JP2020051305A - On-vehicle control device and management server - Google Patents

Abstract

To provide an on-vehicle control device capable of improving prediction accuracy of a plant model using a statistical method.SOLUTION: A vehicle control device comprises: a data storage section 150 which stores control data representing an operation quantity of a plant while a vehicle is in operation, in association with sensor data indicating behavior of the plant in response to the operation quantity, in a storage section of the on-vehicle control device; a data transmission section 160 which transmits the control data and the sensor data to a management server 200 as leaning data of a plant model; and a data update section 170 which receives data about the plant model, subjected to machine learning through the management server 200 using the learning data, from the management server 200 and updates the data about the plant model stored in the on-vehicle control device.SELECTED DRAWING: Figure 5

Description

  The present disclosure relates to an in-vehicle control device and a management server.

  As a technique for controlling each part of a vehicle, a technique called model predictive control (MPC) has been known. The model predictive control is a control method of predicting a future behavior using a model of a control target, and determining an operation amount of the control target while solving an optimization problem at each time.

  FIG. 1 is a diagram simply showing a general system configuration for implementing model predictive control. FIG. 2 is a timing chart showing a general control mode of the model predictive control.

  In this type of control system, the model prediction controller T1 predicts the behavior of a plant using a prediction model (hereinafter, also referred to as a "plant model") that models a control target (hereinafter, also referred to as a "plant"). And an optimizing unit T1a that determines an operation amount of the plant T2 so that the behavior of the plant T2 is optimized based on the prediction result of the predicting unit T1b.

  When the target value γ is set, the model prediction control controller T1 controls the plant T2 to bring the output of the plant T2 closer to the target value γ while considering the current behavior (y (t)) of the plant T2. The mode (ie, the operation amount) is determined. Specifically, the prediction unit T1b uses the plant model to determine, for various control modes of the plant T2 for satisfying the set target value γ, the plant in the prediction section (time t to t + N) from the present time onward. A simulation of the behavior of T2 (that is, the plant output prediction yp) is executed. Then, based on the result obtained from the simulation, the optimization unit T1a controls the plant T2 in the control section (time t to t + 1) such that the predetermined loss function that defines the behavior of the plant T2 is minimized. (That is, the operation amount) is determined, and the plant T2 is controlled in the control mode. Hereinafter, the model prediction controller T1 performs the same processing after time t + 1.

  In a vehicle, in particular, a model of an internal combustion engine (hereinafter, referred to as an “engine”) is modeled to predict the engine model so that the fuel efficiency of the engine is improved and the amount of NOx exhausted from the engine is reduced. Various modes of performing control have been studied (for example, see Patent Document 1).

JP 2017-223229 A

Jaehoon Lee, et al. "Deep Neural Networks as Gaussian Processes", arXiv: 1711.00165v3 [stat.ML], 3 Mar 2018, ("URL: https://arxiv.org/abs/1711.00165")

  Meanwhile, a plant model (hereinafter, also referred to as a “statistical model”) constructed by machine learning based on data collected by a test bench is known. In this type of plant model, prediction accuracy may be degraded depending on running conditions due to a shortage of data points or the like. In addition, in practice, the mode of use of the vehicle (for example, the traveling environment of the vehicle) differs for each vehicle, and a uniform plant model does not always mean that the plant model is optimized.

  However, in such a plant model, the prediction accuracy may be deteriorated depending on the traveling state due to a shortage of data points or the like. In addition, in practice, the mode of use of the vehicle (for example, the traveling environment of the vehicle) differs for each vehicle, and a uniform plant model does not always mean that the plant model is optimized.

  The present disclosure has been made in view of the above problems, and has as its object to provide an in-vehicle control device and a management server that can improve the prediction accuracy of a plant model using a statistical model.

The main disclosure for solving the above-mentioned problems is as follows.
A vehicle model that performs model predictive control of the plant based on a plant model of a plant of the vehicle configured by a learning device and road information of a section in front of a traveling position of the vehicle so that the vehicle travels at a target speed. A control device,
Data storage for storing in a storage unit of the in-vehicle control device the control data indicating the operation amount of the plant when the vehicle is running and the sensor data indicating the behavior of the plant with respect to the operation amount in association with each other. Department and
A data transmitting unit that transmits the control data and the sensor data to a management server as learning data of the plant model,
Receiving, from the management server, data on the plant model on which machine learning has been performed using the learning data in the management server, and updating the data on the plant model stored in the on-vehicle control device. A data update unit,
It is an in-vehicle control device provided with.

In other aspects,
A vehicle model that performs model predictive control of the plant based on a plant model of a plant of the vehicle configured by a learning device and road information of a section in front of a traveling position of the vehicle so that the vehicle travels at a target speed. A management server communicatively connected to the control device,
Data acquisition for acquiring control data indicating an operation amount of the plant when the vehicle is running from the vehicle and sensor data indicating behavior of the plant with respect to the operation amount as learning data of the plant model. Department and
When receiving the learning data from the vehicle, using the learning data, a learning processing unit that performs machine learning on the plant model of the vehicle,
An update command unit that transmits data related to the learned plant model to the vehicle,
It is a management server provided with.

  According to the vehicle-mounted control device according to the present disclosure, it is possible to improve the prediction accuracy of a plant model using a statistical model.

Diagram showing a simplified system configuration for realizing model predictive control Timing chart showing general control mode of model predictive control The figure which shows the structure of the vehicle which concerns on one Embodiment. FIG. 1 is a diagram illustrating a configuration of a vehicle ECU according to an embodiment. FIG. 1 is a diagram illustrating a system configuration for implementing model prediction control of a vehicle ECU according to an embodiment. The figure which shows the relationship between the loss regarding the fuel consumption of an engine, the loss regarding the urea water injection amount in the urea water injection device, and the loss regarding the NOx amount in the exhaust gas when the vehicle is going to run at the target vehicle speed. 4 is a flowchart showing the operation of the vehicle ECU according to one embodiment. 5 is a flowchart illustrating an operation of the management server according to the embodiment.

  Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the present specification and the drawings, components having substantially the same function are denoted by the same reference numerals, and redundant description is omitted.

[Vehicle configuration]
Hereinafter, an example of a configuration of the vehicle 1 according to the embodiment will be described with reference to FIG.

  FIG. 3 is a diagram illustrating a configuration of the vehicle 1 according to the present embodiment.

  The vehicle 1 according to the present embodiment includes an engine 11, a clutch 12, a transmission 13, a driving wheel 14, a braking device 20, an exhaust purification device 30, a target vehicle speed setting device 40, a current location information acquisition device 50, various sensors 60, and a vehicle. An ECU (Electronic Control Unit) 100 is provided.

  The engine 11 burns fuel to generate a driving force for running the vehicle 1. The engine 11 is, for example, a diesel engine, and generates an arbitrary driving force by controlling a fuel injection amount and the like.

  The engine 11 has an injector for injecting fuel and an engine ECU (not shown) for controlling valve timing of intake and exhaust valves. Then, the engine 11 controls the fuel injection pressure, the fuel injection timing, the fuel injection amount, the valve timing of the intake / exhaust valve, and the like based on the control signal from the vehicle ECU 100 so as to obtain predetermined driving characteristics. I do.

  The clutch 12 is interposed between the engine 11 and the transmission 13, and connects the output shaft of the engine 11 and the input shaft of the transmission 13 so as to be able to be connected and disconnected.

  The transmission 13 changes the speed of rotation input from the engine 11 and outputs the rotation to the drive wheels 14. As the transmission 13, for example, a belt-type continuously variable transmission is used. The clutch 12 and the transmission 13 each have an actuator (not shown) for controlling these states. Then, the actuator is controlled based on a control signal from vehicle ECU 100.

  The drive wheels 14 cause the vehicle 1 to run by driving force transmitted via the transmission 13.

  The braking device 20 decelerates the vehicle 1 by applying a resistance force due to frictional resistance to the drive wheels 14 and the like.

  The exhaust gas purification device 30 is, for example, a device that purifies NOx in exhaust gas discharged from the engine 11, and includes an SCR (Selective Catalytic Reduction) 30 a provided in an exhaust pipe of the engine 11 and an SCR 30 a. It is configured to include a urea water injection device 30b that injects urea water in an upstream exhaust pipe (also referred to as a post-treatment device).

  In the exhaust gas purification device 30, the SCR 30a adsorbs, for example, ammonia that is obtained by hydrolyzing the urea water supplied from the urea water injection device 30b, and selectively reduces and purifies NOx in the exhaust gas by the adsorbed ammonia. I do. The amount of urea water injected from the urea water injection device 30b into the exhaust pipe is controlled by a control signal from the vehicle ECU 100.

  The target vehicle speed setting device 40 sets a target vehicle speed during automatic traveling of the vehicle 1 in the vehicle ECU 100. The target vehicle speed setting device 40 includes, for example, an information input interface such as a display with a touch panel arranged on a dashboard (not shown) in a driver's seat, and receives setting of a target vehicle speed from a driver. Note that the target vehicle speed set by the target vehicle speed setting device 40 may be set as a target vehicle speed width.

  The current position information acquisition device 50 acquires information indicating the current position of the vehicle 1 and outputs the information to the vehicle ECU 100. The current location information acquisition device 50 is, for example, a satellite positioning system (GPS) receiver.

  The various sensors 60 are provided to detect the state of each part of the vehicle 1 and the like. Specifically, the various sensors 60 include an engine rotation sensor that detects the number of rotations of the crankshaft of the engine 11, a cam position sensor that detects the rotation phase of the camshaft of the engine 11, and a pressure detection inside the combustion chamber of the engine 11. Pressure sensor, a torque sensor for detecting an output torque of the engine 11, an accelerator opening sensor for detecting an accelerator opening, a cooling water temperature sensor for detecting a temperature of a cooling water flowing in a cooling water circulation path (not shown), and a temperature of exhaust gas. An exhaust temperature sensor for detecting, an A / F (air-fuel ratio) sensor for detecting an air-fuel ratio of exhaust gas, a temperature sensor for detecting a catalyst temperature of the SCR 30a, and the like are provided. Incidentally, these sensors can be realized by known sensors.

  The detection values detected by these various sensors 60 (hereinafter, also referred to as “sensor data”) are transmitted to the vehicle ECU 100.

  The vehicle ECU 100 performs overall control of each unit of the vehicle 1, and includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), an input port, an output port, and the like. Have been.

  The vehicle ECU 100 according to the present embodiment is configured to automatically control each part of the vehicle 1 so that the vehicle 1 automatically runs at the target speed set in the target vehicle speed setting device 40 (also referred to as an auto cruise mode). ). When automatically driving the vehicle 1 at the target speed, the vehicle ECU 100 predicts the behavior of the vehicle 1 in a predetermined section in front of the traveling position of the vehicle 1 based on road information, vehicle information, and the like, and Each part of the vehicle 1 such as the engine 11 is controlled so as to be optimal from the viewpoint of the NOx amount in the exhaust gas discharged from the engine 11 (details will be described later).

  The vehicle ECU 100 includes various parts of the vehicle 1 (the engine 11, the clutch 12, the transmission 13, the braking device 20, the exhaust gas purification device 30, the target vehicle speed setting device 40, the current position information acquisition device 50, and various sensors 60, etc.). They are interconnected by a network and exchange necessary data and control signals with each other. Dotted arrows in FIG. 3 indicate signal paths.

[Configuration of ECU]
Next, an example of the configuration of the vehicle ECU 100 (corresponding to the “vehicle-mounted control device” of the present invention) according to the present embodiment will be described with reference to FIGS.

  FIG. 4 is a diagram illustrating a configuration of the vehicle ECU 100 according to the present embodiment. FIG. 5 is a diagram showing a system configuration for implementing model prediction control of the vehicle ECU 100 according to the present embodiment.

  The vehicle ECU 100 according to the present embodiment includes a road information acquisition unit 110, a vehicle information acquisition unit 120, a driving state determination unit 130, a control unit 140, a data storage unit 150, a data transmission unit 160, and a data update unit 170. I have.

  The road information acquisition unit 110 acquires road information in a predetermined section (hereinafter, referred to as a “forward prediction section”) in front of the position where the vehicle 1 currently travels.

  Specifically, based on the current location of the vehicle 1 acquired by the current location information acquisition device 50, the road information acquisition unit 110 determines a section away from the current position of the vehicle 1 by a predetermined distance to the front. It is determined as a forward prediction section. Then, the road information acquisition unit 110 acquires the road information of the forward prediction section from map data or an external server stored in advance.

  The road information is, for example, data describing slope information, curves, information on the presence / absence of a signal, and the like at each point on the road in association with the horizontal position (latitude / longitude information or the like) of each point on the road.

  The vehicle information acquisition unit 120 acquires vehicle information necessary for causing the vehicle 1 to automatically travel. The vehicle information acquisition unit 120 includes, for example, sensor data of various sensors 60, control data of each unit of the vehicle 1 at the present time (for example, a fuel injection amount in the engine 11, a urea water injection amount in the urea water injection device 30b), and a target. The target vehicle speed data and the like set in the vehicle speed setting device 40 are acquired.

  The driving state determination unit 130 is optimal while the vehicle 1 is running at the target vehicle speed set in the target vehicle speed setting device 40, from the viewpoint of the fuel efficiency of the engine 11, the amount of NOx in the exhaust gas discharged from the engine 11, and the like. The control mode of the engine 11 and the like to be used is determined. The operating state determination unit 130 is a functional unit that performs the model prediction control described with reference to FIGS.

  More specifically, the target vehicle speed, road information of the predicted forward section, and sensor data of the various sensors 60 are input to the driving state determination unit 130. Then, the operation state determination unit 130 predicts the behavior of each part of the vehicle 1 from the current time to a time before the current time by using the control model 132a and the plant model 132b while considering these pieces of information. An optimal control mode of the engine 11 and the like in the control section is determined.

  The operating state determination unit 130 includes a prediction unit 132 and an optimization unit 131.

  The optimization unit 131 sets a plurality of control modes of the engine 11 and the like for the vehicle 1 to run at the target speed in the forward prediction section, and causes the prediction unit 132 to execute a simulation for each of the plurality of control modes. Then, based on the simulation result output from the prediction unit 132, the optimization unit 131 calculates a loss in each of the control modes of the plurality of patterns using a predetermined loss function, and the engine 11 or the like that minimizes the loss. Is determined as the optimal operation of the engine 11.

  The control mode (for example, the operation amount) of the engine 11 or the like for which the optimization unit 131 causes the prediction unit 132 to execute the simulation is, for example, the state information, the target speed, and the road information acquisition unit 110 of each part of the vehicle 1 at the present time. The required driving force or the like estimated from the road information of the acquired forward prediction section is set as a reference.

  At this time, the state information of each unit at the present time acquired by the optimization unit 131 includes, for example, the operation amount of the control unit 140 (the injection system control unit 142, the SCR control unit 143, the transmission control unit 144) at the present time, which will be described later. , And sensor data from various sensors 60 that detect the actual state of the plant (for example, the engine 11, the exhaust gas purification device 30, and the transmission 13).

The optimization unit 131 calculates the loss of the entire forward prediction section for each control mode of the engine 11 or the like, for example, using the loss function of the following equation (1).

  Equation (1) represents the fuel consumption of the engine 11 at each timing when traveling in the forward prediction section, the NOx amount in the exhaust gas discharged from the engine 11, and the urea water injection amount in the urea water injection device 30b in unit time. As the loss per hit, the integrated value up to N time ahead when traveling in the forward prediction section is determined as the loss in the entire forward prediction section. Then, the optimization unit 131 determines the control mode of the engine 11 or the like that minimizes the loss in Expression (1) as the control mode of the engine 11 or the like at the next timing.

  FIG. 6 shows the relationship between the loss related to the fuel efficiency of the engine 11, the loss related to the urea water injection amount in the urea water injection device 30b, and the loss related to the NOx amount in the exhaust gas when the vehicle 1 is going to run at the target vehicle speed. FIG. Normally, there is a trade-off between the loss related to the fuel efficiency of the engine 11 (or the loss related to the urea water injection amount) and the loss related to the NOx amount in the exhaust gas, as shown in FIG. Equation (1) is set in consideration of such a trade-off relationship.

  The prediction unit 132 simulates each of the control modes of the plurality of patterns of the engine 11 and the like set in the optimization unit 131, using the control model 132a of each unit in the vehicle 1 and the plant model 132b of the engine 11 and the like in the vehicle 1. Execute

  The control model 132a models the configuration of the control unit 140 (the control unit 142 of the injection system of the engine 11, the control unit 143 of the urea water injection of the SCR 30a, and the control unit 144 of the transmission 13). Control model 132aa (for example, a model for defining the fuel injection amount and fuel injection timing into the combustion chamber) of the injection system, and a control model 132ab for the urea water injection in the SCR 30a (for example, defining the urea water injection amount and the urea water injection timing). And a control model 132ac of the transmission 13 (for example, a model that defines the transmission ratio of the transmission 13).

  The plant model 132b is a model of the configuration of the plant (the engine 11, the exhaust gas purification device 30, and the transmission 13). For example, the plant model 132ba of the engine 11 (for example, the engine speed, the engine output, and the exhaust gas Model that defines the state), the plant model 132bb of the exhaust treatment device 30 (for example, a model that defines the purification rate of the amount of NOx contained in the exhaust gas), and the plant model 132bc of the power transmission mechanism of the vehicle 1 (for example, the power transmission efficiency is reduced). Specified model).

  The plant model 132ba of the engine 11 is based on, for example, state information (eg, engine output, engine speed, exhaust gas temperature, exhaust gas flow rate) based on the control mode of the injection system control model 132aa of the engine 11 mainly. , And the amount of NOx in the exhaust gas). Further, the plant model 132bb of the exhaust gas purification device 30 outputs its state information (for example, the purification rate or the like in the SCR 30a) mainly based on the control mode of the control model 132ab of the urea water injection in the SCR 30a. Further, the plant model 132bc of the power transmission mechanism of the vehicle 1 has state information (for example, power transmission efficiency from the engine 11 to the driving wheels 14, and power efficiency based on the control mode of the control model 132ac of the transmission 13). (E.g., acceleration of the vehicle 1).

  The prediction unit 132 applies, for example, the control mode of the engine 11 or the like set by the optimization unit 131 (for example, the mode relating to the operation amount or the control timing) to the control model 132a, and the state in the plant model 132b at that time. Calculate information. The state information in the plant model 132b calculated by the prediction unit 132 is, for example, information on the state of the plant model 132b at each timing in the forward prediction section.

  Then, the prediction unit 132 outputs the state information in the plant model 132b to the optimization unit 131. Thereby, the optimization unit 131 calculates the loss for each of the control modes of the engine 11 and the like in a plurality of patterns, and determines the optimum control mode.

  The control unit 140 controls each unit based on the control mode of the engine 11 and the like determined by the operating state determination unit 130 (optimization unit 131). Then, under the control of the control unit 140, the plant (here, the engine 11, the exhaust purification device 30, and the transmission 13) operates.

  The control unit 140 acquires, for example, the control mode of the engine 11 and the like determined by the operation state determination unit 130 and deploys the control mode to each unit, the control unit 142 that controls the injection system of the engine 11, the urea water The control unit 143 controls the urea water injection of the injection device 30b and the control unit 144 controls the transmission 13.

  Note that the model prediction control itself according to the present embodiment is the same as a known method, and a description thereof will be omitted.

  Here, a method of constructing the plant model 132b according to the present embodiment will be described.

  The plant model 132b according to the present embodiment is configured by a learning device that has been subjected to machine learning (also referred to as a statistical model). As a learning device used for the plant model 132b, for example, a known Gaussian process model (for example, see Non-Patent Document 1) is used. However, the plant model 132b may be constituted by any known statistical model, and a neural network, a Bayes model, an SVM (Support Vector Machine), or the like may be used. Further, only a part of the plant model 132b may be constituted by a learning device, and the other part may be constituted by a physical model such as a fluid equation.

  Before the vehicle 1 is shipped, the plant model 132b mounted on the vehicle 1 is subjected to machine learning based on data collected on a test bench. However, in the plant model 132b, as described above, the prediction accuracy may be deteriorated depending on the traveling state due to a shortage of data points or the like. In addition, in practice, the usage mode of the vehicle 1 (for example, the traveling environment of the vehicle 1) differs for each vehicle 1, and in a uniform plant model, the plant model 132b is not necessarily optimized. I can not say. For example, a plant model of a vehicle used in a cold region cannot secure high prediction accuracy unless it is adapted to the cold region.

  Therefore, the plant model 132b according to the present embodiment is individually machine-learned after shipment, based on the state information of the vehicle 1 when the vehicle 1 is actually used. Specifically, the control data indicating the operation amount of the engine 11 and the like when the engine 11 is operating, and the sensor data indicating the behavior of the engine 11 and the like with respect to the operation amount perform machine learning on the plant model 132b. This is the learning data D1 to be applied. The data storage unit 150, the data transmission unit 160, and the data update unit 170 are functional units for realizing the update of the plant model 132b.

  When the engine 11 is operating, the data storage unit 150 sequentially stores the control data indicating the operation amount of the engine 11 and the like and the sensor data indicating the behavior of the engine 11 and the like with respect to the operation amount in the learning data D1. Is stored in a storage unit (for example, RAM) of the vehicle ECU 100. In other words, the learning data D1 is a data set relating to an operation amount of each unit of the plant of the engine 11 and an output value of each unit of the plant when the plant is controlled by the operation amount.

  Here, the control data includes, for example, the accelerator opening, the fuel injection amount in the engine 11, the urea water injection amount in the urea water injection device 30b, and the like. The sensor data includes the NOx amount in exhaust gas discharged from the engine 11 (sensor value of the NOx sensor), the output of the engine 11 (sensor value of the engine rotation sensor, sensor value of the torque sensor), and the like. It is.

  The data transmission unit 160 transmits control data indicating the operation amount of the engine 11 and the like and sensor data indicating the behavior of the engine 11 and the like with respect to the operation amount to the management server 200 as learning data D1. At this time, the data transmission unit 160 transmits the learning data D1 to the management server 200 together with the identification information of the vehicle 1.

  The data updating unit 170 receives, from the management server 200, data related to the plant model 132b after machine learning has been performed using the learning data D1 in the management server 200, and stores the plant model 132b stored in the vehicle ECU 100. Update the data of.

  The timing at which the data transmitting unit 160 and the data updating unit 170 perform data communication with the management server 200 may be arbitrary. For example, the data transmission unit 160 transmits the learning data D1 (that is, control data and sensor data) obtained during the traveling to the management server 200 when the vehicle 1 travels and the driving ends. I do. Further, for example, the data updating unit 170 receives the data of the plant model 132b from the management server 200 when the traveling of the vehicle 1 is started (for example, at the time of key-on).

  FIG. 7 is a flowchart illustrating the operation of the vehicle ECU 100 according to the present embodiment.

  First, vehicle ECU 100 (vehicle information acquisition unit 120) acquires a target vehicle speed from target vehicle speed setting device 40 (step S1). Next, vehicle ECU 100 (road information acquisition unit 110, vehicle information acquisition unit 120) acquires vehicle information and road information (step S2). Next, the vehicle ECU 100 (the driving state determination unit 130) executes a simulation of the driving state of the engine 11 and the like for various control modes (step S3). Next, the vehicle ECU 100 (the driving state determining unit 130) determines the driving state of the engine 11 and the like based on the simulation result of step S3 (step S4). Next, vehicle ECU 100 (control unit 140) controls engine 11 and the like in the operating state of engine 11 and the like determined in step S4 (step S5). Next, the vehicle ECU 100 (data storage unit 150) acquires the control data determined in step S4 and the sensor data of the engine 11 and the like detected when the control of the engine 11 and the like is performed in step S5. (Step S6). Then, vehicle ECU 100 repeatedly executes the processing of steps S1 to S6.

  The management server 200 includes a database 201, a data acquisition unit 202, a learning processing unit 203, and an update command unit 204.

  The database 201 stores data of the plant model 132b of the vehicle 1 in association with the identification information of the vehicle 1. The database 201 stores data of the plant model 132b separately for each of the plurality of vehicles 1 to be managed.

  The data acquisition unit 202 acquires the learning data D1 of the plant model from the vehicle ECU 100. When receiving the learning data D1 from the vehicle 1, the learning processing unit 203 performs machine learning on the plant model 132b of the vehicle 1 using the learning data D1. The update command unit 204 transmits data on the learned plant model 132b to the vehicle 1.

  Note that the learning process of the machine learning in the management server 200 is the same as a known method, and a description thereof will be omitted.

  FIG. 8 is a flowchart illustrating the operation of the management server 200 according to the present embodiment.

  The management server 200 waits for the learning data D1 to be transmitted from the vehicle 1 (step S11: NO). When the management server 200 receives the learning data D1 from the vehicle 1 (step S11: YES), the management server 200 matches the learning data D1 based on the identification information of the vehicle 1 attached to the learning data D1. The data of the first plant model 132b is acquired from the database 201, and machine learning is performed on the data of the plant model 132b using the learning data D1 (step S12). Then, the management server 200 updates the plant model of the vehicle 1 stored in the database 201 with the data of the plant model 132b that has been subjected to the machine learning, and also updates the data of the plant model 132b that has been learned for the vehicle 1. Is transmitted (step S13).

  Thus, vehicle ECU 100 (data updating section 170) updates the data of plant model 132b stored in its own storage section.

[effect]
As described above, the vehicle ECU (in-vehicle control device) 100 according to the present embodiment includes the control data indicating the operation amount of the plant (the engine 11, the exhaust gas purification device 30, and the transmission 13) when the vehicle 1 is running. A data storage unit 150 that stores, in a storage unit of the vehicle ECU 100, sensor data indicating behavior of the plant (the engine 11, the exhaust gas purification device 30, the transmission 13) with respect to the manipulated variable, and the control data and the sensor data. From the management server 200 to the data transmission unit 160 which transmits the data as the plant model learning data D1 to the management server 200, and the data relating to the plant model 132b that has been machine-learned by the management server 200 using the learning data D1. The plant model 132 received and stored in the vehicle ECU 100 And it includes a data updating unit 170 for updating the data.

  Therefore, according to the vehicle ECU 100 according to the present embodiment, when the vehicle 1 is used, the plant model 132b can be updated at any time, so that the plant model 132b is used as the usage mode of the vehicle 1 ( Typically, it can be optimized according to the driving environment of the vehicle 1). For example, the plant model 132b of a vehicle used in a cold region will adapt to a cold region.

  According to the control system of the vehicle 1 according to the present embodiment, the prediction accuracy of the plant model 132b including the statistical model (learning device) depends on the data amount of the learning data D1 and the accuracy of the learning data D1. This is particularly useful in that traveling data obtained from an actual use mode of the vehicle 1 can be used as learning data D1 for updating the plant model 132b of the vehicle 1.

  Further, according to the vehicle ECU 100 according to the present embodiment, more travel data (ie, learning data D1) than experimental data (ie, learning data D1) used in the development stage when constructing the plant model 132b is used. , In real time. In this way, it is possible to supplement traveling data in the extrapolated area, which is not obtained from the experimental data, as learning data D1 for updating the plant model 132b.

(Other embodiments)
The present invention is not limited to the above embodiment, and various modifications can be made.

  In the above-described embodiment, as an example of the operation state determining unit 130, a method using a loss function when determining the control mode of the plant has been described. However, the method by which the operating state determination unit 130 determines the control mode of the plant is not limited to this method. For example, the optimization unit 131 sets a reference trajectory in a prediction section based on the state information and the target value at the present time, and when the prediction unit 132 simulates the behavior of the plant in various control modes. The control mode approaching the reference trajectory may be determined as the optimal solution.

  In the above-described embodiment, an example in which the engine 11 is a plant has been described as an example of the model prediction control of the vehicle ECU 100. However, when an electric vehicle is applied as the vehicle 1, a motor may be used as a target of the model predictive control instead of the engine 11.

  In the above-described embodiment, as an example of the configuration of the vehicle ECU 100, functions of the road information acquisition unit 110, the vehicle information acquisition unit 120, the driving state determination unit 130, the control unit 140, the data transmission unit 160, and the data update unit 170 Has been described as being realized by one computer, but may of course be realized by a plurality of computers.

  As mentioned above, although the specific example of this invention was demonstrated in detail, these are only illustrations and do not limit a claim. The technology described in the claims includes various modifications and changes of the specific examples illustrated above.

  According to the vehicle-mounted control device according to the present disclosure, it is possible to improve the prediction accuracy of a plant model using a statistical model.

DESCRIPTION OF SYMBOLS 1 Vehicle 11 Engine 12 Clutch 13 Transmission 14 Driving wheel 20 Braking device 30 Exhaust purification device 30a SCR
30b Urea water injection device 40 Target vehicle speed setting device 50 Current location information acquisition device 60 Various sensors 100 Vehicle ECU
110 road information acquisition unit 120 vehicle information acquisition unit 130 driving state determination unit 131 optimization unit 132 prediction unit 140 control unit 150 data storage unit 160 data transmission unit 170 data update unit 200 management server 201 database 202 data acquisition unit 203 learning processing unit 204 Update command section

Claims (6)

A vehicle model that performs model predictive control of the plant based on a plant model of a plant of the vehicle configured by a learning device and road information of a section in front of a traveling position of the vehicle so that the vehicle travels at a target speed. A control device,
Data storage for storing in a storage unit of the in-vehicle control device the control data indicating the operation amount of the plant when the vehicle is running and the sensor data indicating the behavior of the plant with respect to the operation amount in association with each other. Department and
A data transmitting unit that transmits the control data and the sensor data to a management server as learning data of the plant model,
Receiving, from the management server, data on the plant model on which machine learning has been performed using the learning data in the management server, and updating the data on the plant model stored in the on-vehicle control device. A data update unit,
An in-vehicle control device comprising: The plant includes an engine mounted on the vehicle, and an exhaust gas purification device that purifies exhaust gas discharged from the engine.
The in-vehicle control device according to claim 1. An operation state determination unit that determines an operation amount of the engine so as to optimize a balance between fuel efficiency of the engine and an amount of NOx in exhaust gas of the engine using the plant model,
The on-vehicle control device according to claim 2. The control data includes a fuel injection amount in the engine, and data on an accelerator opening of the vehicle,
The sensor data includes an amount of NOx in exhaust gas of the engine,
The vehicle-mounted control device according to claim 2 or 3. The plant model is stored in the management server in association with identification information of the vehicle.
The in-vehicle control device according to any one of claims 1 to 4. A vehicle model that performs model predictive control of the plant based on a plant model of a plant of the vehicle configured by a learning device and road information of a section in front of a traveling position of the vehicle so that the vehicle travels at a target speed. A management server communicatively connected to the control device,
Data acquisition for acquiring control data indicating an operation amount of the plant when the vehicle is running from the vehicle and sensor data indicating behavior of the plant with respect to the operation amount as learning data of the plant model. Department and
When receiving the learning data from the vehicle, using the learning data, a learning processing unit that performs machine learning on the plant model of the vehicle,
An update command unit that transmits data related to the learned plant model to the vehicle,
Management server comprising:

Images