherein the operation target device includes a real-time behavior generated based on the latest sensor information output from a sensor. a matching unit for matching a comparative action condition containing information with an action condition included in the second prediction model received from the action prediction support device; A behavior prediction system comprising: a model identification unit that identifies a model; and a control unit that performs device control based on a behavior prediction result included in the identified second prediction model.

A third aspect of the present invention is a behavior prediction support method for supporting behavior prediction of a person or an animal, comprising a receiving step of receiving sensor information from a sensor, and behavior information creation of creating behavior information based on the sensor information. a first prediction model generation step of generating a first prediction model by machine learning using a plurality of the behavior information created in the past as teacher data; A second prediction that associates a prediction processing step of performing behavior prediction on behavior information, a behavior prediction result of the prediction processing step, and an action condition based on the past behavior information when the behavior prediction result was obtained. a second prediction model generation step of generating a model; a device identification step of identifying an operation target device corresponding to the behavior prediction result included in the second prediction model; and a control device controlling the identified operation target device. and a transmission step of transmitting the second prediction model to the behavior prediction support method by a computer.

A fourth aspect of the present invention includes an operation target device having a communication function, a control device that controls the operation target device, and the behavior prediction support device described above, wherein the control device includes the latest a matching unit for matching a comparison action condition including real-time action information created based on sensor information with the action condition included in the second prediction model received from the action prediction support device; and a degree of matching by the matching unit. is equal to or greater than a reference value; and a device control unit that controls the operation target device based on behavior prediction results included in the identified second prediction model. It is a behavior prediction system that

This has the effect of simplifying the arithmetic processing for behavior prediction.

BRIEF DESCRIPTION OF THE DRAWINGS It is the figure which showed roughly the whole structure of the action prediction system which concerns on one Embodiment of this invention. It is the figure which showed an example of the hardware constitutions of the action prediction assistance apparatus which concerns on one Embodiment of this invention. 1 is a functional block diagram showing an example of a behavior prediction support function included in a behavior prediction support device according to one embodiment of the present invention; FIG. It is the figure which showed an example of the action information which concerns on one Embodiment of this invention. FIG. 4 is a diagram showing an example of a matrix of transition probabilities between states in a hidden Markov model; FIG. 4 is a diagram showing an example of action conditions according to one embodiment of the present invention; It is a figure showing an example of hardware constitutions of a control device with which an operation target device concerning one embodiment of the present invention is provided. 3 is a functional block diagram showing an example of functions possessed by an operation target device according to an embodiment of the present invention; FIG. It is the flowchart which showed an example of the procedure of the action information creation process performed by the action prediction assistance apparatus which concerns on one Embodiment of this invention. It is the flow chart which showed an example of the procedure of the 2nd prediction model generation processing performed by the action prediction support device concerning one embodiment of the present invention. It is the flowchart which showed an example of the procedure of the 2nd prediction model delivery process performed by the action prediction assistance apparatus and operation target device which concern on one Embodiment of this invention. 4 is a flowchart showing an example of the procedure of action prediction processing executed by the operation target device according to one embodiment of the present invention;

An embodiment of a behavior prediction support device, a behavior prediction support method, and a behavior prediction system according to the present invention will be described below with reference to the drawings.
FIG. 1 is a diagram schematically showing the overall configuration of a behavior prediction system 1 according to this embodiment. As shown in FIG. 1, the behavior prediction system 1 includes a behavior prediction support device 10, multiple sensors 40a to 40d, and multiple operation target devices 50a and 50b. Hereinafter, the sensors 40a, 40b, etc. will be referred to as the sensors 40a, 40b, etc. when they need to be distinguished from each other, and simply referred to as the sensor 40 when there is no need to distinguish them. Also, other configurations are treated in the same way.

In FIG. 1, two sensors 40 and two operation target devices 50 are illustrated, but the number of installations is not limited to the illustrated embodiment, and the number of installations may be increased or decreased depending on the situation. is possible. The behavior prediction support device 10 , the plurality of sensors 40 , and the plurality of operation target devices 50 are communicably connected to each other via the network 15 . An example of the network 15 is, for example, the Internet.

The sensor 40 and the operation target device 50 are so-called IoT (Internet of Things) devices having communication functions. The sensor 40 is, for example, a sensor capable of recognizing the action, motion, or speech of a person or animal, and examples thereof include a human sensor, a temperature sensor, a camera, a microphone, and the like.

For example, assuming daily life at home, the operation target device 50 is a home appliance such as a lighting device, an air conditioner, a sound device, a television, or a clock. Arithmetic processing is performed, and it operates according to the processing result. Further, the operation target device 50 may include the sensor 40 . In FIG. 1, the operation target device 50a includes a sensor 40c, and the operation target device 50b includes a sensor 40d. Details of the operation target device 50 will be described later.

FIG. 2 is a diagram showing an example of the hardware configuration of the action prediction support device 10. As shown in FIG. As shown in FIG. 2, the behavior prediction support device 10 is a so-called computer, and includes a CPU (Central Processing Unit) 11, a main memory 12, a storage section 13, and a communication section 14, for example. Also, the behavior prediction support device 10 may include an input unit, a display unit, and the like. These units are connected to each other directly or indirectly via a bus, and cooperate with each other to perform various processes.

For example, the CPU 11 controls the overall action prediction support device 10 by an OS (Operating System) stored in a storage unit 13 connected via a bus, and executes various programs stored in the storage unit 13. Executes various processes.

The main memory 12 is composed of a writable memory such as a cache memory and a RAM (Random Access Memory), and is used as a working area for reading the execution program of the CPU 11 and writing processing data by the execution program.

The storage unit 13 is, for example, a ROM (Read Only Memory), a HDD (Hard Disk Drive), a flash memory, or the like. It stores an OS for controlling the entire support device 10, various device drivers for hardware operation of peripheral devices, various application software, various data and files, and the like. The storage unit 13 also stores programs for implementing various processes and various data required for implementing various processes.

The communication unit 14 has a communication interface for connecting to a network, for example, connects to the network 15 via a wireless line, establishes communication with other devices, and realizes mutual communication of information. Examples of wireless lines include cellular lines such as LTE, 3G, 4G, and 5G, WiFi, wireless LAN, Bluetooth (registered trademark), and the like.

FIG. 3 is a functional block diagram showing an example of the action prediction support function of the action prediction support device 10 according to this embodiment.

A series of processes for realizing various functions described later are stored in the storage unit 13 in the form of a program (for example, a behavior prediction support program), for example, and the CPU 11 reads this program into the main memory 12, Various functions are realized by executing information processing and arithmetic processing. The program may be pre-installed in the storage unit 13, may be provided in a state stored in another computer-readable storage medium, or may be distributed via wired or wireless communication means. may apply. Computer-readable storage media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, and the like.

The behavior prediction support device 10 is a device that supports behavior prediction of a person or an animal. As shown in FIG. 3, the behavior prediction support device 10 includes, for example, a receiving unit 21, an behavior information creation unit 22, an behavior information accumulation unit 23, a first prediction model generation unit 24, a first prediction model storage unit 25, a prediction process It has a unit 26 , a second prediction model generation unit 27 , a second prediction model storage unit 28 , a device identification unit 29 and a transmission unit 30 .

In the following, for convenience of explanation, as an example, the case where the behavior prediction system 1 according to the present embodiment predicts the behavior of a person in a living space will be described.

The receiving unit 21 receives sensor information via the network 15 from the plurality of sensors 40a to 40d. For example, the sensors 40a-40d output sensor information periodically, at predetermined timings, or when their states change.

The behavior information creation unit 22 creates behavior information based on the sensor information received from the sensor 40 . For example, the action information creating unit 22 creates action information based on the time information when the sensor information is received from the sensor 40 and the position information of the sensor 40 which acquires the sensor information.
In addition, when the action information creating unit 22 receives sensor information from a plurality of sensors 40 within a predetermined period, it may create one piece of action information from the plurality of sensor information.

FIG. 4 is a diagram showing an example of action information. As shown in FIG. 4, for example, the action information is information in which date and time information, position information, user information, and action information are associated. For example, in FIG. 4, as an example of the action information, "May 21, 2018, 6:27 am, bedroom, Mr. A moved", "May 21, 2018, 6:30 am, bedroom, Mr. A is out", "At 6:31 am on May 21, 2018, Mr. A, in the living room, the light is turned on", etc. are shown.

The action information created by the action information creation unit 22 is accumulated in the action information accumulation unit 23 and transmitted to the operation target device 50 via the transmission unit 30 .
A large amount of behavior information created by the behavior information creating unit 22 is accumulated in the behavior information accumulating unit 23 .
Hereinafter, the action information output from the action information creating unit 22 will be referred to as "real-time action information", and the action information accumulated in the action information accumulation unit 23 will be referred to as "past action information".

The first prediction model generation unit 24 generates a first prediction model by machine learning using a plurality of pieces of action information created in the past, in other words, a plurality of pieces of action information accumulated in the action information accumulation unit 23 as teacher data. to generate

For example, the first prediction model generating unit 24 acquires a plurality of behavioral information including new behavioral information from the behavioral information storage unit 23 at a predetermined timing, and uses the acquired plurality of behavioral information as teacher data for machine learning. to create a first prediction model.

Here, as an example of the predetermined timing, a predetermined time of the day (for example, noon, midnight, etc.), a predetermined time interval (for example, one hour interval, etc.), and a predetermined number of new action information The accumulated timing, when an update instruction is input by the user, when a predetermined condition is satisfied, and the like can be mentioned.

In addition, the first prediction model generation unit 24 may acquire all the behavior information stored in the behavior information storage unit 23, or new behavior information (that is, when the first prediction model was created last time Some behavioral information including behavioral information that was not used for
Then, the first prediction model generation unit 24 performs machine learning using the acquired plurality of pieces of behavior information as teacher data, and creates a first prediction model.

Machine learning, for example, generates a matrix of transition probabilities between states that change according to user behavior, like a hidden Markov model, and uses it as a prediction model. It is possible to apply various machine learning methods. FIG. 5 is a diagram showing an example of a matrix of transition probabilities between states in a hidden Markov model. Machine learning for behavior prediction is not limited to the hidden Markov model, and any known model can be used as appropriate.

The first prediction model generated by the first prediction model generation unit 24 is stored in the first prediction model storage unit 25 . Thus, the first prediction model stored in the first prediction model storage unit 25 is updated as needed by the first prediction model generation unit 24 generating the first prediction model at a predetermined timing. .

The prediction processing unit 26 uses the first prediction model stored in the first prediction model storage unit 25 to perform action prediction on past action information. For example, the prediction processing unit 26 reads one or more pieces of behavior information stored in the behavior information storage unit 23 in time series, and uses the read past behavior information as input information to perform behavior prediction using the first prediction model. conduct. As a result, future action prediction results for past action information are output. Note that the behavior prediction executed by the prediction processing unit 26 can appropriately use a known behavior prediction method similar to that used by the first prediction model generation unit 24 .

The second prediction model generation unit 27 generates a second prediction model that associates the behavior prediction result of the prediction processing unit 26 with one or more pieces of past behavior information used as input information when obtaining the behavior prediction result. Generate. For example, the second prediction model generation unit 27 combines a plurality of time-series behavior information read by the prediction processing unit 26, and a behavior prediction result obtained when behavior prediction is performed using the combination of behavior information. A second prediction model is generated by associating with .

For example, if the input information is a combination of two pieces of time-series action information: "6:27, bedroom, Mr. A has moved" and "6:30, bedroom, Mr. A has left." When the action prediction result of "living room, Mr. A, turn on the light" is obtained, a second prediction model is generated by associating the combination of these two pieces of time-series action information with the action prediction result.
In this way, the second prediction model includes an "action condition" including a combination of one or more time-series action information, and when the action prediction process is executed using the first prediction model based on the action condition It becomes a model in which the obtained "behavior prediction results" are associated with each other.

In addition, the second prediction model generating unit 27 may combine the same prediction results into one model when the same prediction result is obtained for the same action condition at different times. For example, according to the action information shown in FIG. 4, on May 21, 2018 and May 22, 2018, the same action condition exists between 6 am and 9 am. In such a case, the second prediction model may be generated by providing a temporal width.

For example, in this case, as shown in FIG. 6, the action condition is, for example, a plurality of time-series data such as "6:00, bedroom, Mr. A, moved" and "6:00, bedroom, Mr. A, left". is represented by a combination of behavioral information. Then, with this action condition, for example, the action prediction result of "living room, Mr. A, turned on the light" is associated. In addition, as described above, the action condition does not necessarily have to be composed of a combination of a plurality of pieces of action information, and may be composed of one piece of action information. Also, the action condition may include other information besides the action information.

The second predictive model generated by the second predictive model generator 27 is accumulated in the second predictive model accumulation unit 28 . In this way, the second prediction model generation unit 27 generates the second prediction model based on the prediction result of the prediction processing unit 26, so that the second prediction model is accumulated in the second prediction model storage unit 28 as needed. The Rukoto.

The device identifying unit 29 reads the second predictive model from the second predictive model storage unit 28 at a predetermined timing, and identifies the operation target device 50 based on the action prediction result included in the read second predictive model.
Here, examples of the predetermined timing include a predetermined time of the day (eg, 6:00 am, 6:00 pm, etc.), a predetermined time interval (eg, 1 hour interval, etc.), and an update instruction from the user. Examples include when an input is made and when a predetermined condition is satisfied.

The device identification unit 29 has, for example, device information of each operation target device 50 connected via the network 15 . This device information includes, for example, the name of the device (e.g., lighting, air conditioner, audio equipment, television, clock, etc.), location information where the device is installed (bedroom, living room, entrance, etc.), information (address), etc. Then, the device identification unit 29 uses this device information to identify the operation target device 50 corresponding to the action prediction result of the second prediction model.

For example, referring to the second prediction model shown in FIG. 6, the action prediction result is "Living room, Mr. A, turned on the light", which indicates that the lighting in the living room is related. Therefore, the device identifying unit 29 identifies the lighting installed in the living room as the operation target device 50 corresponding to the second prediction model.

The transmission unit 30 transmits the second prediction model corresponding to the device 50 identified by the device identification unit 29 to the operation target device 50 . Thereby, each second prediction model stored in the second prediction model accumulation unit 28 is transmitted to the operation target device 50 corresponding to the model.

The transmission unit 30 also transmits the real-time behavior information created by the behavior information creation unit 22 to each operation target device 50 connected via the network 15 .

Next, the operation target device 50 according to this embodiment will be described with reference to the drawings.
FIG. 7 is a diagram showing an example of the hardware configuration of the control device 60 included in the operation target device 50 according to this embodiment. The operation target device 50 is, for example, lighting, an air conditioner, an audio device, a television, a clock, etc., and has a control device 60 for controlling these.

As shown in FIG. 7, the control device 60 includes a CPU 61, a main memory 62, a storage section 63, a communication section 64, and a sensor 40, for example. These units are connected to each other directly or indirectly via a bus, and cooperate with each other to perform various processes. The details of each part are the same as those of a general computer, so detailed descriptions are omitted here. Further, the sensor 40 generates sensor information corresponding to the control state of the operation target device 50 . For example, when the operation target device 50 is lighting, the sensor 40 detects whether it is in a lighting state or a lighting state, and generates sensor information according to the detection result. Sensor information generated by the sensor 40 is transmitted to the behavior prediction support device 10 via the communication unit 64 .

FIG. 8 is a functional block diagram showing an example of the functions of the operation target device 50 according to this embodiment. A series of processes for realizing various functions to be described later are stored in the storage unit 63 in the form of a program, for example, and the CPU 61 reads this program into the main memory 62 and executes information processing and arithmetic processing. By doing so, various functions are realized. The program may be pre-installed in the storage unit 63, may be provided in a state stored in another computer-readable storage medium, or may be distributed via wired or wireless communication means. may apply. Computer-readable storage media include magnetic disks, magneto-optical disks, CD-ROMs, DVD-ROMs, semiconductor memories, and the like.

As shown in FIG. 8, the operation target device 50 includes, for example, a receiving unit 71, a second predictive model storage unit 72, a matching unit 73, a model specifying unit 74, a device control unit 75, a sensor 40, and a transmitting unit 76. ing.

The receiving unit 71 receives the second prediction model transmitted from the behavior prediction support device 10 via the network 15, for example. Upon receiving the second prediction model, the reception unit 71 stores it in the second prediction model storage unit 72 . As a result, the second prediction model generated and provided by the behavior prediction support device 10 is stored in the second prediction model storage unit 72 .

The receiving unit 71 also receives real-time action information transmitted from the action prediction support device 10 via the network 15, for example. Upon receiving the second prediction model, receiving portion 71 outputs it to matching portion 73 .

The collation unit 73 collates the comparison action condition including the real-time action information input via the reception unit 71 with the action condition included in the second prediction model. Here, the comparative action condition to be matched with the second prediction model may be composed of one piece of real-time action information, or may be composed of a combination of one or more pieces of action information received prior to the real-time action information. may have been

For example, the matching unit 73 creates a comparison action condition consisting of a combination of a plurality of pieces of time-series action information based on the real-time action information and one or a plurality of pieces of action information received most recently. Then, the created comparison action condition and the action condition included in the second prediction model stored in the second prediction model storage unit 72 are collated. The degree of matching between the two obtained as a matching result is output to the model specifying unit 74 .

The model identification unit 74 identifies a second prediction model whose degree of matching by the matching unit 73 is equal to or greater than a reference value. Here, the reference value is a pre-registered value. Also, this reference value may be set so as to be changeable by the user. In this way, by configuring the reference value of the degree of coincidence to be changeable, it is possible to easily adjust the accuracy of action prediction.

The device control unit 75 performs device control based on the action prediction result included in the second prediction model identified by the model identification unit 74 . For example, when the operation target device 50 is lighting, the lighting is turned on or off based on the behavior prediction result included in the second prediction model.

The sensor 40 generates sensor information under the control of the device control section 75 . For example, when the operation target device 50 is a lighting device and the device control unit 75 turns on the lighting device, sensor information indicating the lighting is generated. Further, when the device control unit 75 turns off the light, it generates sensor information indicating that the light is turned off.

The transmission unit 76 transmits sensor information generated by the sensor 40 to the behavior prediction support device 10 via the network 15 . Further, the sensor information may be transmitted (broadcast) to all devices connected via the network 15 .

Next, action information creation processing executed by the action prediction support device 10 according to the present embodiment will be described with reference to FIG. FIG. 9 is a flowchart showing an example of the procedure of action information creation processing executed by the action prediction support device 10 according to this embodiment. The action information creation process is executed, for example, each time the action prediction support device 10 receives sensor information via the network 15 .

First, when the action prediction support device 10 receives sensor information from the sensor 40 via the network 15 (SA1), it creates real-time action information based on the received sensor information (SA2), and stores it in the action information accumulation unit 23. (SA3). As a result, for example, a plurality of pieces of behavior information are accumulated in the behavior information accumulation unit 23 . Also, the behavior prediction support device 10 transmits real-time behavior information to the operation target device 50 connected to the network 15 (SA4). As a result, behavior prediction processing is executed in the operation target device 50, as will be described later.

FIG. 10 is a flow chart showing an example of the procedure of the second prediction model generation process executed by the behavior prediction support device 10 according to this embodiment. The second prediction model generation process is, for example, a process that is repeatedly performed at predetermined timings.

First, all action information stored in the action information accumulation unit 23 is read (SB1). Note that instead of reading all the action information, some action information including new action information (for example, action information created in a predetermined period in the past, a predetermined number of pieces of action information in the past) may be read. .
Subsequently, using the read behavior information as teacher data, a first prediction model is generated by machine learning (SB2).

Subsequently, the first prediction model is stored in the first prediction model storage unit 25 (SB3). Next, one or a plurality of pieces of behavior information stored in the behavior information accumulation unit 23 are read in time series, and a combination of the read one or a plurality of pieces of past behavior information is used as input information to use the first prediction model, Behavior prediction is performed (SB4).

Next, an action condition is created from the past action information when the action prediction result was obtained, and a second prediction model is generated by associating this action condition with the action prediction result (SB5). Subsequently, the second prediction model is stored in the second prediction model storage unit 28 (SB6). As a result, the second prediction model is accumulated in the second prediction model accumulation unit 28 as needed.

FIG. 11 is a flowchart showing an example of the procedure of the second prediction model distribution process executed by the action prediction support device 10 and the operation target device 50 according to this embodiment. The second prediction model distribution process is, for example, a process that is repeatedly performed at predetermined timings.

First, the behavior prediction support device 10 reads the second prediction model stored in the second prediction model storage unit 28 (SC1). Here, the read second prediction model may be limited to a newly created or updated second prediction model.

Subsequently, the behavior prediction support apparatus 10 identifies an operation target device based on the behavior prediction result included in the second prediction model (SC2), and for the identified operation target device 50, the second prediction corresponding to the device. Send the prediction model (SC3).
Each operation target device 50 receives the second prediction model transmitted from the behavior prediction support device 10 (SC4), and stores the received second prediction model in the second prediction model storage unit 72 (SC5).

FIG. 12 is a flowchart showing an example of the action prediction process performed by the operation target device 50 according to this embodiment. The behavior prediction process is executed, for example, each time the operation target device 50 receives real-time behavior information from the behavior prediction support device 10 .

First, when real-time action information is received from the action prediction support device 10 via the network 15 (SD1), comparison action conditions including the received real-time action information are created (SD2). Subsequently, the created comparison action condition and the action condition included in the second prediction model stored in the second prediction model storage unit 72 are collated (SD3). Subsequently, it is determined whether or not there is a second prediction model whose degree of matching is equal to or higher than the reference value (SD4). As a result, if there is no second prediction model whose degree of matching is equal to or higher than the reference value (SD4: NO), this processing is terminated. On the other hand, if there is a second prediction model whose degree of matching is equal to or higher than the reference value (SD4: YES), the second prediction model is specified (SD5), and based on the behavior prediction results included in the specified second prediction model to control the device (SD6). Subsequently, sensor information is generated based on device control (SD7), and the generated sensor information is transmitted to behavior prediction support device 10 via network 15 (SD8).

As described above, according to the behavior prediction support device 10 and the behavior prediction support method according to the present embodiment, a plurality of pieces of past behavior information are used as teacher data to generate the first prediction model by machine learning. Behavior prediction is performed by performing behavior prediction using one or a plurality of pieces of past behavior information in the first prediction model as input information. Then, a second prediction model is generated in which the action prediction result and the input information (combination of one or more pieces of action information) when this action prediction result is obtained are associated with each other. Subsequently, the operation target device 50 corresponding to the action prediction result included in the second prediction model is specified, and the corresponding second prediction model is transmitted to the specified operation target device 50 .

As described above, according to the behavior prediction support device 10 and the behavior prediction support method according to the present embodiment, the second prediction model simplified from the first prediction model used in machine learning is distributed to the operation target device 50. becomes possible. Furthermore, the destination of delivery of the second prediction model is limited to the operation target device 50 related to the action prediction result included in the second prediction model, so the storage capacity of the operation target device 50 can be used effectively. .

Further, according to the operation target device 50 according to the present embodiment, the comparison action condition including the real-time action information and the action condition included in the second prediction model distributed from the action prediction support device 10 are collated, and the collation result is obtained. A second prediction model having a matching rate equal to or higher than a reference value is specified, and device control is performed based on the action prediction result included in the specified second prediction model.

As a result, behavior prediction can be performed by relatively simple processing called matching processing. As a result, even an IoT device with a relatively small memory capacity and relatively low computing power can execute behavior prediction.
In addition, according to the operation target device 50 according to the present embodiment, it is possible to link behavior prediction results with device control. It becomes possible.

Furthermore, by making it possible to change the reference value used by the device 50 to be operated, it is possible to easily adjust the accuracy of behavior prediction. When adjusting the accuracy of behavior prediction with the first prediction model used in machine learning, large-scale parameter adjustment is required, and it is very difficult to adjust to the desired accuracy. On the other hand, in the present embodiment, behavior prediction is performed using the second prediction model, so it is possible to easily adjust the accuracy of behavior prediction.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. Various changes or improvements can be made to the above-described embodiments without departing from the gist of the invention, and forms with such changes or improvements are also included in the technical scope of the present invention. Also, the above embodiments may be combined as appropriate.
Further, the processing procedure described in the above embodiment is also an example, and unnecessary steps may be deleted, new steps added, or the processing order may be changed without departing from the gist of the present invention. .

For example, in the above-described embodiment, each operation target device 50 receives real-time action information from the action prediction support device 10, but the present invention is not limited to this example. For example, each operation target device 50 may receive sensor information from the sensor 40 connected via the network 15 and create real-time action information based on the received sensor information. In other words, the operation target device 50 may have the same function as the action information creating section 22 included in the action prediction support device 10 .

Further, in the above-described embodiment, the case where the operation target device 50 includes the control device 60 has been described, but the present invention is not limited to this example. For example, the control device 60 may be separate from the operation target device 50, and the control device 60 and the operation target device 50 may exchange information to realize the functions described above.
For example, the control device 60 may be connected to a network (for example, the network 15) to which the operation target device 50 is connected, and the operation target device 50 may be controlled by the control device 60 via the network. . Further, in this case, the sensor function provided in the control device 60 may remain installed in the operation target device 50 .
Furthermore, the control device 60 may be shared by a plurality of devices 50 to be operated, and may be configured to integrally control the devices 50 to be operated from the control device 60 .

1: Action prediction system 10: Action prediction support device 11: CPU
12: Main memory 13: Storage unit 14: Communication unit 15: Network 21: Reception unit 22: Action information creation unit 23: Action information storage unit 24: First prediction model generation unit 25: First prediction model storage unit 26: Prediction Processing unit 27: second prediction model generation unit 28: second prediction model storage unit 29: device identification unit 30: transmission unit 40 (40a to 40d): sensor 50 (50a to 50b): operation target device 60: control device 61 : CPU
62: main memory 63: storage unit 64: communication unit 71: reception unit 72: second prediction model storage unit 73: matching unit 74: model identification unit 75: device control unit 76: transmission unit

Claims (10)

A behavior prediction support device for supporting behavior prediction of a person or an animal,
a receiver that receives sensor information from the sensor;
an action information creation unit that creates action information based on the sensor information;
a first prediction model generation unit that generates a first prediction model by machine learning using the behavior information created in the past as teacher data;
a prediction processing unit that performs behavior prediction for the past behavior information using the first prediction model;
a second prediction model generation unit that generates a plurality of second prediction models that associate the action prediction result of the prediction processing unit with action conditions based on the past action information when the action prediction result was obtained;
a device identification unit that identifies , for each of the second prediction models, an operation target device corresponding to the behavior prediction result included in the second prediction model;
In order to perform device control based on behavior prediction results included in each of the second prediction models, the second prediction model corresponding to the operation target device is provided to a control device that controls the specified operation target device. A behavior prediction support device comprising a transmitting unit for transmitting. The first prediction model generation unit acquires, at a predetermined timing, a plurality of pieces of behavior information including the new behavior information from a behavior information storage unit in which a plurality of pieces of behavior information is accumulated, 2. The behavior prediction support device according to claim 1, wherein said first prediction model is updated by performing machine learning using information as teacher data. The prediction processing unit performs the behavior prediction using a combination of the plurality of time-series behavior information as input information,
The second prediction model generation unit associates the behavior prediction result of the prediction processing unit with an action condition including a combination of the plurality of time-series behavior information when the behavior prediction result is obtained, thereby 3. The behavior prediction support device according to claim 1, which generates a second prediction model. an operation target device having a communication function;
A behavior prediction support device according to any one of claims 1 to 3,
The operation target device is
a matching unit that compares a comparative action condition including real-time action information created based on the latest sensor information output from a sensor with the action condition included in the second prediction model received from the action prediction support device; ,
a model identification unit that identifies the second prediction model whose degree of matching by the matching unit is equal to or greater than a reference value;
a control unit that performs device control based on behavior prediction results included in the specified second prediction model;
A behavior prediction system comprising: 5. The behavior prediction system according to claim 4, wherein said reference value is changeable by a user. The action prediction system according to claim 4 or 5, wherein the operation target device is equipped with the sensor. 7. The behavior prediction system according to any one of claims 4 to 6, wherein the operation target device acquires the real-time behavior information from the behavior prediction support device. The operation target device is
a receiving unit that receives sensor information from the sensor;
7. The behavior prediction system according to any one of claims 4 to 6, further comprising a behavior information creation unit that creates the real-time behavior information based on the sensor information. A behavior prediction support method for supporting behavior prediction of a person or an animal, comprising:
a receiving step for receiving sensor information from the sensor;
an action information creating step of creating action information based on the sensor information;
A first prediction model generation step of generating a first prediction model by machine learning using a plurality of the behavior information created in the past as teacher data;
A prediction processing step of performing behavior prediction for the past behavior information using the first prediction model;
a second prediction model generation step of generating a plurality of second prediction models in which the action prediction result of the prediction processing step and the action condition based on the past action information when the action prediction result was obtained are associated;
a device identification step of identifying , for each of the second prediction models, an operation target device corresponding to the behavior prediction result included in the second prediction model;
In order to perform device control based on behavior prediction results included in each of the second prediction models, the second prediction model corresponding to the operation target device is provided to a control device that controls the specified operation target device. A behavior prediction support method in which a computer executes a transmission step of transmitting. an operation target device having a communication function;
a control device that controls the device to be operated;
A behavior prediction support device according to any one of claims 1 to 3,
The control device is
a matching unit that compares a comparative action condition including real-time action information created based on the latest sensor information output from a sensor with the action condition included in the second prediction model received from the action prediction support device; ,
a model identification unit that identifies the second prediction model whose degree of matching by the matching unit is equal to or greater than a reference value;
a device control unit that controls the operation target device based on a behavior prediction result included in the identified second prediction model;
A behavior prediction system comprising: