JP7432321B2 - Control program, control method and control device - Google Patents

Description

The present invention relates to a control program, a control method , and a control device .

BACKGROUND ART Conventionally, there is a technique for relearning a model when the estimation accuracy of a model generated by machine learning has decreased (see, for example, Patent Document 1).

International Publication No. 2016/152053

However, with the conventional technology, there is a possibility that the estimation accuracy of the model cannot be appropriately ensured. Specifically, in the conventional technology, relearning is performed using all input data, so the estimation accuracy of the model for some input data does not necessarily improve.

The present invention has been made in view of the above, and an object of the present invention is to provide a control device and a control method that can appropriately maintain model estimation accuracy.

In order to solve the above-mentioned problems and achieve the purpose, a control device according to an embodiment includes a storage section, a determination section, and a setting section. The storage unit stores parameters to be applied to a model generated by machine learning according to an application range of the parameters. The determination unit determines to which application range input data input to the model corresponds. The setting unit sets the parameters to the model based on the application range determined by the determination unit.

According to the present invention, the estimation accuracy of the model can be maintained appropriately.

FIG. 1A is a diagram showing an example of mounting a control device. FIG. 1B is a diagram showing an overview of the control method. FIG. 2 is a block diagram of the control device. FIG. 3 is a diagram showing an example of the parameter DB. FIG. 4 is a diagram illustrating an example of processing by the search unit. FIG. 5 is a flowchart (part 1) showing the processing procedure executed by the control device. FIG. 6 is a flowchart (Part 2) showing the processing procedure executed by the control device.

Hereinafter, a control device and a control method according to an embodiment will be described with reference to the accompanying drawings. Note that the present invention is not limited to the embodiments described below.

First, an overview of a control device and a control method according to an embodiment will be described using FIG. 1A and FIG. 1B. FIG. 1A is a diagram showing an example of mounting a control device. FIG. 1B is a diagram showing an overview of the control method.

As shown in FIG. 1A, the control device 1 according to the embodiment is mounted on a vehicle 100. The control device 1 generates output data Do by inputting input data Di based on sensor information input from an on-vehicle sensor of the vehicle 100 into a model M. Here, the model M has layers called an input layer, an intermediate layer (hidden layer), and an output layer, and transmits input data Di input from the input layer to the output layer while propagating each connection path. .

At this time, output data Do corresponding to the input data Di is generated by performing arithmetic processing based on parameters set for each connection path. Note that the parameters include weight parameters, bias parameters, and the like.

By the way, as described above, the control device 1 performs arithmetic processing using the model M based on sensor information input from an on-vehicle sensor. Here, installation errors may occur for each vehicle 100 in the vehicle-mounted sensor, or parts of the vehicle 100 may deteriorate.

Therefore, it is necessary to optimize model M according to the current state of vehicle 100. As a method of optimizing the model M, for example, a method of relearning the model M using learning data corresponding to all conditions can be considered.

However, in this case, it is necessary to verify whether the model M after relearning is effective under all conditions, which is not preferable because it increases cost. Furthermore, in this case, if relearning is performed using learning data corresponding to all the conditions, there is a risk that the accuracy of output data Do for some input data Di may deteriorate before and after relearning.

Therefore, in the control method according to the embodiment, the parameters to be applied to the model M are stored according to the applicable range of the parameters, and the parameters are used depending on the input data Di.

In the example shown in FIG. 1B, the application range R1 and the application range R2 are ranges set based on the first condition, which is the engine load factor, and the second condition, which is the engine rotational speed, respectively.

The first parameter corresponds to the applicable range R1, and the second parameter different from the first parameter corresponds to the applicable range R2. In addition, below, when application range R1 and application range R2 are not distinguished, they are simply described as application range R.

In the control method according to the embodiment, it is determined which application condition the input data Di corresponds to, parameters of the model M are set according to the determination result, and the input data Di is input to the model M.

Specifically, if the input data Di is in the applicable range R1, the first parameter is set to the model M, and if the input data Di is in the applicable range R2, the second parameter is set to the model M.

In this manner, in the control method according to the embodiment, a plurality of parameters are stored according to the applicable range R of the parameter, and optimization is performed so that the model M covers the entire range using the plurality of parameters.

Therefore, in the control method according to the embodiment, when the application range R of the parameter changes due to deterioration of parts, etc., there is no need to relearn all the parameters, and only some parameters corresponding to some conditions are learned. Just re-learn.

That is, in the control method according to the embodiment, it is possible to re-learn parameters corresponding to a part of the application range using learning data of a part of the conditions. In other words, only the parameters of the conditions whose accuracy has deteriorated are changed without changing the parameters of the conditions whose accuracy was originally good.

Therefore, in the control method according to the embodiment, the estimation accuracy of the model can be maintained appropriately. In addition, in the above example, the application range R for two conditions, the first condition and the second condition, is shown, but it is not limited to this, and the number of such conditions may be one, or there may be three or more conditions. It may be.

Further, in the embodiment described above, the case where there are two applicable ranges is shown, but the present invention is not limited to this, and the applicable ranges R may be three or more.

Next, a configuration example of the control device 1 according to the embodiment will be described using FIG. 2. FIG. 2 is a block diagram of the control device 1. Note that FIG. 2 shows a control system S including the control device 1 and the server device 50. Further, it is assumed that the control system S includes a plurality of control devices 1.

Further, FIG. 2 also shows on-vehicle sensors 101. The on-vehicle sensors 101 are, for example, sensors that detect the state of the internal combustion engine of the vehicle 100. The control device 1 can determine a state such as engine misfire by inputting sensor information input from the on-vehicle sensors 101 to the model M.

Further, the server device 50 is a server that centrally manages the models M owned by each control device 1. As described later, the server device 50 provides the model M to the control device 1, and when receiving a relearning request for the model M from the control device 1, performs relearning based on the relearning request. Equipped with the function to perform

Thereafter, the server device 50 transmits the parameters, which are the learning results of the relearning, to the control device 1. Thereby, the control device 1 can optimize the parameters, and the model M can be optimized.

As shown in FIG. 2, the control device 1 according to the embodiment includes a storage section 2 and a control section 3. The storage unit 2 is realized by, for example, a semiconductor memory element such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk.

Further, as shown in FIG. 2, the storage unit 2 includes a model DB 21, a parameter DB 22, and a learning data DB 23. The model DB 21 is a database that stores models M.

The parameter DB 22 is a database that stores parameters applied to the model M. As described above, the parameter DB 22 stores a plurality of parameters depending on the applicable range of the parameters.

FIG. 3 is a diagram showing an example of the parameter DB 22. As shown in FIG. 3, in the parameter DB 22, "parameter ID", "applicable range", "weight parameter", "bias parameter", etc. are stored in association with each other.

The parameter ID is an identifier that identifies each parameter. The applicable range indicates the applicable range of the corresponding parameter. In this embodiment, the applicable range indicates the applicable range provided for the first condition and the second condition shown in FIG. 1B.

Weight parameters and bias parameters are examples of the parameters described above. These parameters are values derived by machine learning. In addition, in the example shown in FIG. 3, the weight parameters and bias parameters are schematically shown as "C01" and "D01", but specific information is written in "C01" and "D01". shall be taken as a thing. Further, in "C01" and "D01", parameters corresponding to the number of nodes of the model M are written.

Returning to the explanation of FIG. 2, the learning data DB 23 will be explained. The learning data DB 23 is a database that stores data used during relearning. As described later, the control device 1 according to the embodiment has a function of searching the applicable range of parameters, and stores data outside the applicable range in the learning data DB 23 as learning data.

The control unit 3 is a controller, and is realized by, for example, a CPU, an MPU, or the like executing various programs stored in the storage unit 2 using the RAM as a work area. Further, the control unit 3 can be realized by, for example, an integrated circuit such as an ASIC or an FPGA.

As shown in FIG. 2, the control unit 3 includes an acquisition unit 31, a determination unit 32, a setting unit 33, a calculation unit 34, a search unit 35, and an update unit 36. The acquisition unit 31 acquires sensor information input from the on-vehicle sensors 101. Note that, for example, the on-vehicle sensors 101 are sensors that are provided in the internal combustion engine of the vehicle 100 and detect the state of the internal combustion engine.

The determination unit 32 determines to which application range R the input data Di input to the model M corresponds. Specifically, the determination unit 32 refers to the applicable range of the parameter DB 22 and determines to which applicable range the above sensor information (input data) corresponds.

The setting unit 33 sets parameters based on the application range determined by the determining unit 32. Specifically, the setting unit 33 retrieves the parameters of the applicable range determined by the determining unit 32 from the parameter DB 22 and sets the parameters to the model M.

At this time, if the previous applicable range and the current applicable range are the same, the setting unit 33 may retain the parameters of the model M as they are. In other words, the setting unit 33 may set the parameters only when the applicable range R changes.

The calculation unit 34 generates output data Do by inputting the sensor information acquired by the acquisition unit 31 to the model M as input data Di. Furthermore, the calculation unit 34 can also generate the evaluation output data Do by inputting the evaluation data into the model M.

The search unit 35 searches for a parameter application range R based on output data Do obtained by inputting evaluation data into the model M. Specifically, the search unit 35 searches for a range in which the estimation accuracy of the model M exceeds a threshold value as an applicable range R, and a range in which the estimation accuracy falls below the threshold value as a non-applicable range Rn.

In the following, it is assumed that the model M is a model that generates the probability of knocking in each cylinder as the output data Do. First, the search unit 35 notifies an engine control device (not shown) of an instruction to generate evaluation data.

Specifically, after specifying the first condition and the second condition, for example, the engine control device is instructed to operate the engine under conditions where knocking occurs. This allows the search unit 35 to acquire sensor information indicating knocking as evaluation data. The search unit 35 can obtain all the evaluation data by instructing the engine control device while sequentially changing the first condition and the second condition.

Subsequently, the search unit 35 obtains the output data Do obtained by inputting the evaluation data into the model M from the calculation unit 34, and searches for the applicable range of the parameter based on the output data Do and the teacher label. .

Here, the output data Do obtained from the evaluation data should originally indicate knocking.

Therefore, when the output data Do indicates knocking, the search unit 35 can determine that the parameters are compatible, and when the output data Do does not indicate knocking, the search unit 35 can determine that the parameters are compatible. It can be determined that there is no

For example, the search unit 35 sets the condition where the probability of knocking indicated by the output data Do (hereinafter referred to as knock probability) is 80% or more as the applicable range R, and the condition where the knock probability is less than 80% as the non-applicable range Rn. Explore.

Then, the search unit 35 stores the input data Di and output data Do of the non-applicable range in the learning data DB 23. At this time, the search unit 35 does not need to search the entire range using one parameter, and may search only a part of the range.

For example, the search unit 35 searches the current appropriate range R+α using the corresponding parameter. This makes it possible to reduce the search range for one parameter, thereby improving search efficiency.

In this way, the search unit 35 can identify non-applicable ranges by searching for applicable ranges and non-applicable ranges of each parameter. In other words, it becomes possible to promptly respond to non-applicable ranges. It is assumed that the search unit 35 performs the search process at predetermined intervals.

Further, when the non-applicable range is adjacent to a plurality of applicable ranges R, the search unit 35 can perform the search again using the parameters of the plurality of applicable ranges R.

FIG. 4 is a diagram illustrating an example of processing by the search unit 35. For example, the search unit 35 searches for the non-applicable range Rn1 using only the first parameter corresponding to the applicable range R1, and for the non-applicable range Rn2, uses the first parameter and the second parameter corresponding to the applicable range R2. The non-applicable range Rn1 is searched using each of the parameters.

This is because the non-applicable range Rn1 is not adjacent to the applicable range R2, so even if the second parameter is used, it is unlikely to become the applicable range R of the second parameter. On the other hand, the non-applicable range Rn2 is adjacent to each of the applicable range R1 and the applicable range R2.

Therefore, even if the non-applicable range Rn2 changes from the first parameter applicable range R to the non-applicable range Rn, it is assumed that the non-applicable range Rn2 may become the second parameter applicable range R. In this way, the search unit 35 searches for the non-applicable range Rn using each of the parameters of the adjacent applicable ranges R, thereby making it possible to omit relearning described later.

Returning to the explanation of FIG. 2, the update unit 36 will be explained. The updating unit 36 updates the parameters of the non-applicable range Rn to parameters that have been relearned based on the evaluation data and output data Do of the non-applicable range Rn.

Specifically, the update unit 36 extracts the evaluation data of the non-applicable range Rn and the teacher label (label indicating whether or not it is actually knocking) from the learning data DB 23, and sends them to the server device 50 along with the learning request. Send.

The server device 50 performs relearning using the model M based on the evaluation data and the teacher label, and generates parameters optimized in the non-applicable range Rn.

Thereafter, the server device 50 transmits the re-learned parameters to the control device 1, and the update unit 36 updates the parameter DB 22 based on the parameters. That is, the updating unit 36 can update the applicable range of parameters and add or delete parameters.

In this way, the updating unit 36 can reduce the non-applicable range Rn by updating the parameters of the non-applicable range Rn. Further, in the control device 1 according to the present embodiment, relearning is performed based on the input data Di (evaluation data) of the non-applicable range Rn and the corresponding teacher label. Therefore, it is possible to change the parameters of the non-applicable range Rn without changing the parameters of the applicable range R.

That is, after ensuring the estimation accuracy of the model M in the applicable range R, it is possible to newly ensure the estimation accuracy of the non-applicable range Rn. Therefore, according to the control device 1 according to the embodiment, the estimation accuracy of the model can be maintained appropriately.

Next, the processing procedure executed by the control device 1 according to the embodiment will be described using FIGS. 5 and 6. 5 and 6 are flowcharts showing the processing procedure executed by the control device 1.

First, the normal control procedure will be explained using FIG. 5. As shown in FIG. 5, first, when the control device 1 acquires the input data Di (step S100), it determines the application range R corresponding to the input data Di (step S101).

Next, the control device 1 sets the parameters of the model M based on the application range R determined in step S101 (step S102), inputs input data to the model M (step S103), and ends the process. do.

Next, a series of processes associated with parameter update processing will be described using FIG. 6. As shown in FIG. 6, the control device 1 determines whether it is update timing (step S111). Here, the control device 1 can determine the update timing using, for example, a predetermined cycle or the timing at which a component is replaced as a trigger, and the update timing as a trigger.

When the control device 1 determines that it is the search timing in step S111 (step S111, Yes), it acquires evaluation data (step S112) and performs a search process (step S113).

Subsequently, the control device 1 determines whether or not there is a non-applicable range Rn as a result of the search process (step S114), and if there is a non-applicable range Rn (step S114, Yes), the control device 1 determines whether or not there is a non-applicable range Rn. At the same time, a learning request is transmitted to the server device 50 (step S115).

After that, the control device 1 acquires the re-learned parameters from the server device 50 (step S116), updates the parameter DB 22 based on the acquired parameters (step S117), and ends the process.

As described above, the control device 1 according to the embodiment includes the storage section 2, the determination section 32, and the setting section 33. The storage unit 2 stores parameters to be applied to the model M generated by machine learning according to the application range R of the parameters. The determination unit 32 determines to which application range R the input data input to the model M corresponds.

The setting unit 33 sets parameters for the model M based on the application range R determined by the determining unit 32. Therefore, according to the control device 1 according to the embodiment, the estimation accuracy of the model M can be maintained appropriately.

Incidentally, in the embodiment described above, a case has been described in which the control device 1 is mounted on the vehicle 100, but the present invention is not limited to this. That is, the present invention can be applied to various devices including personal computers.

Further, in the above-described embodiment, a case has been described in which the server device 50 relearns parameters in the non-applicable range, but the present invention is not limited to this. That is, the control device 1 (updating unit 36) may perform relearning.

Further advantages and modifications can be easily deduced by those skilled in the art. Therefore, the broader aspects of the invention are not limited to the specific details and representative embodiments shown and described above. Accordingly, various changes may be made without departing from the spirit or scope of the general inventive concept as defined by the appended claims and their equivalents.

1 Control device 31 Acquisition unit 32 Judgment unit 33 Setting unit 34 Calculation unit 35 Search unit 36 Update unit M model

Claims (5)

A control program that is executed by a computer,
Normal processing of acquiring parameters according to the applicable range of input data to be input to the model from a storage unit and setting the acquired parameters to the model;
A range in which the estimation accuracy of the model based on output data obtained by inputting evaluation data to the model exceeds a threshold value is defined as the applicable range, and a range in which the estimation accuracy is less than or equal to the threshold value is searched as a non- applicable range. , updating the parameters stored in the storage unit regarding the searched non-applicable range with parameters re-learned based on the evaluation data of the non-applicable range and the teacher label of the evaluation data; or updating the parameters stored in the storage unit regarding the searched non-applicable range. A control program that performs an update/addition process that adds parameters to the . The control program according to claim 1, wherein in the update/addition process, the range searched as the non-applicable range is searched again using parameters of another applicable range adjacent to the range. 3. The control program according to claim 1, wherein, in the update/addition process, the evaluation data of the non-applicable range and the teacher label of the evaluation data are sent to a relearning execution destination to request relearning. A control method executed by a computer,
Normal processing of acquiring parameters according to the applicable range of input data to be input to the model from a storage unit and setting the acquired parameters to the model;
A range in which the estimation accuracy of the model based on output data obtained by inputting evaluation data into the model exceeds a threshold is defined as the applicable range, and a range in which the estimation accuracy is less than or equal to the threshold is searched as a non- applicable range. , updates the parameters of the non-applicable range stored in the storage unit, or updates the parameters to the storage unit with the parameters re-learned based on the evaluation data of the non-applicable range and the teacher label of the evaluation data. A control method including an update/addition process that performs addition. Normal processing of acquiring parameters according to the applicable range of input data to be input to the model from a storage unit and setting the acquired parameters to the model;
A range in which the estimation accuracy of the model based on output data obtained by inputting evaluation data to the model exceeds a threshold value is defined as the applicable range, and a range in which the estimation accuracy is less than or equal to the threshold value is searched as a non- applicable range. , update the parameters of the non-applicable range stored in the storage unit with the parameters re-learned based on the evaluation data of the non-applicable range and the teacher label of the evaluation data, or update the parameters of the non-applicable range stored in the storage unit. A control device that performs update and addition processing that performs additions.

Images