JP6629779B2 - Gateway device, control method therefor, and program - Google Patents

Description

  The present invention relates to a gateway device capable of collecting data from sensors existing under the device and dynamically collecting data from peripheral gateway devices in an autonomous and decentralized manner, a control method thereof, and a program.

  Sensor processing that is indispensable to the IoT (Internet of Things) environment in which many sensors are connected to a network is generally realized by directly collecting data acquired by each sensor in the cloud and processing the data. (Non-Patent Document 1). In this case, as the scale of the IoT environment increases, the amount of data transmitted from the sensor to the cloud increases, and the power consumption of the entire network due to data transmission increases.

  To solve such a problem, a network configuration in which a gateway (for example, a fog node) having a cloud function is installed between a cloud and a sensor is known (Non-Patent Document 2). Each gateway is provided at a position relatively close to the sensors under its control, collects data acquired by those sensors, performs predetermined processing (averaging processing, etc.) on the collected data, and stores the result in the cloud. Send to This eliminates the need to directly transmit data from each sensor to the cloud, so that a reduction in the amount of data transmitted to the cloud and a reduction in power consumption of the entire network due to data transmission can be expected.

A. Al-Fuqaha et al., “Internet of Things: A Survey on Enabling Technologies, Protocols, and Applications,” in IEEE Communication Survey & Tutorials, vol. 17, no. 4, 2015. M. Chiang et al., “Fog and IoT: An Overview of Research Opportunities,” in IEEE Internet of Things Journal, vol. 3, no. 6, 2016.

  However, in the above-described network configuration, each gateway cannot grasp the reliability of the processing result obtained by the data processing, and always transmits the processing result to the cloud regardless of the reliability of the processing result. As described above, the power consumption of the entire network increases due to the fact that each gateway transmits not only a highly reliable processing result but also a low-reliable processing result. Further, the calculation model used for data processing in the gateway cannot cope with the daily fluctuation of the data set collected from each sensor, and it may be difficult to realize highly accurate data processing.

  The present invention has been made in view of the above problems. The present invention provides a technology for transmitting a highly reliable data processing result to a server using a determiner generated by machine learning in a gateway device capable of autonomously and decentralized data collection from peripheral gateway devices. It is intended to be.

  The gateway device according to one embodiment of the present invention is a gateway device that has a plurality of sensors that output sensor data under its control, and is a determiner that performs data processing on the sensor data. Generating means for generating, by machine learning using sensor data collected in the past, the determination unit that outputs the determination result of the target sensor and the reliability of the determination result; and a collection range corresponding to the target sensor. Processing means for collecting sensor data output from one or more of the sensors, and performing the data processing on the collected sensor data using the determiner generated by the generating means; Transmitting means for transmitting a result of the data processing to a server in accordance with the determination result output from the device, wherein the processing means comprises: When a sensor under the control of the gateway device is included in the collection range, sensor data output from the sensor is collected from the other gateway device by inter-gateway communication, and the processing unit includes: The sensor data collection and the data processing by the determiner on the collected sensor data are repeated while expanding the collection range based on the reliability output from the determiner.

  According to the present invention, in a gateway device capable of autonomously and decentralized data collection from a peripheral gateway device, a highly reliable data processing result is transmitted to a server by using a determiner generated by machine learning. Becomes possible.

FIG. 1 is a diagram illustrating an example of a network configuration according to an embodiment. FIG. 1 is a block diagram illustrating a configuration example of a gateway device according to an embodiment. Flowchart showing the procedure of preprocessing according to one embodiment The figure which shows the example of generation | occurrence | production of the determiner in the preprocessing concerning one Embodiment 4 is a flowchart illustrating a procedure of a determination process according to an embodiment. FIG. 7 is a diagram illustrating an example of extending a data collection range in a determination process according to an embodiment. 4 is a flowchart illustrating a procedure of a feedback process according to an embodiment.

  Hereinafter, exemplary embodiments of the present invention will be described with reference to the drawings. In the following drawings, components that are not necessary for the description of the embodiment are omitted from the drawings.

<Network configuration>
FIG. 1 is a diagram illustrating an example of a network configuration including a gateway device (GW) according to the present embodiment. In the network configuration of the present embodiment, a plurality of gateway devices (G1, G2, G3,...) Are respectively provided between a plurality of subordinate sensors and a cloud (a server on the cloud). In the example of FIG. 1, nine sensors S11, S12,..., S19 are provided under the gateway device G1. Similarly, nine sensors S21, S22, ..., S29 are provided under the gateway device G2, and nine sensors S31, S32, ..., S39 are provided under the gateway device G3. . Note that an arbitrary number of sensors can be provided under each gateway device.

  Each gateway device is communicably or wirelessly connected to each subordinate sensor and can acquire data (sensor data) output from each sensor. Each gateway device has a gateway function of collecting sensor data from a plurality of subordinate sensors, performing predetermined processing on the collected data, and transmitting the processing result to the cloud.

  In the present embodiment, gateway-to-gateway communication in which one gateway device directly communicates with another gateway device or indirectly communicates via one or more gateway devices is introduced. With the introduction of the inter-gateway communication, it is possible to mutually acquire sensor data between the gateway devices. For example, the gateway device G1 can acquire sensor data output from the sensors S21 to S29 from the gateway device G2 by communication between gateways. In addition, the gateway device G1 can acquire the sensor data output from the sensors S31 to S39 from the gateway device G3 by communication between the gateways.

  Each gateway device in the network previously holds arrangement information indicating a gateway device having each sensor in the network under its control (that is, which gateway device each sensor is under its control). In addition, each gateway device holds in advance route information indicating a route for reaching (accessing) any other gateway device. Each gateway device can perform gateway-to-gateway communication by using these pieces of information without going through the cloud. The gateway device of the present embodiment can collect sensor data autonomously and decentralized from peripheral gateway devices using such inter-gateway communication.

  If a new sensor or gateway device is added to the network, or if a sensor or gateway device that has already been placed is removed, each gateway device sends the placement information or information based on the notification from the cloud. The route information may be updated. Alternatively, each gateway device may update the arrangement information or the route information based on a notification from another gateway device.

<Overview of processing by gateway device>
The gateway device according to the present embodiment collects sensor data, performs data processing on the collected data by using a calculation model (determiner) generated by machine learning in advance, and only processes a processing result (or a processing result if necessary). Information indicating the result) to the cloud. The gateway device, based on the reliability of the result of the data processing as a reference, extends the data collection range not only to the sensors under its own device but also to the sensors under the nearby gateway device as necessary, and Collect data. As described above, the gateway device increases the sensor data to be subjected to the data processing, thereby improving the accuracy of the data processing and enabling the highly reliable processing result to be transmitted to the cloud. In other words, by transmitting only highly reliable processing results from each gateway device to the cloud as needed, the power consumption of the entire network due to data transmission from the gateway device to the cloud is reduced as compared to the past. I do.

More specifically, the gateway device performs the following three stages of processing.
Preprocessing:
First, the gateway device performs machine learning using data collected in the past, thereby generating a calculation model (determiner) used for data processing on the collected sensor data.
Determination process:
Next, the gateway device processes the sensor data collected from the sensors under its own device and, if necessary, from the peripheral gateway devices, using the determiner generated by the preprocessing, and when necessary, Only send the processing result to the cloud.
Feedback processing:
Further, the gateway device updates the determiner based on the sensor data collected in the determination process. This makes it possible for the determiner used in the determination process to respond to daily fluctuations in sensor data collected from each sensor.

<Functional configuration of gateway device>
FIG. 2 is a block diagram illustrating a functional configuration example of the gateway device according to the present embodiment. As shown in FIG. 2, the gateway device roughly includes a pre-processing unit 21, a determination processing unit 22, a feedback processing unit 23, and a data storage unit 24. The pre-processing unit 21, the determination processing unit 22, and the feedback processing unit 23 are functional units that implement the above-described pre-processing, determination processing, and feedback processing, respectively. The pre-processing unit 21, the determination processing unit 22, and the feedback processing unit 23 can be realized in the gateway device by hardware or software, or a combination thereof. When implemented by software, one or more processors (not shown) such as CPUs included in the gateway device read and execute a control program stored in a storage device (not shown) such as a ROM. Can be done.

(Preprocessing)
The preprocessing unit 21 performs machine learning using sensor data collected in the past to generate a determiner used for data processing on the collected sensor data. The preprocessing unit 21 according to the present embodiment generates a plurality of determiners by performing machine learning using different sets (sizes) of sensor data to be subjected to machine learning. Further, the preprocessing unit 21 confirms the accuracy of the determination result obtained by each of the generated determiners, and determines one of the determiners that satisfies a predetermined criterion as a default determiner for the determination process by the determination processing unit 22. I do. If the plurality of determiners satisfy the criterion, the optimal determiner that is the determiner that requires the least amount of data (the required data amount) for data processing is determined as the default determiner among the plurality of determiners. May be. Further, the preprocessing unit 21 determines a data amount required for data processing by the default determiner. The data corresponding to the required data amount may include not only sensor data collected from a sensor under its own device but also sensor data collected from a peripheral gateway device. The preprocessing unit 21 stores the generated plurality of determiners, the default determiner, and information indicating the required data amount in the data storage unit 24.

(Determination process)
The determination processing unit 22 collects the sensor data, processes the collected sensor data using the determiner stored in the data storage unit 24, and transmits information indicating the processing result of the sensor data to the cloud ( Upload). At that time, the determination processing unit 22 uses a default determination unit among a plurality of determination units stored in the data storage unit 24.

  The determination processing unit 22 expands the collection range of the sensor data as necessary based on the reliability of the determination result by the determiner, and collects the sensor data within the expanded range. If the extended collection range includes (a subordinate sensor of) a peripheral gateway device, the determination processing unit 22 collects sensor data from the peripheral gateway device using communication between gateways. The determination processing unit 22 reprocesses the collected sensor data using the determiner to determine again whether information indicating the processing result of the sensor data should be transmitted to the cloud.

  In this manner, the determination processing unit 22 collects (acquires) sensor data from other gateway devices using inter-gateway communication as necessary, thereby improving the accuracy of data processing using the determiner. Let it. The determination processing unit 22 acquires a determination result with a higher degree of reliability, and when the determination result indicates transmission to the cloud, transmits (uploads) information indicating the processing result of the sensor data to the cloud. In addition, based on the sensor data collected by the determination processing unit 22, the preprocessing unit 21 periodically generates a plurality of determiners by machine learning (that is, every time a predetermined time elapses) and determines a default determiner. May be performed.

(Feedback processing)
The feedback processing unit 23 acquires the sensor data used for the determination processing by the determination processing unit 22 and performs machine learning using the acquired sensor data, thereby updating the determiner used for the determination processing. The feedback processing unit 23 stores the updated determiner in the data storage unit 24 so that the updated determiner can be used for the determination process by the determination processing unit 22.

  Hereinafter, specific processing executed by the gateway device will be described more specifically with reference to FIGS. Here, a temperature sensor that measures a temperature (air temperature) and outputs a measured value is used as a sensor under the control of each gateway device, and a certain target sensor is connected to a temperature output from a peripheral (nearby) sensor. An example will be described in which it is determined whether or not a sensor outputs extremely different temperatures (hereinafter, also referred to as a “hot spot”).

<Pre-processing procedure>
FIG. 3 is a flowchart illustrating a procedure of the pre-processing performed by the pre-processing unit 21. This pre-processing is executed in a gateway device having a sensor (target sensor) to be a hot spot determination target under its control. FIG. 4 is a diagram illustrating an example of generation of a determiner in preprocessing.

  In step S301, the preprocessing unit 21 acquires a data set including sensor data obtained in the past. The past data set acquired here includes the sensor data output from the target sensor and the sensor data output from the sensors under the gateway device under the target sensor and the gateway device therearound. Can be. Further, the acquired data set includes not only sensor data indicating the measured value of the temperature by each sensor, but also information indicating whether or not the target sensor is a hot spot.

  Next, in step S302, the preprocessing unit 21 classifies each sensor data included in the acquired data set into learning data or test data, as shown in FIG. Further, in step S303, the preprocessing unit 21 determines a plurality of data sets of different sizes, each of which includes learning data, and advances the processing to step S304. FIG. 4 shows an example in which N data sets in which some data overlap between data sets are determined.

  In step S304, the preprocessing unit 21 generates a plurality of determiners by performing machine learning (for example, a support vector machine) using each of the data sets determined in step S303. Here, machine learning is performed for each of the N data sets (K = 1, 2, 3,..., N), so that N determiners (K = 1, 2, 3,...). ., N) are generated.

  The input of each determiner generated in step S304 is a sensor data string output from the target sensor and peripheral sensors. The determiner determines whether the target sensor is a hot spot by performing data processing on the input sensor data sequence. The output of the determiner is the determination result and the reliability of the determination result. The determination result indicates whether or not the target sensor is a hot spot (that is, whether or not information indicating the processing result of the sensor data should be transmitted (uploaded) to the cloud). The reliability indicates the likelihood of the processing result and corresponds to the probability of uploading information indicating the processing result to the cloud.

  Thereafter, in step S305, the preprocessing unit 21 uses the test data to check the accuracy of the determination result of each of the determiners generated in step S304, and compares the accuracy between the determiners. Further, in step S306, the preprocessing unit 21 determines, by the determination processing unit 22, the determination unit (K = X), of the plurality of determination units generated in step S304, in which the accuracy of the determination result satisfies a predetermined criterion. Determined as the default determiner for processing. The preprocessing unit 21 determines, for example, the optimal determiner as a default determiner. The preprocessing unit 21 stores the generated plurality of determiners, the default determiner, and the information indicating the required data amount in the data storage unit 24, and ends the preprocessing.

  When all data included in the data set acquired in step S301 is used as learning data, the preprocessing unit 21 does not compare the accuracy of the determination result between the generated determiners, and selects any one of them. The determiner is determined as a default determiner for the determination process. In this case, for example, a determiner generated by machine learning using a data set consisting of past sensor data of only the target sensor, or a machine using the smallest data set (K = 1 in FIG. 4) A determiner generated by learning (that is, a determiner that requires the least amount of data for data processing by the determiner) may be determined as the default determiner.

<Procedure of judgment processing>
FIG. 5 is a flowchart illustrating a procedure of the determination process performed by the determination processing unit 22. This determination processing is executed in a gateway device having a sensor (target sensor) to be a hot spot determination target similarly to the preprocessing. FIG. 6 is a diagram illustrating an example of extending the data collection range in the determination process.

  In step S501, the determination processing unit 22 collects sensor data at a certain time, which is necessary for data processing by the default determiner obtained by the pre-processing. When it is necessary to collect sensor data output from a sensor under a gateway device different from the target sensor, the determination processing unit 22 uses sensor-to-gateway communication to transfer sensor data from a peripheral gateway device. collect. Note that the collection range of the sensor data in step S501 is a range that is predetermined as an initial collection range, may be a range that includes only the sensor data of the target sensor, or corresponds to the above-described necessary data amount. It may be a range.

  Next, in step S502, the determination processing unit 22 determines whether the target sensor is a hot spot by performing data processing on the collected sensor data using a default determiner. Further, in step S503, the determination processing unit 22 compares the reliability of the determination result by the determiner with a predetermined threshold, and determines whether the reliability exceeds the threshold. If the reliability exceeds the threshold, the determination processing unit 22 proceeds with the process to step S504.

  In step S504, the determination processing unit 22 determines whether it is necessary to transmit sensor information indicating the target sensor to the cloud (a server on the cloud) as information indicating the processing result of the sensor data. Specifically, if the determination result by the default determiner is a result of transmitting (uploading) the processing result of the sensor data to the cloud (that is, the target sensor is a hot spot), the transmission of the sensor information is performed. Is determined to be necessary. On the other hand, in other cases, it is determined that transmission of sensor information is not necessary. When determining that the transmission of the sensor information is necessary, the determination processing unit 22 transmits the sensor information (for example, the identification ID that is the identification information of the target sensor) in step S505, and ends the process.

  On the other hand, in step S503, when the reliability of the determination result by the default determiner does not exceed the threshold value (is equal to or less than the threshold value), the determination processing unit 22 advances the processing to step S506. In step S506, the determination processing unit 22 extends the sensor data collection range for each minimum unit as shown in FIG. Here, FIG. 6 illustrates an example in which the sensor S12 under the gateway device G1 is a target sensor, and the gateway device G1 collects sensor data. In FIG. 6, the sensors S11 to S19 are under the gateway device G1, the sensors S21 to S29 are under the gateway device G2, the sensors S31 to S39 are under the gateway device G3, and the sensors S41 to S49 are under the gateway device G4.

  FIG. 6A shows an example of the initial collection range, in which only the sensor data output from the target sensor (sensor S12) is included in the initial collection range. In this case, the gateway device G1 collects sensor data output from the sensor S12 without performing inter-gateway communication. FIG. 6B shows an example in which the collection range is extended in step S506, and the collection range is extended to a range of one hop from the target sensor with one hop as a minimum unit. In this case, the gateway device G1 collects the sensor data output from the sensors S11 to S16 under its own device, and also collects the sensor data output from the sensors S27 to S29 from the gateway device G2 by inter-gateway communication. I do.

  Next, in step S507, the determination processing unit 22 determines whether the extended collection range has exceeded a predetermined limit (reached the limit), and if not, advances the process to step S508. If it has exceeded, the process proceeds to step S509. In step S508, the determination processing unit 22 increases the sensor data used for data processing by the default determiner by collecting sensor data in the extended collection range as described above, and returns the processing to step S502. Thus, the determination processing unit 22 performs data processing again on the collected sensor data in step S502.

  In this way, the determination processing unit 22 repeats the data processing by the default determiner in step S502 while expanding the collection range as necessary until the sensor data collection range exceeds the limit, thereby setting the threshold value. Obtain a judgment result with higher reliability. Further, the determination processing unit 22 transmits (uploads) sensor information indicating a result of the data processing to the cloud according to the determination result having the reliability exceeding the threshold. When the data processing using the default determiner is completed, the determination processing unit 22 stores the collected sensor data in the data storage unit 24 for the feedback processing by the feedback processing unit 23.

  On the other hand, in step S509, the determination processing unit 22 performs data processing on the collected sensor data by the heuristic method, and proceeds to step S504. In the heuristic method, the target sensor is not hot-learned based on sensor data obtained in the past, but uses an index (average value or standard deviation) obtained by analysis based on the entire sensor data obtained at present. It is determined whether or not it is a spot. In step S504, sensor information indicating the result of the data processing is transmitted (uploaded) to the cloud according to the determination result by the heuristic method.

<Procedure of feedback processing>
FIG. 7 is a flowchart illustrating a procedure of the feedback processing executed by the feedback processing unit 23. This feedback processing is executed in a gateway apparatus having a sensor (target sensor) to be a hot spot determination target, similarly to the preprocessing and the determination processing.

  The basic procedure of the feedback process is the same as the pre-process shown in FIG. However, in step S701, the feedback processing unit 23 acquires collected sensor data that is used in the determination process by the determination processing unit 22 and stored in the data storage unit 24. In step S704, the feedback processing unit 23 updates each of the determiners already stored in the data storage unit 24 by machine learning for each data set including the collected sensor data. Further, the feedback processing unit 23 stores the plurality of updated determiners and the updated default determiner and the information indicating the required data amount in the data storage unit 24, so that the existing determiners and the information are stored. Perform an update. As a result, the updated determiner and information stored in the data storage unit 24 are used (feedback) in the determination processing by the determination processing unit 22.

  As described above, in the present embodiment, the gateway device has a plurality of sensors that output sensor data under its control. The pre-processing unit 21 is a determiner that performs data processing on sensor data, and outputs a determination result for a target sensor among a plurality of sensors and a reliability of the determination result, Generated by machine learning using collected sensor data. The determination processing unit 22 collects sensor data output from one or more sensors within the collection range corresponding to the target sensor, and determines the determined sensor ( Data processing is performed using a default determiner). At this time, the determination processing unit 22 repeats the collection of the sensor data and the data processing of the collected sensor data by the determiner while expanding the collection range based on the reliability output from the determiner. When a sensor under the control of another gateway device is included in the collection range, the determination processing unit 22 collects sensor data output from the sensor from the other gateway device by communication between the gateways. I do. Thereafter, the determination processing unit 22 determines the data according to the determination result output from the determiner (for example, only when the determination result indicates that the result of the data processing should be transmitted to the server (uploaded to the cloud)). Send the processing result to the server.

  According to the present embodiment, by using the determiner generated by machine learning, the result of low-reliability data processing is not transmitted to the server (cloud), and only the result of high-reliability data processing is transmitted to the server. It becomes possible to send. That is, the result of the data processing is transmitted to the server only when necessary, and the power consumption of the entire network due to the data transmission can be significantly reduced. Further, since only the result of the highly reliable data processing can be transmitted to the server, it is possible to guarantee the reliability of the data processing for the sensor data. Further, by performing data processing on the sensor data by the gateway device and transmitting the processing result to the server (cloud), it is not necessary to perform the same data processing on the server, and the processing load on the server can be reduced.

  Further, in the present embodiment, the feedback processing unit 23 performs a feedback process of updating the determiner generated by the preprocessing unit 21 by machine learning using the sensor data collected by the determination processing unit 22. As a result, it is possible to make the determiner used for the determination processing by the determination processing unit 22 correspond to the daily fluctuation of the sensor data collected from each sensor, thereby realizing more accurate data processing. Will be possible.

  Note that the gateway device according to the present embodiment can be realized by a computer program for causing a computer to function as a gateway device. The computer program is stored in a computer-readable storage medium and can be distributed, or can be distributed via a network.

G1, G2, G3, G4: gateway devices S11 to S19, S21 to S29: sensors S31 to S39, S41 to S49: sensor 21: preprocessing unit 22: determination processing unit 23: feedback processing unit 24: data storage unit

Claims (13)

A gateway device having a plurality of sensors that output sensor data under its control,
A determinator that performs data processing on sensor data, wherein the determinator that outputs a determination result for a target sensor of the plurality of sensors and the reliability of the determination result includes a sensor collected in the past. Generating means for generating by machine learning using data,
Collecting sensor data output from one or more sensors within a collection range corresponding to the target sensor, and performing the data processing on the collected sensor data using the determiner generated by the generating unit. Processing means for performing
Transmitting means for transmitting a result of the data processing to a server according to a determination result output from the determiner,
When the sensor under the control of another gateway device is included in the collection range, the processing unit collects sensor data output from the sensor from the other gateway device by inter-gateway communication. ,
The processing unit repeats the collection of sensor data and the data processing by the determiner on the collected sensor data while expanding the collection range based on the reliability output from the determiner. Gateway device. The gateway device according to claim 1, further comprising an update unit that updates the determiner generated by the generation unit by machine learning using the sensor data collected by the processing unit. When the data processing using the determiner is completed, the processing unit stores the collected sensor data in a storage unit,
The gateway device according to claim 2, wherein the updating unit updates the determiner by machine learning using sensor data stored in the storage unit. The processing unit repeats the collection of sensor data and the data processing by the determiner on the collected sensor data while expanding the collection range until the reliability output from the determiner exceeds a threshold. The gateway device according to any one of claims 1 to 3, wherein: The processing means, when the expanded collection range reaches a predetermined limit, without using the determiner, performs data processing by heuristic method on the collected sensor data,
5. The gateway device according to claim 1, wherein the transmitting unit transmits a result of the data processing to a server according to a determination result by the heuristic method. 6. The gateway device previously stores arrangement information indicating whether each sensor in the network is under which gateway device, and route information indicating a route for accessing another gateway device in advance,
The said processing means collects sensor data from another gateway apparatus by performing the said inter-gateway communication based on the said arrangement | positioning information and the said route information. The Claim 1 characterized by the above-mentioned. The gateway device as described. The said gateway apparatus updates the said arrangement | positioning information and the said route information based on the notification from the said server, when a sensor or a gateway apparatus is added or removed in the said network. The claim of Claim 6 characterized by the above-mentioned. Gateway device. The said transmission means transmits the result of the said data processing to the said server only when the said judgment result shows that the result of the said data processing should be transmitted to the said server. The gateway device according to any one of the above. The gateway device according to any one of claims 1 to 8, wherein the transmitting unit transmits the identification information of the target sensor as a result of the data processing. The generation means,
Acquisition means for acquiring sensor data collected in the past,
Classification means for classifying each sensor data acquired by the acquisition means for learning or for testing,
A first determination unit that determines a plurality of data sets of different sizes, each of the data sets including the learning sensor data;
Learning means for generating a plurality of determiners by performing machine learning for each data set determined by the first determining means;
Using the sensor data for the test, the determination accuracy of each of the determiners generated by the learning unit is confirmed, and the determination unit that satisfies a predetermined criterion among the plurality of determiners is subjected to the processing. Second determining means for determining a determiner used by the means,
The gateway device according to any one of claims 1 to 9, wherein the processing unit performs the data processing using a determiner determined by the second determination unit. The second determining means determines, from among the determiners satisfying the predetermined criterion, a determiner that requires the least amount of sensor data for the data processing as a determiner used by the processing means. The gateway device according to claim 10, wherein: A method for controlling a gateway device having a plurality of sensors that output sensor data under the control thereof,
A determinator that performs data processing on sensor data, wherein the determinator that outputs a determination result for a target sensor of the plurality of sensors and the reliability of the determination result includes a sensor collected in the past. A generation step generated by machine learning using data;
Collecting sensor data output from one or more sensors within a collection range corresponding to the target sensor, and performing the data processing on the collected sensor data using the determiner generated in the generation step. A processing step of performing
A transmitting step of transmitting a result of the data processing to a server according to the determination result output from the determiner,
In the processing step, when a sensor under the control of another gateway device is included in the collection range, sensor data output from the sensor is collected from the other gateway device by inter-gateway communication. ,
In the processing step, the collection of sensor data and the data processing by the determiner on the collected sensor data are repeated while expanding the collection range based on the reliability output from the determiner. A method of controlling a gateway device.   A program for causing a computer to execute each step of the method for controlling a gateway device according to claim 12.

Images