Machine learning apparatus

Abstract

To enable the efficient generation of trained models by utilizing a plurality of datasets acquired under different environments.SOLUTION: A machine learning apparatus comprises: an acquisition unit for acquiring a first dataset including one or more explanatory variables and one or more response variables collected in a first environment, and a second dataset including the same explanatory variables and response variables as the first dataset collected in a second environment; a pre-processing unit for performing pre-processing to set to zero the values of specific explanatory variables that are predetermined as the explanatory variables of the second data set that have a large deviation from the values of the explanatory variables of the first data set; and a learning processing unit for performing machine learning using training data including the first dataset and the pre-processed second dataset.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a machine learning device. [Background technology]

[0002] In recent years, trained models generated by machine learning are used to control the damping force in vehicle suspension systems that can adjust the damping force according to the road surface, driving conditions, etc. As machine learning for generating such trained models, a technology that uses model-based transfer learning that reuses an already generated trained model (Patent Document 1) and a technology that selects data to be used for machine learning from data on multiple vehicles (Patent Document 2) have been disclosed. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2022-21107 [Patent Document 2] JP 2021-174433 A Summary of the Invention [Problem to be solved by the invention]

[0004] One machine learning technique is instance-based transfer learning, which uses multiple data sets (data sets showing the relationship between explanatory variables and target variables) collected under different environments as training data. It is believed that this technique can efficiently generate a trained model for controlling a suspension system. However, between multiple data sets collected under different environments, the values ​​of explanatory variables may diverge significantly due to differences in the environments. If multiple data sets with such large deviations in explanatory variables are used simultaneously as training data (for machine learning of one model), the learning efficiency may decrease.

[0005] The present invention has been made in consideration of the above, and provides a machine learning device that can efficiently generate a trained model by using multiple datasets collected under different environments. [Means for solving the problem]

[0006] One embodiment of the present invention is a machine learning device that generates a trained model used to control a vehicle suspension system through machine learning, and includes an acquisition unit that acquires a first dataset including one or more explanatory variables and one or more objective variables collected in a first environment, and a second dataset collected in a second environment including the same explanatory variables and objective variables as the first dataset, a pre-processing unit that performs pre-processing to set to zero the values ​​of specific explanatory variables that are predetermined as explanatory variables of the second dataset that deviate greatly from the values ​​of the explanatory variables of the first dataset, and a learning processing unit that performs machine learning using teacher data including the first dataset and the second dataset after pre-processing. Effect of the Invention

[0007] According to the machine learning device of the present invention, when machine learning is performed using multiple data sets as training data, the influence of explanatory variables that have a large deviation between multiple data sets collected under different environments is eliminated, making it possible to efficiently generate a trained model using multiple data sets collected under different environments. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram illustrating an example of a hardware configuration of a machine learning device according to an embodiment. [Diagram 2] FIG. 2 is a block diagram illustrating an example of a functional configuration of the machine learning device according to the embodiment. [Diagram 3] FIG. 3 is a block diagram illustrating an example of an overall flow of processing in the machine learning device of the embodiment. [Figure 4] FIG. 4 is a block diagram illustrating an example of a flow of pre-processing according to the embodiment. [Diagram 5] FIG. 5 is a diagram illustrating an example of a data structure of the second data set according to the embodiment. [Figure 6] FIG. 6 is a diagram conceptually illustrating an example of the learning process according to the embodiment. [Figure 7] FIG. 7 is a flowchart illustrating an example of processing by the machine learning device according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Exemplary embodiments of the present invention are disclosed below. The configurations of the embodiments described below, and the actions, results, and effects brought about by the configurations are merely examples. The present invention can be realized with configurations other than those disclosed in the following embodiments, and it is possible to obtain at least one of various effects based on the basic configurations and derivative effects.

[0010] 1 is a block diagram showing an example of a hardware configuration of a machine learning device 1 according to an embodiment. The machine learning device 1 is a device that executes machine learning to generate a trained model for controlling the damping force of a vehicle suspension system.

[0011] The machine learning device 1 illustrated here has a configuration similar to that of a general-purpose computer, and includes a calculation device 11, a storage device 12, a user I / F (Interface) 13, and a communication I / F .

[0012] The arithmetic device 11 is configured using, for example, a central processing unit (CPU), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), etc., and executes a predetermined arithmetic process according to a program stored in the storage device 12. The storage device 12 includes a read only memory (ROM), a random access memory (RAM), a storage, etc., and is configured using, for example, a dynamic random access memory (DRAM), a static random access memory (SRAM), a solid state drive (SSD), and a hard disk drive (HDD). The user I / F 13 is a device that enables the reception of operations from a user (such as a person who performs machine learning) and the output of information to the user, and may be, for example, a keyboard, a pointing device, a touch panel mechanism, a display, a speaker, a microphone, etc. The communication I / F 14 is a device for establishing communication with an external device via an appropriate computer network. The arithmetic device 11, the storage device 12, the user I / F 13, and the communication I / F 14 are connected via a data bus.

[0013] 1 is merely an example, and the hardware configuration of the machine learning device 1 is not limited to the above. For example, the machine learning device 1 may be a system in which multiple computers operate in cooperation with each other.

[0014] 2 is a block diagram showing an example of a functional configuration of the machine learning device 1 of the embodiment. The machine learning device 1 of the embodiment includes an acquisition unit 101, a preprocessing unit 102, a setting unit 103, and a learning processing unit 104. These functional units are realized by cooperation between hardware elements and software elements (programs, etc.) of the machine learning device 1. Furthermore, at least one of these functional units may be configured by dedicated hardware (circuits, etc.).

[0015] The acquiring unit 101 acquires a first dataset collected in a first environment and a second dataset collected in a second environment different from the first environment. The first dataset includes one or more explanatory variables and one or more objective variables. The second dataset includes the same explanatory variables and objective variables as the first dataset.

[0016] The explanatory variables are variables corresponding to input values ​​to the trained model, and may be, for example, the sprung acceleration of each wheel, the wheel speed of each wheel, the damping force of the shock absorber of each wheel, the longitudinal acceleration applied to the vehicle body, the lateral acceleration applied to the vehicle body, the steering angle, the accelerator opening, the brake operation amount, etc. The objective variables are variables corresponding to output values ​​(estimated values) from the trained model, and may be, for example, the stroke speed and stroke displacement amount of the suspension device of each wheel, etc. The suspension system mounted on the vehicle can determine the target value of the damping force of the shock absorber of each wheel and the drive current value to each shock absorber, etc., based on the stroke speed and stroke displacement amount estimated by the trained model.

[0017] The combination of the first environment and the second environment should be appropriately determined according to the specifications of the vehicle in which the suspension system controlled by the trained model is mounted. The first environment may be, for example, an environment in which the vehicle is actually driven on a road surface such as a test course. The second environment may be, for example, an environment in which the wheels of a stopped vehicle are vibrated in a specific direction (e.g., vertical direction) by a vibration generating device.

[0018] The preprocessing unit 102 performs preprocessing to set the values ​​of predetermined specific explanatory variables among the explanatory variables of the second data set to zero. The specific explanatory variables are explanatory variables that are predetermined to have a large deviation from the values ​​of the explanatory variables of the first data set.

[0019] The specific explanatory variable should be appropriately set according to the combination of the first environment and the second environment. For example, if the first environment is an environment in which a vehicle runs on a road surface and the second environment is an environment in which the wheels of a vehicle in a stopped state are vibrated by a vibration generator, the specific explanatory variable may be, for example, the wheel speed. This is because the wheels do not rotate in the second environment, and therefore there is a large discrepancy between the wheel speed value in the first data set collected in the first environment and the wheel speed value in the second data set collected in the second environment. Note that the specific explanatory variable is not limited to the wheel speed.

[0020] The setting unit 103 enables the specific explanatory variable to be arbitrarily set by a user operation. For example, the setting unit 103 enables an operation to change the specific explanatory variable to an explanatory variable other than the wheel speed or to add an explanatory variable other than the wheel speed via the user I / F 13 constituting the machine learning device 1.

[0021] The learning processing unit 104 performs machine learning to generate a trained model (adjust parameters of the trained model) using training data including the first data set and the preprocessed second data set. At this time, the learning processing unit 104 may readjust all parameters of the trained model using the training data.

[0022] Fig. 3 is a block diagram showing an example of an overall flow of processing in the machine learning device 1 of the embodiment. As shown in Fig. 3, preprocessing is performed on the second data set 202 to generate a preprocessed second data set 211 in which the values ​​of specific explanatory variables are changed to zero, and teacher data 221 including the first data set 201 and the preprocessed second data set 211 is generated. Then, a trained model 251 is generated by machine learning using the teacher data 221.

[0023] Fig. 4 is a block diagram showing an example of a flow of pre-processing according to an embodiment. As shown in Fig. 4, first, the second data set 202 is classified into a target variable 301 and explanatory variables 302, and then the explanatory variables 302 are classified into normal explanatory variables 311 and specific explanatory variables 312. The normal explanatory variables 311 are explanatory variables other than the specific explanatory variables 312 among the multiple explanatory variables 302.

[0024] Then, a zero replacement process is performed on the specific explanatory variable 312. The zero replacement process is a process of setting the value of the specific explanatory variable 312 to zero. For example, if the specific explanatory variable 312 is a wheel speed, the value of the wheel speed (measured in the second environment) included in the second data set 202 is changed to zero.

[0025] After that, preprocessed explanatory variables 302' are generated, which include the normal explanatory variables 311 and the specific explanatory variables after the 0 replacement process 312. Then, a preprocessed second data set 211 is generated, which includes the response variable 301 and the preprocessed explanatory variables 302'.

[0026] FIG. 5 is a diagram showing an example of the data structure of the second data set 202 of the embodiment. The explanatory variables 302 of the second data set 202 illustrated here include the sprung acceleration, damping force, and wheel speed for each of the four wheels of the vehicle. In FIG. 5, FR indicates the right front wheel, FL indicates the left front wheel, RR indicates the right rear wheel, and RL indicates the left rear wheel. Note that the data configuration of the explanatory variables 302 is not limited to this. For example, the explanatory variables 302 may include variables other than the sprung acceleration, damping force, and wheel speed, and may include, for example, longitudinal acceleration, lateral acceleration, steering angle, accelerator opening, brake operation amount, etc. Also, here, the explanatory variables 302 include the wheel speeds of all four wheels, but the data configuration may include only the wheel speed of a specific wheel (for example, wheel speed_RR, etc.).

[0027] The objective variable 301 includes the stroke speed and stroke displacement amount for each of the four wheels. Note that the data configuration of the objective variable 301 is not limited to this, and the objective variable 301 may include variables other than the stroke speed and stroke displacement amount, or may include only either the stroke speed or the stroke displacement amount.

[0028] In the explanatory variables 302 of the second data set 202 illustrated here, the sprung acceleration and the damping force are classified as normal explanatory variables 311, and the wheel speed is classified as a specific explanatory variable 312. The value of the wheel speed as the specific explanatory variable 312 is changed to zero by the zero replacement process as the preprocessing described above.

[0029] FIG. 6 is a diagram conceptually illustrating an example of processing in the learning processing of the embodiment. As shown in FIG. 6, the teacher data 221 used in the learning processing of the embodiment includes a first data set 201 collected under a first environment and a preprocessed second data set 211 collected under a second environment in which the value of a specific explanatory variable 312 having a large deviation from the explanatory variable of the first data set 201 is changed to zero. Then, all parameters of the trained model 251 are readjusted by machine learning (instance-based transfer learning) using such teacher data 221. In this way, by readjusting all parameters rather than only some parameters of the trained model 251 (for example, specific parameters such as an output layer), it is possible to effectively improve the estimation accuracy of the trained model 251 by using multiple data sets collected under different environments.

[0030] 7 is a flowchart showing an example of processing by the machine learning device 1 of the embodiment. In step S101, the acquisition unit 101 acquires the first data set 201 and the second data set 202. In step S102, the preprocessing unit 102 separates the specific explanatory variables 312 from the second data set 202. Then, in step S103, the preprocessing unit 102 executes preprocessing (zero replacement processing) on ​​the specific explanatory variables 312. As a result, a preprocessed second data set 211 in which the value of the specific explanatory variable 312 (e.g., wheel speed) is changed to zero is generated.

[0031] After that, in step S104, the preprocessing unit 102 generates training data 221 including the first data set 201 and the preprocessed second data set 211. Then, in step S105, the learning processing unit 104 performs machine learning using the training data 221 and adjusts parameters of the trained model.

[0032] As described above, according to the machine learning device 1 of this embodiment, when machine learning is performed using multiple data sets as training data, the influence of explanatory variables that have a large deviation between multiple data sets collected under different environments is eliminated. This makes it possible to efficiently generate a trained model using multiple data sets collected under different environments.

[0033] In the above embodiment, the explanatory variables 302 of the first data set 201 and the explanatory variables 302 of the second data set 202 include sprung acceleration and wheel speed, the first data set 201 is collected when the vehicle is driven on a road surface, the second data set 202 is collected when the wheels of the vehicle in a stopped state are vibrated by a vibration generator, and the specific explanatory variables 312 include wheel speed. This makes it possible to efficiently generate a trained model for controlling the suspension system of a vehicle using a data set collected when the vehicle is actually driven and a data set collected when the vehicle is stopped.

[0034] Furthermore, in the above embodiment, the learning processing unit 104 readjusts all the parameters of the trained model 251 by using the teacher data 221. In this manner, by readjusting all the parameters rather than adjusting only some of the parameters of the trained model, it is possible to effectively improve the estimation accuracy of the trained model.

[0035] Furthermore, in the above embodiment, the machine learning device 1 includes a setting unit 103 that allows a user to arbitrarily set the specific explanatory variables 312. This makes it possible to appropriately set the specific explanatory variables 312 depending on the environment in which the dataset is collected, etc.

[0036] A program for causing a computer (such as the arithmetic device 11) to realize the functions of the machine learning device 1 of the above embodiment may be configured to be provided by recording it on a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, or a digital versatile disk (DVD) in the form of an installable or executable file.

[0037] The program may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. The program may be provided or distributed via a network such as the Internet.

[0038] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims. [Explanation of symbols]

[0039] Reference Signs List 1...machine learning device, 101...acquisition unit, 102...preprocessing unit, 103...setting unit, 104...learning processing unit, 201...first data set, 202...second data set, 211...second data set after preprocessing, 221...teaching data, 251...trained model, 301...objective variable, 302...explanatory variable, 312...specific explanatory variable

Claims

1. A machine learning device that generates a trained model used to control a suspension system of a vehicle by machine learning, an acquisition unit that acquires a first dataset including one or more explanatory variables and one or more objective variables collected in a first environment, and a second dataset that is collected in a second environment and includes the same explanatory variables and objective variables as the first dataset; a pre-processing unit that performs pre-processing to set to zero a value of a specific explanatory variable that is predetermined as a variable that is largely different from the values ​​of the explanatory variables of the first dataset among the explanatory variables of the second dataset; A learning processing unit that performs the machine learning using teacher data including the first dataset and the second dataset after preprocessing; A machine learning device comprising:

2. The explanatory variables of the first data set and the explanatory variables of the second data set include sprung acceleration and wheel speed, the first data set is collected while the vehicle is driven on a road surface; The second data set is collected when a wheel of the vehicle in a stationary state is vibrated by a vibration generator; The specific explanatory variables include the wheel speed. The machine learning device according to claim 1 .

3. The learning processing unit readjusts all parameters of the trained model using the teacher data. The machine learning device according to claim 1 or 2.

4. a setting unit that enables the specific explanatory variables to be arbitrarily set by a user operation; The machine learning device according to claim 1 or 2, further comprising:

Images