阅读说明：本技术 机器学习装置 (Machine learning device ) 是由 横山大树 浅原则己 于 2021-06-07 设计创作，主要内容包括：本发明提供一种机器学习装置,在能够向外部供给电力的车辆中,抑制在灾害时能够从车辆向外部供给的电力量因与机器学习模型的学习相关的处理而减少的情况。机器学习装置设置于能够向外部供给电力的车辆(3),并具备实施与在车辆中使用的机器学习模型的学习相关的处理的学习部(51)。学习部在取得了灾害信息的情况下,与未取得灾害信息的情况相比,使与学习相关的处理中的电力消耗量降低。(The invention provides a machine learning device, which can restrain the reduction of the electric power amount supplied from a vehicle to the outside during disaster caused by the processing related to the learning of a machine learning model in the vehicle capable of supplying electric power to the outside. The machine learning device is provided in a vehicle (3) capable of supplying electric power to the outside, and is provided with a learning unit (51) that performs processing relating to learning of a machine learning model used in the vehicle. When disaster information is acquired, the learning unit reduces the amount of power consumption in the processing related to learning, as compared to when disaster information is not acquired.)

1. A machine learning device provided in a vehicle capable of supplying electric power to the outside,

the machine learning device includes a learning unit that performs processing related to learning of a machine learning model used in the vehicle,

the learning unit reduces the amount of power consumption in the process related to the learning when disaster information is acquired, as compared to when the disaster information is not acquired.

2. The machine learning apparatus of claim 1,

the machine learning device further includes a position information acquisition unit that acquires position information of the vehicle,

when power supply from the vehicle to the outside is predicted based on the disaster information and the location information, the learning unit reduces the amount of power consumption in the process related to the learning, as compared to a case where the power supply is not predicted.

3. The machine learning apparatus of claim 2,

the position information acquiring unit acquires a destination of the vehicle,

when the supply of electric power from the vehicle to the outside is predicted based on the disaster information and the destination, the learning unit reduces the amount of electric power consumption compared to a case where the supply of electric power is not predicted.

4. The machine learning device of any one of claims 1 to 3,

the learning portion reduces the amount of power consumption by stopping processing related to the learning.

5. The machine learning apparatus of claim 4,

the machine learning device further includes an output device control unit that controls an output device provided in the vehicle,

the output device control portion confirms permission to stop the process related to the learning to a driver of the vehicle via the output device,

the learning portion does not stop the process related to the learning when the driver does not allow the stop of the process related to the learning.

6. A machine learning device is provided with:

a communication device capable of communicating with a vehicle capable of supplying electric power to the outside; and

a control device that performs learning of a machine learning model and transmits the learned machine learning model to the vehicle via the communication device,

the control device stops transmission of the learned machine learning model to the vehicle when disaster information is acquired.

Technical Field

The present invention relates to a machine learning device.

Background

In recent years, with the development of AI (artificial intelligence) technology, control using a machine learning model such as a neural network model is being studied in a vehicle. For example, in the machine learning device described in patent document 1, an electronic control unit provided in a vehicle learns a neural network model, and outputs an estimated value of the temperature of an exhaust gas purification catalyst from the learned neural network model.

Documents of the prior art

Patent document

Patent document 1: japanese patent laid-open publication No. 2019-183698

Disclosure of Invention

Problems to be solved by the invention

However, in a vehicle having a large battery capacity such as a plug-in hybrid vehicle (PHV), the electric power stored in the battery can be supplied to the outside of the vehicle. Therefore, when a power failure occurs due to a disaster, the vehicle can be effectively used as a power supply source.

However, when processing related to learning of the machine learning model is performed in the vehicle, electric power necessary for learning is consumed in addition to electric power necessary for traveling of the vehicle. As a result, the amount of power consumption in the vehicle increases, and there is a possibility that the power necessary for disaster cannot be secured.

In view of the above problems, an object of the present invention is to suppress, in a vehicle capable of supplying electric power to the outside, a decrease in the amount of electric power that can be supplied from the vehicle to the outside in a disaster due to processing related to learning of a machine learning model.

Means for solving the problems

The gist of the present disclosure is as follows.

(1) A machine learning device provided in a vehicle capable of supplying electric power to the outside, wherein the machine learning device includes a learning unit that performs processing related to learning of a machine learning model used in the vehicle, and the learning unit reduces the amount of electric power consumed in the processing related to learning when disaster information is acquired as compared with a case where the disaster information is not acquired.

(2) The machine learning device according to the above (1), further comprising a position information acquisition unit that acquires position information of the vehicle, wherein when power supply from the vehicle to the outside is predicted based on the disaster information and the position information, the learning unit reduces power consumption in the process related to the learning, as compared to a case where the power supply is not predicted.

(3) The machine learning device according to (2) above, wherein the location information acquisition unit acquires a destination of the vehicle, and when power supply from the vehicle to the outside is predicted based on the disaster information and the destination, the learning unit reduces the amount of power consumption compared to a case where the power supply is not predicted.

(4) The machine learning device according to any one of the above (1) to (3), wherein the learning section reduces the amount of power consumption by stopping a process related to the learning.

(5) The machine learning device according to the above (4), further comprising an output device control unit that controls an output device provided in the vehicle, wherein the output device control unit confirms permission to stop the process related to the learning to a driver of the vehicle via the output device, and wherein the learning unit does not stop the process related to the learning when the driver does not permit the stop of the process related to the learning.

(6) A machine learning device is provided with: a communication device capable of communicating with a vehicle capable of supplying electric power to the outside; and a control device that performs learning of a machine learning model and transmits the learned machine learning model to the vehicle via the communication device, wherein the control device stops transmission of the learned machine learning model to the vehicle when disaster information is acquired.

Effects of the invention

According to the present invention, in a vehicle capable of supplying electric power to the outside, it is possible to suppress a reduction in the amount of electric power that can be supplied from the vehicle to the outside in a disaster due to processing related to learning of a machine learning model.

Drawings

Fig. 1 is a schematic configuration diagram of a machine learning system according to a first embodiment of the present invention.

Fig. 2 is a diagram schematically showing the configuration of a vehicle provided with a machine learning device according to a first embodiment of the present invention.

Fig. 3 is a functional block diagram of the ECU in the first embodiment.

Fig. 4 shows an example of a neural network model having a simple configuration.

Fig. 5 is a flowchart showing a control routine of disaster information transmission processing in the first embodiment of the present invention.

Fig. 6 is a flowchart showing a control routine of the learning stop process in the first embodiment of the present invention.

Fig. 7 is a diagram schematically showing the configuration of a vehicle provided with a machine learning device according to a second embodiment of the present invention.

Fig. 8 is a flowchart showing a control routine of the learning stop process in the second embodiment of the present invention.

Fig. 9 is a functional block diagram of an ECU in the third embodiment.

Fig. 10 is a flowchart showing a control routine of the vehicle determination processing in the third embodiment of the present invention.

Fig. 11 is a flowchart showing a control routine of the learning stop process in the third embodiment of the present invention.

Fig. 12 is a flowchart showing a control routine of the learning stop process in the fourth embodiment of the present invention.

Fig. 13 is a functional block diagram of an ECU in the fifth embodiment.

Fig. 14 is a flowchart showing a control routine of the learning stop process in the fifth embodiment of the present invention.

Fig. 15 is a flowchart showing a control routine of a vehicle specifying process in the sixth embodiment of the present invention.

Fig. 16 is a flowchart showing a control routine of the model transmission stop process in the seventh embodiment of the present invention.

Detailed Description

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following description, the same reference numerals are given to the same constituent elements.

< first embodiment >

First, a first embodiment of the present invention will be described with reference to fig. 1 to 6. Fig. 1 is a schematic configuration diagram of a machine learning system 1 according to a first embodiment of the present invention. The machine learning system 1 includes a server 2 and a vehicle 3.

As shown in fig. 1, the server 2 is provided outside the vehicle 3, and includes a communication interface 21, a storage device 22, a memory 23, and a processor 24. The server 2 may further include an input device such as a keyboard and a mouse, an output device such as a display, and the like. The server 2 may be configured by a plurality of computers.

The communication interface 21 is capable of communicating with the vehicle 3, and enables the server 2 to communicate with the vehicle 3. Specifically, the communication interface 21 has an interface circuit for connecting the server 2 with the communication network 5. The server 2 communicates with the vehicle 3 via the communication interface 21, the communication network 5, and the wireless base station 6. The communication interface 21 is an example of a communication device.

The storage device 22 has, for example, a Hard Disk Drive (HDD), a Solid State Drive (SSD), or an optical recording medium. The storage device 22 stores various data, such as information relating to the vehicle 3, a computer program for causing the processor 24 to execute various processes, and the like.

The memory 23 includes a semiconductor memory such as a Random Access Memory (RAM). The memory 23 stores various data and the like used when various processes are executed by the processor 24, for example.

The communication interface 21, the storage device 22, and the memory 23 are connected to the processor 24 via signal lines. The processor 24 has one or more CPUs and peripheral circuits thereof, and executes various processes. The processor 24 may further include an arithmetic circuit such as a logical operation unit or a numerical operation unit. The processor 24 is an example of a control device.

Fig. 2 is a diagram schematically showing the configuration of a vehicle 3 provided with a machine learning device according to a first embodiment of the present invention. The vehicle 3 is a vehicle capable of supplying electric power to the outside of the vehicle 3, and examples thereof include a plug-in hybrid vehicle (PHV), an Electric Vehicle (EV), and a Fuel Cell Vehicle (FCV).

As shown in fig. 2, the vehicle 3 includes a Human Machine Interface (HMI) 31, a GPS receiver 32, a map database 33, a navigation system 34, an actuator 35, a sensor 36, a communication module 37, and an electronic Control unit (ecu) 40. The HMI31, the GPS receiver 32, the map database 33, the navigation system 34, the actuator 35, the sensor 36, and the communication module 37 are communicatively connected to the ECU40 via an in-vehicle Network conforming to CAN (Controller Area Network) or the like.

The HMI31 is an input/output device that inputs and outputs information between the driver and the vehicle 3. The HMI31 includes, for example, a display for displaying information, a speaker for generating sound, an operation button or a touch panel for the driver to perform input operations, a microphone for receiving the driver's sound, and the like. The output of the ECU40 is transmitted to the driver via the HMI31, and the input from the driver is sent to the ECU40 via the HMI 31.

The GPS receiver 32 receives signals from three or more GPS satellites and detects the current position of the vehicle 3 (for example, the latitude and longitude of the vehicle 3). The output of the GPS receiver 32 is sent to the ECU 40.

The map database 33 stores map information. The ECU40 acquires map information from the map database 33.

The navigation system 34 sets a travel route of the vehicle to the destination based on the current position of the vehicle detected by the GPS receiver 32, the map information of the map database 33, an input by the driver of the vehicle, and the like. The travel route set by the navigation system 34 is sent to the ECU 40. In addition, the GPS receiver 32 and the map database 33 may also be combined with the navigation system 34.

The actuator 35 is a working member required for traveling of the vehicle 3. In the case where the vehicle 3 is a PHV, the actuator 35 includes, for example, an electric motor, a fuel injection valve, an ignition plug, a throttle valve drive actuator, an EGR control valve, and the like. The ECU40 controls the actuator 35.

The sensors 36 detect state quantities of the vehicle 3, the internal combustion engine, the battery, and the like, and include a vehicle speed sensor, an accelerator opening sensor, an air flow meter, an air-fuel ratio sensor, a crank angle sensor, a torque sensor, a voltage sensor, and the like. The output of the sensor 36 is sent to the ECU 40.

The communication module 37 is a device that enables the vehicle 3 to communicate with the outside of the vehicle 3. The communication module 37 is, for example, a data communication module (dcm) that can communicate with the communication network 5 via the wireless base station 6. In addition, as the communication module 37, a mobile terminal (for example, a smartphone, a tablet terminal, a WiFi router, or the like) may also be used.

The ECU40 includes the communication interface 41, the memory 42, and the processor 43, and executes various controls of the vehicle 3. In the present embodiment, one ECU40 is provided, but a plurality of ECUs may be provided for each function.

The communication interface 41 has an interface circuit for connecting the ECU40 to an in-vehicle network conforming to the CAN or the like standard. The ECU40 communicates with other in-vehicle devices as described above via the communication interface 41.

The memory 42 has, for example, a volatile semiconductor memory (e.g., RAM) and a nonvolatile semiconductor memory (e.g., ROM). The memory 42 stores a program executed in the processor 43, various data used when various processes are executed by the processor 43, and the like.

The processor 43 has one or more CPUs (Central Processing units) and peripheral circuits thereof, and performs various processes. The processor 43 may further include an arithmetic circuit such as a logical operation unit or a numerical operation unit. The communication interface 41, the memory 42, and the processor 43 are connected to each other via signal lines.

In the present embodiment, the ECU40 functions as a machine learning device. Fig. 3 is a functional block diagram of the ECU40 in the first embodiment. The ECU40 has a learning portion 51. The learning portion 51 is a functional block realized by causing the processor 43 of the ECU40 to execute a program stored in the memory 42 of the ECU 40.

The learning unit 51 performs processing related to learning of a machine learning model used in the vehicle 3. In the present embodiment, the learning unit 51 performs processing related to learning of the neural network model using the neural network model as the machine learning model. First, an outline of the neural network model will be described with reference to fig. 4. Fig. 4 shows an example of a neural network model having a simple configuration.

The circle marks in fig. 4 represent artificial neurons. Artificial neurons are generally referred to as nodes or units (in this specification, referred to as "nodes"). In fig. 4, L ═ 1 denotes an input layer, L ═ 2 and L ═ 3 denote hidden layers, and L ═ 4 denotes an output layer. In addition, the hidden layer is also referred to as an intermediate layer.

In FIG. 4, x₁And x₂Each node of the input layer (L ═ 1) and the output value from the node are indicated, and y indicates the node of the output layer (L ═ 4) and the output value thereof. Likewise, z₁ ^(L＝2)、z₂ ^(L＝2)And z₃ ^(L＝2)Each node representing a hidden layer (L ═ 2) and an output value, z, from that node₁ ^(L＝3)And z₂ ^(L＝3)Each node of the hidden layer (L ═ 3) and the output value from that node are represented.

At each node of the input layer, the input is directly output. On the other hand, the output value x of each node of the input layer is input to each node of the hidden layer (L ═ 2)₁And x₂At each node of the hidden layer (L ═ 2), the total input value u is calculated using the weight w and the offset b, which correspond to each other. For example, in fig. 4 z by hidden layer (L ═ 2)_k ^(L＝2)Total input value u calculated at each node represented by (k ═ 1, 2, 3)_k ^(L＝2)As shown below (M is the number of nodes of the input layer).

[ mathematical formula 1 ]

Then, the total input value u_k ^(L＝2)Transforming from z by a hidden layer (L-2) by an activation function f_k ^(L＝2)Node shown as output value z_k ^(L＝2)(＝f(u_k ^(L＝2)) ) and output. On the other hand, the output value z of each node of the hidden layer (L ═ 2) is input to each node of the hidden layer (L ═ 3)₁ ^(L＝2)、z₂ ^(L＝2)And z₃ ^(L＝2)At each node of the hidden layer (L ═ 3), a total input value u (═ Σ z · w + b) is calculated using the weight w and the offset b corresponding to each node. The total input value u is similarly transformed by an activation function, and the output value z is obtained from each node of the hidden layer (L: 3)₁ ^(L＝3)、z₂ ^(L＝3)And the output, the activation function is for example a Sigmoid function σ.

In addition, the output value z of each node of the hidden layer (L ═ 3) is input to the node of the output layer (L ═ 4)₁ ^(L＝3)And z₂ ^{(L ＝3)}At the nodes of the output layer, the total input value u (∑ z · w + b) is calculated using the respectively corresponding weights w and the deviations b, or the total input value u (∑ z · w) is calculated using only the respectively corresponding weights w. For example, at the nodes of the output layer, identity functions are used as activation functions. In this case, the total input value u calculated at the node of the output layer is directly output as the output value y from the node of the output layer.

The neural network model used in the vehicle 3 is stored in the memory 42 of the ECU40 or other storage device provided in the vehicle 3. The ECU40 causes the neural network model to output at least one output parameter by inputting a plurality of input parameters to the neural network model. At this time, as the value of each input parameter, for example, a value detected by the sensor 36 or the like or a value calculated by the ECU40 is used. By using the neural network model, a value of an appropriate output parameter corresponding to an input parameter of a prescribed value can be obtained.

In order to improve the accuracy of such a neural network model, it is necessary to learn the neural network model in advance. In the present embodiment, the ECU40 of the vehicle 3 learns the neural network model. That is, the neural network model is learned not in the server 2 but in the vehicle 3.

In the learning of the neural network model, a training data set is used which is composed of a combination of actual measurement values of a plurality of input parameters and actual measurement values (correct answer data) of at least one output parameter corresponding to the actual measurement values. Therefore, the learning unit 51 of the ECU40 creates a training data set as a process related to learning of the neural network model (hereinafter referred to as "learning related process"). Specifically, the learning unit 51 acquires measured values of a plurality of input parameters and measured values of at least one output parameter corresponding to the measured values, and combines the measured values of the input parameters and the measured values of the output parameters to create a training data set.

The actual measurement values of the input parameters and the actual measurement values of the output parameters are acquired as values detected by the sensor 36 or the like, or values calculated or determined by the ECU40, for example. The training data set created by the learning unit 51 is stored in the memory 42 of the ECU40 or other storage device provided in the vehicle 3. Further, the measured values of the input parameters used as the training data set may be normalized or normalized.

In addition, the learning unit 51 performs learning of the neural network model as the learning related process. Specifically, the learning unit 51 repeatedly updates the weight w and the deviation b in the neural network model by a known error back propagation algorithm using a large number of training data sets so that the difference between the output value of the neural network model and the actually measured value of the output parameter becomes small. As a result, the neural network model is learned, and a learned neural network model is generated. The information of the learned neural network model (the structure of the model, the weight w, the deviation b, and the like) is stored in the memory 42 of the ECU40 or in another storage device provided in the vehicle 3. By using the neural network model learned in the vehicle 3, it is possible to predict the value of the output parameter corresponding to the input parameter of the predetermined value without detecting the actual value of the output parameter by the sensor 36 or the like.

However, as described above, in the vehicle 3, the electric power stored in the battery of the vehicle 3 can be supplied to the outside of the vehicle 3. Therefore, when a power failure occurs due to a disaster, the vehicle 3 can be effectively used as a power supply source.

However, when the learning related processing is performed in the vehicle 3, the electric power necessary for learning is consumed in addition to the electric power necessary for traveling of the vehicle 3. As a result, the amount of power consumption in the vehicle 3 increases, and there is a possibility that the power necessary for disaster cannot be secured.

Therefore, in the present embodiment, when disaster information is acquired, the learning unit 51 reduces the amount of power consumption in the learning-related process as compared with the case where disaster information is not acquired. This can suppress a reduction in the amount of electric power that can be supplied from the vehicle 3 to the outside in the event of a disaster due to the learning-related process.

The disaster information includes information (location information of a disaster area, etc.) related to natural disasters (earthquakes, typhoons, volcanic eruptions, floods, etc.) and artificial disasters (power outage, etc. due to operation errors). For example, the learning unit 51 receives disaster information from the outside of the vehicle 3 to acquire the disaster information. In this case, the server 2 receives disaster information from public institutions (meteorological offices, national transportation provinces, etc.), electric power companies, and the like, and transmits the disaster information to the vehicle 3.

On the other hand, the learning unit 51 of the ECU40 stops the learning related process when the disaster information is received from the server 2. That is, the learning unit 51 stops the learning related process to reduce the amount of power consumption (set to zero) in the learning related process.

The control described above will be described below with reference to flowcharts of fig. 5 and 6. Fig. 5 is a flowchart showing a control routine of disaster information transmission processing in the first embodiment of the present invention. The present control routine is repeatedly executed by the processor 24 of the server 2 at predetermined execution intervals.

First, in step S101, the processor 24 determines whether disaster information is received. When it is determined that the disaster information has not been received, the control routine is terminated. On the other hand, if it is determined that disaster information has been received, the control routine proceeds to step S102.

In step S102, the processor 24 transmits disaster information to the vehicle 3. After step S102, the present control routine ends.

Further, disaster information may be input to the server 2 by an operator of the server 2 or the like, and the processor 24 may determine whether disaster information is input to the server 2 in step S101.

Fig. 6 is a flowchart showing a control routine of the learning stop process in the first embodiment of the present invention. The present control routine is repeatedly executed by the ECU40 of the vehicle 3 at predetermined execution intervals.

First, in step S201, the learning unit 51 determines whether disaster information is received from the server 2. When it is determined that the disaster information has not been received from the server 2, the control routine is terminated. On the other hand, when it is determined that disaster information is received from the server 2, the control routine proceeds to step S202.

In step S202, the learning unit 51 stops the learning related process. Specifically, the learning section 51 stops the creation of the training data set and the learning of the neural network model. At this time, the driver may be notified of the fact that the learning related process has stopped in order to suppress power consumption by characters or voice via the HMI 31. After step S202, the present control routine ends. In this case, the learning unit 51 restarts the learning related process when, for example, a predetermined time has elapsed, the vehicle 3 is restarted, the driver of the vehicle 3 instructs the restart of the learning related process via the HMI31, or the server 2 notifies the elimination of the disaster.

In addition, since the amount of power consumption for learning the neural network model is the largest in the learning related process, the learning unit 51 may stop only the learning of the neural network model in step S202.

The learning unit 51 may reduce the amount of power consumption in the learning related process without stopping the learning related process. In this case, the learning unit 51 reduces the amount of power consumption by, for example, reducing the frequency of creating a training data set, reducing the frequency of learning a neural network model, or slowing down the learning rate of a neural network model.

The learning unit 51 may receive disaster information directly from public institutions (meteorological offices, national transportation provinces, and the like), electric power companies, and the like, without passing through the server 2. The learning unit 51 may receive disaster information from another vehicle by vehicle-to-vehicle communication or acquire disaster information from a roadside machine by road-to-vehicle communication using the communication module 37. In these cases, the control routine of fig. 5 is omitted, and the learning unit 51 determines whether or not disaster information is received in step S201.

< second embodiment >

The configuration and control of the machine learning device of the second embodiment are basically the same as those of the machine learning device of the first embodiment except for the points described below. Therefore, the second embodiment of the present invention will be described below mainly focusing on differences from the first embodiment.

Fig. 7 is a diagram schematically showing the configuration of a vehicle 3' provided with a machine learning device according to a second embodiment of the present invention. As shown in fig. 7, the vehicle 3' further includes an exterior camera 38. The vehicle exterior camera 38 captures the periphery of the vehicle 3 'to generate a peripheral image of the vehicle 3'. For example, the exterior camera 38 is disposed in front of the vehicle 3 '(for example, a rear surface of an interior mirror in the vehicle, a front bumper, and the like) to photograph the front of the vehicle 3'. The vehicle exterior camera 38 may be a stereo camera capable of distance measurement.

In the second embodiment, the vehicle 3' detects a disaster. That is, the learning unit 51 of the ECU40 acquires disaster information by detecting a disaster. For example, the learning unit 51 determines whether or not there is a disaster around the vehicle 3' based on the surrounding image generated by the vehicle exterior camera 38. Specifically, the learning unit 51 analyzes the surrounding image by using an image recognition technique such as machine learning (neural network, support vector machine, or the like), thereby determining whether or not a disaster is present. For example, when recognizing a power failure of a traffic light, a collapse of a building, a crack in a road surface, a fallen tree, water immersion in a road, a landslide, and the like from the surrounding image, the learning unit 51 determines that a disaster has occurred around the vehicle 3'. The sensor 36 includes a gyro sensor or the like, and the learning unit 51 may detect a disaster by detecting an earthquake by the sensor 36.

Fig. 8 is a flowchart showing a control routine of the learning stop process in the second embodiment of the present invention. The present control routine is repeatedly executed by the ECU40 of the vehicle 3' at predetermined execution intervals.

First, in step S301, the learning unit 51 determines whether or not a disaster is detected around the vehicle 3'. If it is determined that no disaster is detected, the control routine is terminated. On the other hand, if it is determined that a disaster is detected, the control routine proceeds to step S302.

In step S302, the learning unit 51 stops the learning related process, similarly to step S202 of fig. 6. After step S302, the present control routine ends. In this case, the learning unit 51 restarts the learning related process when, for example, a predetermined time has elapsed, the vehicle 3 'is restarted, or the driver of the vehicle 3' instructs the restart of the learning related process via the HMI 31. In addition, the control routine of fig. 8 may be modified in the same manner as the control routine of fig. 6.

< third embodiment >

The configuration and control of the machine learning device of the third embodiment are basically the same as those of the machine learning device of the first embodiment except for the points described below. Therefore, the third embodiment of the present invention will be described below mainly focusing on differences from the first embodiment.

Fig. 9 is a functional block diagram of the ECU40 in the third embodiment. In the third embodiment, the ECU40 includes the position information acquiring unit 52 in addition to the learning unit 51. The learning unit 51 and the positional information acquisition unit 52 are functional blocks realized by the processor 43 of the ECU40 executing programs stored in the memory 42 of the ECU 40.

The positional information acquisition unit 52 acquires positional information of the vehicle 3. For example, the position information acquisition unit 52 acquires the current position of the vehicle 3 based on the output of the GPS receiver 32. The position information of the vehicle 3 is periodically transmitted from the vehicle 3 to the server 2 together with the identification information (e.g., identification number) of the vehicle 3, and is stored in the storage device 22 of the server 2.

However, even if a disaster occurs, the necessity of using the power supply of the vehicle 3 is low when the vehicle 3 travels in a place away from the disaster area. Therefore, in the third embodiment, when the supply of electric power from the vehicle 3 to the outside is predicted based on the disaster information and the positional information of the vehicle 3, the learning unit 51 reduces the amount of electric power consumption in the learning-related process as compared with the case where the supply of electric power is not predicted. This makes it possible to reduce the amount of power consumed by the learning-related processing at a more appropriate timing in preparation for power supply at the time of a disaster.

For example, the server 2 receives disaster information from public institutions (weather authorities, national transportation centers, and the like), electric power companies, and the like, and specifies a disaster area where supply of electric power from vehicles to the outside is predicted. The position information of a plurality of traveling vehicles is periodically transmitted to the server 2, and the server 2 identifies the vehicle in the disaster area by comparing the position information of the vehicle and the disaster area.

In a disaster, it is predicted that electric power is supplied from the vehicle to the outside to cope with a power failure or the like. Therefore, the server 2 transmits an instruction to stop the learning related process to the vehicle in the disaster area. Upon receiving an instruction to stop the learning related process from the server 2, the learning unit 51 of the ECU40 stops the learning related process. That is, the learning unit 51 stops the learning related process to reduce the amount of power consumption (set to zero) in the learning related process.

The control described above will be described below with reference to flowcharts of fig. 10 and 11. Fig. 10 is a flowchart showing a control routine of the vehicle determination processing in the third embodiment of the present invention. The present control routine is repeatedly executed by the processor 24 of the server 2 at predetermined execution intervals.

First, in step S401, the processor 24 determines whether disaster information is received. When it is determined that the disaster information has not been received, the control routine is terminated. On the other hand, if it is determined that disaster information has been received, the control routine proceeds to step S402.

In step S402, the processor 24 identifies a vehicle in the disaster area by comparing the location information of the disaster area included in the disaster information with the location information of the vehicle (the current location of the vehicle) stored for each vehicle.

Next, in step S403, the processor 24 transmits a stop instruction of the learning related process to the vehicle determined in step S402. After step S403, the present control routine ends.

Further, the disaster information may be input to the server 2 by an operator of the server 2 or the like, and the processor 24 may determine whether the disaster information is input to the server 2 in step S401.

Fig. 11 is a flowchart showing a control routine of the learning stop process in the third embodiment of the present invention. The present control routine is repeatedly executed by the ECU40 of the vehicle 3 at predetermined execution intervals.

First, in step S501, the learning unit 51 determines whether or not a stop instruction of the learning related process is received from the server 2. When it is determined that the instruction to stop the learning related process has not been received, the control routine is terminated. On the other hand, if it is determined that the instruction to stop the learning related process has been received, the control routine proceeds to step S502.

In step S502, the learning unit 51 stops the learning related process, similarly to step S202 of fig. 6. After step S502, the present control routine ends. In this case, the learning unit 51 restarts the learning related process when, for example, a predetermined time has elapsed, the vehicle 3 is restarted, the driver of the vehicle 3 instructs the restart of the learning related process via the HMI31, or the server 2 instructs the restart of the learning related process.

In addition, since the amount of power consumption for learning the neural network model is the largest in the learning related process, the learning unit 51 may stop only the learning of the neural network model in step S502.

In addition, the processor 24 of the server 2 may transmit the power suppression instruction to the vehicle in the disaster area instead of the learning stop instruction. In this case, when receiving the power suppression instruction from the server 2, the learning unit 51 reduces the amount of power consumption in the learning related process without stopping the learning related process. For example, the learning unit 51 reduces the frequency of creating a training data set, reduces the frequency of learning a neural network model, or slows down the learning rate of a neural network model.

< fourth embodiment >

The configuration and control of the machine learning device of the fourth embodiment are basically the same as those of the machine learning device of the first embodiment except for the points described below. Therefore, the fourth embodiment of the present invention will be described below focusing on differences from the first embodiment.

In the fourth embodiment, disaster information is transmitted to the vehicle 3 instead of the server 2, and the learning unit 51 of the ECU40 acquires the disaster information. That is, the learning unit 51 receives disaster information from public institutions (meteorological offices, national transportation provinces, etc.), electric power companies, etc., and specifies a disaster area where supply of electric power from the vehicle to the outside is predicted. The learning unit 51 may receive disaster information from another vehicle by vehicle-to-vehicle communication or acquire disaster information from a roadside machine by road-to-vehicle communication using the communication module 37.

Further, the learning unit 51 reduces the amount of power consumption in the learning-related process when the vehicle 3 is located in a disaster-stricken area. Specifically, the learning unit 51 stops the learning related process when the vehicle 3 is located in the disaster area.

Fig. 12 is a flowchart showing a control routine of the learning stop process in the fourth embodiment of the present invention. The present control routine is repeatedly executed by the ECU40 of the vehicle 3 at predetermined execution intervals.

First, in step S601, the learning unit 51 determines whether disaster information is received. When it is determined that the disaster information has not been received, the control routine is terminated. On the other hand, if it is determined that disaster information has been received, the control routine proceeds to step S602.

Next, in step S602, the learning unit 51 determines whether or not the vehicle 3 is located in the disaster area based on the location information of the disaster area included in the disaster information and the current location of the vehicle 3 acquired by the location information acquiring unit 52. When it is determined that the vehicle 3 is not located in the disaster area, the control routine is terminated. On the other hand, when it is determined that the vehicle 3 is located in the disaster area, the control routine proceeds to step S603.

In step S603, the learning unit 51 stops the learning related process, similarly to step S202 of fig. 6. After step S603, the present control routine ends. In this case, the learning unit 51 restarts the learning related process when, for example, a predetermined time has elapsed, the vehicle 3 is restarted, or the driver of the vehicle 3 instructs the restart of the learning related process via the HMI 31.

In addition, since the amount of power consumption for learning the neural network model is the largest in the learning related process, the learning unit 51 may stop only the learning of the neural network model in step S603.

In step S603, the learning unit 51 may reduce the amount of power consumption in the learning related process without stopping the learning related process. In this case, the learning unit 51 reduces the frequency of creating the training data set, reduces the frequency of learning the neural network model, or slows down the learning rate of the neural network model, for example.

< fifth embodiment >

The configuration and control of the machine learning device of the fifth embodiment are basically the same as those of the machine learning device of the first embodiment except for the points described below. Therefore, the fifth embodiment of the present invention will be described below mainly focusing on differences from the first embodiment.

Fig. 13 is a functional block diagram of the ECU40 in the fifth embodiment. The ECU40 includes an output device control unit 53 in addition to the learning unit 51 and the positional information acquisition unit 52. The learning unit 51, the positional information acquisition unit 52, and the output device control unit 53 are functional blocks realized by the processor 43 of the ECU40 executing programs stored in the memory 42 of the ECU 40.

The output device control unit 53 controls an output device provided in the vehicle 3. In the present embodiment, the output device control unit 53 controls the HMI 31. The HMI31 is an example of an output device.

As described above, the learning unit 51 reduces the amount of power consumption in the learning related process in preparation for supplying power to the outside of the vehicle 3 in the event of a disaster. However, power failure does not necessarily occur in a disaster area. In addition, in a disaster, the driver may not want to supply electric power from the vehicle 3 to the outside.

Therefore, the output device control unit 53 confirms the permission to stop the learning related process to the driver of the vehicle 3 via the HMI 31. Further, the learning unit 51 can stop the learning related process in preparation for the power supply at the time of the disaster based on the intention of the driver by stopping the learning related process when the driver permits the stop of the learning related process and not stopping the learning related process when the driver does not permit the stop of the learning related process.

In the fifth embodiment, the control routine of the disaster information transmission process shown in fig. 5 is executed in the same manner as in the first embodiment, and the control routine of the learning termination process shown in fig. 14 is executed. Fig. 14 is a flowchart showing a control routine of the learning stop process in the fifth embodiment of the present invention. The present control routine is repeatedly executed by the ECU40 of the vehicle 3 at predetermined execution intervals.

First, in step S701, the output device control unit 53 determines whether disaster information is received from the server 2. When it is determined that the disaster information has not been received from the server 2, the control routine is terminated. On the other hand, when it is determined that disaster information is received from the server 2, the control routine proceeds to step S702.

In step S702, the output device control unit 53 confirms the permission to stop the learning related process to the driver of the vehicle 3 via the HMI 31. For example, the output device control unit 53 confirms the permission to the driver by characters or voice via the HMI 31.

Next, in step S703, the learning unit 51 determines whether the driver permits the stop of the learning related process based on the input to the HMI31 by the driver. If it is determined that the driver does not permit the stop of the learning related process, the present control routine is ended. On the other hand, if it is determined that the driver permits the stop of the learning related process, the present control routine proceeds to step S704.

In step S704, the learning unit 51 stops the learning related process, similarly to step S202 of fig. 6. After step S704, the present control routine ends. In this case, the learning unit 51 restarts the learning related process when, for example, a predetermined time has elapsed, the vehicle 3 is restarted, the driver of the vehicle 3 instructs the restart of the learning related process via the HMI31, or the server 2 instructs the restart of the learning related process. In addition, the control routine of fig. 14 may be modified in the same manner as the control routine of fig. 6.

< sixth embodiment >

The configuration and control of the machine learning device according to the sixth embodiment are basically the same as those of the machine learning device according to the third embodiment except for the points described below. Therefore, the sixth embodiment of the present invention will be described below centering on differences from the third embodiment.

As described above, in the third embodiment, the amount of power consumption in the learning-related process is reduced in the vehicle in the disaster area. On the other hand, even if the vehicle is not located in the disaster area when a disaster occurs, the destination of the vehicle may be the disaster area. In this case, it is considered that electric power is supplied from the vehicle to the outside after the vehicle reaches the disaster area. Therefore, in the sixth embodiment, when the supply of electric power from the vehicle 3 to the outside is predicted based on the disaster information and the destination of the vehicle 3, the learning unit 51 reduces the amount of electric power consumption in the learning-related process as compared with the case where the supply of electric power is not predicted. This makes it possible to suppress power consumption in an appropriate vehicle in consideration of the destination.

The destination of the vehicle 3 is input by the driver of the vehicle 3, and is stored in the navigation system 34 or the like, for example. The positional information acquisition unit 52 acquires the stored destination of the vehicle 3. The destination of the vehicle 3 is periodically transmitted from the vehicle 3 to the server 2 together with identification information (e.g., an identification number) of the vehicle 3, and is stored in the storage device 22 of the server 2.

In the sixth embodiment, the control routine of the vehicle determination process shown in fig. 15 is executed, and the control routine of the learning stop process of fig. 11 is executed as in the third embodiment. Fig. 15 is a flowchart showing a control routine of a vehicle specifying process in the sixth embodiment of the present invention. The present control routine is repeatedly executed by the processor 24 of the server 2 at predetermined execution intervals.

First, in step S801, the processor 24 determines whether disaster information is received. When it is determined that the disaster information has not been received, the control routine is terminated. On the other hand, if it is determined that disaster information has been received, the control routine proceeds to step S802.

In step S802, the processor 24 identifies a vehicle having a disaster area as a destination by comparing the location information of the disaster area included in the disaster information with the destination of the vehicle stored for each vehicle.

Next, in step S803, the processor 24 transmits a stop instruction of the learning related process to the vehicle determined in step S802. After step S803, the present control routine ends.

Further, the disaster information may be input to the server 2 by an operator of the server 2 or the like, and the processor 24 may determine whether the disaster information is input to the server 2 in step S801. In step S803, the processor 24 may transmit the power suppression instruction to the vehicle addressed to the disaster area instead of the learning stop instruction. Alternatively, the current location and destination of the vehicle may be periodically transmitted to the server 2, and the processor 24 may transmit the stop instruction or the power suppression instruction of the learning related process to the vehicle in the disaster area and the vehicle to which the disaster area is destined.

< seventh embodiment >

The configuration and control of the machine learning device of the seventh embodiment are basically the same as those of the machine learning device of the first embodiment except for the points described below. Therefore, the seventh embodiment of the present invention will be described below focusing on differences from the first embodiment.

In the seventh embodiment, the server 2 learns the neural network model instead of the ECU40 of the vehicle 3. In other words, the server 2 functions as a machine learning device.

A training data set for learning of the neural network model is created in the plurality of vehicles and transmitted from the plurality of vehicles to the server 2. The processor 24 of the server 2 performs learning of the neural network model using a large amount of training data sets, and transmits the learned neural network model to the vehicle via the communication interface 21. At this time, the vehicle 3 consumes power when receiving and storing the learned neural network model from the server 2.

Therefore, when the disaster information is acquired, the processor 24 of the server 2 stops the transmission of the learned neural network model to the vehicle 3. This can suppress a reduction in the amount of power that can be supplied from the vehicle 3 to the outside in the event of a disaster in the vehicle 3.

Fig. 16 is a flowchart showing a control routine of the model transmission stop process in the seventh embodiment of the present invention. The present control routine is repeatedly executed by the processor 24 of the server 2 at predetermined execution intervals.

First, in step S901, the processor 24 determines whether disaster information is received. When it is determined that the disaster information has not been received, the control routine is terminated. On the other hand, if it is determined that disaster information has been received, the control routine proceeds to step S902.

In step S902, the processor 24 stops the transmission of the learned neural network model to the vehicle 3. After step S902, the present control routine ends. In this case, the processor 24 restarts transmission of the learned neural network model when a predetermined time has elapsed or when the disaster has been eliminated, for example.

Further, the disaster information may be input to the server 2 by an operator of the server 2 or the like, and the processor 24 may determine whether the disaster information is input to the server 2 in step S901.

Further, step S402 of fig. 10 may be executed between steps S901 and S902, and in step S902, the processor 24 may stop transmission of the learned neural network model to the vehicle specified in step S402. Similarly, step S802 of fig. 15 may be executed between steps S901 and S902, and in step S902, the processor 24 may stop transmission of the learned neural network model to the vehicle identified in step S802.

In the seventh embodiment, as in the above-described embodiments, control for stopping the learning related processing in the vehicle 3 or control for reducing the power consumption amount of the learning related processing in the vehicle 3 may be executed. In this case, the learning unit 51 stops the creation of the training data set and the transmission of the training data set to the server 2 when the learning related process is stopped, and reduces the generation frequency of the training data set or the transmission frequency of the training data set to the server 2 when the power consumption amount of the learning related process is reduced.

< other embodiment >

While the preferred embodiments of the present invention have been described above, the present invention is not limited to these embodiments, and various modifications and changes can be made within the scope of the claims.

For example, as the input parameters and the output parameters of the neural network model, various parameters can be used according to the object of the neural network model (an internal combustion engine, a motor, a battery, and the like). The sensor for detecting the actual measurement value of the input parameter or the actual measurement value of the output parameter is selected according to the kind of the input parameter and the output parameter.

The machine learning model that is learned by the vehicle 3 or the server 2 may be a machine learning model other than a neural network such as a random forest, a k-neighborhood method, or a support vector machine.

The above embodiments may be implemented in any combination. For example, in the case of combining the second embodiment and the fifth embodiment, in the control routine of the learning stop process of fig. 8, steps S702 to S704 of fig. 14 are executed instead of step S302. In the case of combining the third embodiment and the fifth embodiment, steps S702 to S704 in fig. 14 are executed in the control routine of the learning stop process in fig. 11 instead of step S502. In the case of combining the fourth embodiment and the fifth embodiment, steps S702 to S704 in fig. 14 are executed in the control routine of the learning stop process in fig. 12 instead of step S603.

In the case where the fourth embodiment and the sixth embodiment are combined, in step S602 of the control routine of the learning stop process in fig. 12, the learning unit 51 determines whether or not the destination of the vehicle 3 is a disaster area. In the case where the fifth embodiment and the sixth embodiment are combined, when the control routine of the vehicle specifying process of fig. 15 and the control routine of the learning/stopping process of fig. 11 are executed in the sixth embodiment, steps S702 to S704 of fig. 14 are executed instead of step S502 of fig. 11.

Description of the reference symbols

2 Server

21 communication interface

24 processor

3 vehicle

40 ECU

51 learning part