Machine learning device

The machine learning device addresses the challenge of training data variations by correcting score discrepancies in AI models, enabling autonomous sensory evaluations and reducing the time needed to set vehicle adaptation constants.

JP2024159226APending Publication Date: 2024-11-08SUBARU CORP
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2023075077
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-04-28
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing AI models require extensive man-hours to set vehicle adaptation constants due to variations in training data, which affect their ability to perform sensory evaluations accurately.

Method used

A machine learning device that acquires and groups acceleration waveforms, corrects score variations using predefined rules, and trains an AI model to predict comfort levels during vehicle operation, reducing the need for manual evaluation.

Benefits of technology

The device enables accurate and efficient setting of vehicle adaptation constants by correcting score variations, allowing the AI model to perform sensory evaluations autonomously, thereby reducing the required man-hours.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024159226000001_ABST
    Figure 2024159226000001_ABST
Patent Text Reader

Abstract

To reduce man-hours required to set the adaptation constant of a vehicle.SOLUTION: A machine learning device comprises one or more processors and one or more memories connected to the processors. The processor executes processing of cooperating with a program included in the memories and acquiring multiple types of acceleration waveforms when the vehicle is made to travel, acquiring multiple types of teacher data being scores obtained by evaluating each of the multiple types of acceleration waveforms, grouping the multiple types of acceleration waveforms on the basis of the shapes of the acceleration waveforms, and correcting the scores of the grouped acceleration waveforms on the basis of a correction rule.SELECTED DRAWING: Figure 8
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

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

[0002] Conventionally, in a vehicle control program, an adaptation constant is set to compensate for differences in vehicle characteristics for each vehicle model. For example, this adaptation constant is set by having an experienced driver drive the vehicle and performing a sensory evaluation so that the acceleration during vehicle driving is not uncomfortable. Setting this adaptation constant requires several months of work.

[0003] Patent Document 1 discloses an information processing device equipped with a machine learning model (hereinafter referred to as an AI model) that inputs vehicle acceleration information and estimates the driving behavior of a driver. If such an AI model can be used to perform sensory evaluation during vehicle driving instead of an experienced driver, the man-hours required for setting the adaptation constants can be reduced. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2020 / 035996 Summary of the Invention [Problem to be solved by the invention]

[0005] By the way, training data is necessary for AI model learning. For an AI model to perform sensory evaluation in place of an experienced driver, various acceleration waveforms of the vehicle and various evaluation and numerical score data are required as training data. However, if there is variation in this training data, the AI ​​model cannot learn correctly, and as a result, there is a problem that it is not possible to reduce the man-hours required for setting the vehicle's adaptation constants.

[0006] In view of the above problems, the present invention aims to provide a machine learning device that can reduce the amount of work required to set adaptation constants for a vehicle. [Means for solving the problem]

[0007] In order to solve the above problems, the machine learning device of the present invention includes one or more processors and one or more memories connected to the processors, and the processor works in cooperation with a program contained in the memory to perform processes including acquiring multiple types of acceleration waveforms when a vehicle is traveling, acquiring multiple types of teacher data which are scores evaluating each of the multiple types of acceleration waveforms, grouping the multiple types of acceleration waveforms based on the shapes of the acceleration waveforms, and correcting the scores of the grouped acceleration waveforms based on a correction rule. Effect of the Invention

[0008] According to the present invention, it is possible to reduce the number of steps required to set the adaptation constants for a vehicle. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram showing the configuration of a machine learning system according to this embodiment. [Diagram 2] FIG. 2 is a block diagram showing an example of the functional configuration of the machine learning device according to this embodiment. [Diagram 3] FIG. 3 is a graph showing an example of an acceleration waveform, which is input data, and teacher data, which is a score obtained by evaluating the acceleration waveform. [Figure 4] FIG. 4 is a graph showing acceleration waveforms classified into group A. [Diagram 5] FIG. 5 is a graph showing acceleration waveforms classified into group B. [Figure 6] FIG. 6 is a graph showing acceleration waveforms classified into group C. [Figure 7]FIG. 7 is a graph showing a state after the correction unit corrects the teacher data shown in FIG. [Figure 8] FIG. 8 is a graph showing an acceleration waveform of group D, which is different from groups A, B, and C classified by the grouping unit. [Figure 9] FIG. 9 is a flowchart showing the flow of processing in the machine learning device according to this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] A preferred embodiment of the present invention will be described in detail below with reference to the accompanying drawings. The dimensions, materials, and other specific values ​​shown in the embodiment are merely examples for facilitating understanding of the invention, and do not limit the present invention unless otherwise specified. In this specification and drawings, elements having substantially the same functions and configurations are given the same reference numerals to avoid repeated explanations, and elements not directly related to the present invention are not shown.

[0011] 1 is a schematic diagram showing the configuration of a machine learning system 100 according to the present embodiment. The machine learning system 100 includes a vehicle 200 and a machine learning device 300. The vehicle 200 is, for example, an engine vehicle provided with an engine as a driving source for traveling. Note that the vehicle 200 may be a hybrid vehicle provided with an engine and a motor as driving sources for traveling, or may be an electric vehicle provided with a motor as a driving source for traveling.

[0012] In addition, in this embodiment, an example will be described in which the vehicle 200 is a four-wheeled private automobile. However, the present invention is not limited to this example, and the vehicle 200 may be various automobiles, such as a commercial vehicle such as a bus, truck, or taxi, a special vehicle such as a police car, a fire engine, an ambulance, a tow truck, a snowplow, or a construction vehicle, or a motorcycle.

[0013] The vehicle 200 of this embodiment includes an acceleration sensor 210. The acceleration sensor 210 measures the acceleration of the vehicle 200. The acceleration sensor 210 transmits acceleration data indicating the measured acceleration of the vehicle 200 to the machine learning device 300.

[0014] The machine learning device 300 is a device that performs a sensory evaluation of the comfort level, which is the ride quality when the vehicle 200 accelerates or decelerates, based on acceleration data of the vehicle 200 acquired from the acceleration sensor 210. The machine learning device 300 has two phases, broadly classified as a learning phase and a prediction phase.

[0015] The learning phase is a phase in which learning data is input to the machine learning device 300 and an AI model is constructed based on the learning data. Hereinafter, the AI ​​model constructed in the learning phase is referred to as a trained model.

[0016] The learning data includes acceleration data of the vehicle 200 as input data and teacher data as correct answer data. The teacher data is a score obtained by evaluating and quantifying the acceleration waveform of the acceleration data of the vehicle 200. For example, the teacher data is an integer score obtained by evaluating the acceleration waveform of the vehicle 200 when an experienced driver drives the vehicle 200 and expressing the comfort level of the vehicle 200.

[0017] In this embodiment, the learning data includes multiple types of acceleration data having multiple types of acceleration waveforms when the vehicle 200 is driven multiple times. The teacher data has a score obtained by evaluating and quantifying each of the multiple types of acceleration waveforms. The teacher data is linked to the acceleration data, and a score corresponding to each acceleration waveform is linked to the teacher data. Therefore, the learning data includes multiple types of teacher data linked to multiple types of acceleration waveforms.

[0018] The prediction phase is a phase in which no teacher data is linked to the trained model, for example, acceleration data when a driver who is not an experienced driver is input, and a score is predicted by evaluating the acceleration waveform contained in the acceleration data. In this way, the input data input in the prediction phase is different from the training data input in the training phase.

[0019] The machine learning device 300 includes an I / F unit 310, a data storage unit 320, a system bus 330, one or more processors 340, and one or more memories 350. The I / F unit 310 is an interface for acquiring acceleration data transmitted from the vehicle 200.

[0020] Data storage unit 320 is composed of a RAM, a flash memory, a HDD, etc., and stores various information necessary for the processing of processor 340 described below. For example, learning data used in the learning phase is stored in data storage unit 320. System bus 330 is a transmission path that electrically connects I / F unit 310, data storage unit 320, processor 340, and memory 350 and transmits data among them.

[0021] The processor 340 includes, for example, a CPU (Central Processing Unit). The memory 350 includes, for example, a ROM (Read Only Memory) and a RAM (Random Access Memory). The ROM is a storage element that stores programs and computation parameters used by the CPU. The RAM is a storage element that temporarily stores data such as variables and parameters used in processing executed by the CPU.

[0022] According to the machine learning device 300 of the present embodiment, in the learning phase, input data and teacher data are input to the AI ​​model, and machine learning is performed so that the score predicted from the input data approaches the teacher data, and a trained model is constructed. As a result, in the prediction phase, an acceleration waveform is input to the trained model as input data, and a score appropriately predicted by the trained model can be output. As a result, the trained model can perform sensory evaluation during vehicle driving instead of an experienced driver, and the number of steps required for setting the adaptation constant can be reduced.

[0023] However, when learning an AI model, if there is variation in the teacher data, a problem may occur in which the AI ​​model cannot be trained correctly. The variation in the teacher data is a deviation from the most reliable true value among the evaluations and quantifications of acceleration waveforms having similar shapes by veteran drivers. The variation in the teacher data may occur, for example, when different veteran drivers give different evaluations to similar acceleration data. Therefore, the machine learning device 300 of this embodiment performs a correction process to correct the variation in the teacher data in the learning phase. The correction process of the teacher data in the machine learning device 300 of this embodiment will be described in detail below.

[0024] Fig. 2 is a block diagram showing an example of a functional configuration of a machine learning device 300 according to this embodiment. For example, as shown in Fig. 2, the machine learning device 300 includes an acceleration waveform acquisition unit 300a, a teacher data acquisition unit 300b, a grouping unit 300c, a correction unit 300d, a learning unit 300e, and a predicted score acquisition unit 300f.

[0025] The processor 340 cooperates with and executes the programs contained in the memory 350 to realize various processes including the processes described below which are performed by the acceleration waveform acquisition unit 300a, the teacher data acquisition unit 300b, the grouping unit 300c, the correction unit 300d, the learning unit 300e, and the predicted score acquisition unit 300f.

[0026] The acceleration waveform acquiring unit 300a acquires an acceleration waveform when the vehicle 200 is traveling, based on the acceleration data of the vehicle 200 acquired from the acceleration sensor 210. The teacher data acquiring unit 300b acquires teacher data that is a score obtained by evaluating the acceleration waveform acquired by the acceleration waveform acquiring unit 300a.

[0027] In this embodiment, the acceleration waveform acquisition unit 300a acquires multiple types of acceleration waveforms when one or multiple veteran drivers drive the vehicle 200 multiple times, and the teacher data acquisition unit 300b acquires multiple types of teacher data in which the veteran drivers evaluate each of the multiple types of acceleration waveforms.

[0028] The grouping unit 300c groups the multiple types of acceleration waveforms acquired by the acceleration waveform acquisition unit 300a based on the shapes of the acceleration waveforms. The details of the grouping unit 300c will be described later. The correction unit 300d corrects the scores of the acceleration waveforms grouped by the grouping unit 300c based on a correction rule. The details of the correction unit 300d will be described later.

[0029] The learning unit 300e trains the AI ​​model in the above-mentioned learning phase based on the teacher data of the scores corrected by the correction unit 300d. The learning of the AI ​​model by the learning unit 300e constructs a trained model.

[0030] In the above-mentioned prediction phase, the predicted grade acquisition unit 300f inputs the acceleration waveform acquired by the acceleration waveform acquisition unit 300a into the trained model constructed by the learning unit 300e to predict the grade of the acceleration waveform. Then, the predicted grade acquisition unit 300f acquires the grade predicted and output by the trained model.

[0031] Fig. 3 is a graph showing an example of acceleration waveforms, which are input data, and teacher data, which are scores obtained by evaluating the acceleration waveforms. In Fig. 3, nine types of acceleration waveforms L1, L2, L3, L4, L5, L6, L7, L8, and L9 are shown as examples of acceleration waveforms. In Fig. 3, the dashed line indicates a score of 2, the solid line indicates a score of 3, and the dashed line indicates a score of 4.

[0032] 3, acceleration waveform L1 is evaluated as a score 2, acceleration waveform L2 is evaluated as a score 3, and acceleration waveform L3 is evaluated as a score 2. Furthermore, acceleration waveform L4 is evaluated as a score 4, acceleration waveform L5 is evaluated as a score 4, and acceleration waveform L6 is evaluated as a score 3. Furthermore, acceleration waveform L7 is evaluated as a score 3, acceleration waveform L8 is evaluated as a score 3, and acceleration waveform L9 is evaluated as a score 3.

[0033] In the learning phase, the acceleration waveform acquiring unit 300a acquires the acceleration waveforms L1 to L9. In addition, the teacher data acquiring unit 300b acquires a grade 2 linked to the acceleration waveforms L1 and L3, a grade 3 linked to the acceleration waveforms L2, L6, L7, L8, and L9, and a grade 4 linked to the acceleration waveforms L4 and L5.

[0034] In the learning phase, the grouping unit 300c groups the multiple types of acceleration waveforms acquired by the acceleration waveform acquisition unit 300a based on the shapes of the acceleration waveforms. In this embodiment, the grouping unit 300c groups the multiple types of acceleration waveforms using the K-Shape method. The K-Shape method is a shape-based clustering method targeted at time-series data, and is a method for automatically grouping similar shapes. In this embodiment, the grouping unit 300c uses the K-Shape method to group the multiple types of acceleration waveforms, but is not limited to this and may use other methods as long as they can group based on the shapes of the acceleration waveforms.

[0035] The nine types of acceleration waveforms L1 to L9 shown in FIG. 3 are classified into three groups, for example, A, B, and C, for each similar shape by the K-Shape method. The number of groups classified by this K-Shape method is determined according to the number of types of ratings. Specifically, the greater the number of types of ratings, the greater the number of groups classified. In the example shown in FIG. 3, since there are three types of ratings, the number of groups classified by the K-Shape method is determined to be three. However, this is not limited thereto, and the number of groups classified by the K-Shape method may be the same as the number of similar shapes, regardless of the number of types of ratings.

[0036] Fig. 4 is a graph showing acceleration waveforms L1 to L3 classified into group A. Fig. 5 is a graph showing acceleration waveforms L4 to L6 classified into group B. Fig. 6 is a graph showing acceleration waveforms L7 to L9 classified into group C.

[0037] As shown in Fig. 4, among the acceleration waveforms L1 to L9, the acceleration waveforms L1 to L3 having the first similar shape are classified into group A. As shown in Fig. 5, among the acceleration waveforms L1 to L9, the acceleration waveforms L4 to L6 having the second similar shape are classified into group B. As shown in Fig. 6, among the acceleration waveforms L1 to L9, the acceleration waveforms L6 to L9 having the third similar shape are classified into group C.

[0038] As shown in FIG. 4, in group A, the acceleration waveform L1 is evaluated as score 2, the acceleration waveform L2 is evaluated as score 3, and the acceleration waveform L3 is evaluated as score 2, and there is variation in the teacher data of the acceleration waveforms L1 to L3 having the first similar shape.

[0039] As shown in FIG. 5, in group B, acceleration waveform L4 is evaluated as score 4, acceleration waveform L5 is evaluated as score 4, and acceleration waveform L6 is evaluated as score 3, and there is variation in the teacher data for acceleration waveforms L4 to L6 having the second similar shape.

[0040] Therefore, the correction unit 300d corrects the scores of the acceleration waveforms grouped by the grouping unit 300c based on a correction rule in order to correct the teacher data having variations. Here, the correction rule includes a first correction rule and a second correction rule. The first correction rule is to use as a standard the score of the acceleration waveform having the largest number of acceleration waveforms evaluated with the same score among the grouped acceleration waveforms. If the scores of the acceleration waveforms are real numbers rather than integers, the correction unit 300d may round off the scores of the acceleration waveforms in the group to integers. For example, if the score of an acceleration waveform in the group is 2.312, the correction unit 300d may correct the score of the acceleration waveform to 2.

[0041] 4, of the grouped acceleration waveforms L1 to L3, the acceleration waveforms L1 and L3 are evaluated as a score of 2, and the acceleration waveform L2 is evaluated as a score of 3. The correction unit 300d corrects the score of the acceleration waveform L2 from 3 to 2 according to the first correction rule, based on the score of 2 of the acceleration waveforms L1 and L3 which have the largest number of acceleration waveforms evaluated to the same score among the grouped acceleration waveforms L1 to L3. As a result, the scores of the acceleration waveforms L1 to L3 of group A are all corrected to the same score of 2.

[0042] 5, of the grouped acceleration waveforms L4 to L6, the acceleration waveforms L4 and L5 are evaluated as a score of 4, and the acceleration waveform L6 is evaluated as a score of 3. The correction unit 300d corrects the score of the acceleration waveform L6 from 3 to 4 according to the first correction rule, based on the score of 4 of the acceleration waveforms L4 and L5, which have the largest number of acceleration waveforms evaluated to the same score among the grouped acceleration waveforms L4 to L6. As a result, the scores of the acceleration waveforms L4 to L6 of group B are all corrected to the same score of 4.

[0043] 6, among the grouped acceleration waveforms L7 to L9, the acceleration waveforms L7 to L9 are all evaluated as the same score of 3. If the scores of all the acceleration waveforms in a group are the same, the score is not corrected by the correction unit 300d, and the score is maintained. Therefore, the scores of the acceleration waveforms L7 to L9 of group C are all maintained as score 3.

[0044] Fig. 7 is a graph showing a state after correction unit 300d corrects the teacher data shown in Fig. 3. As shown in Fig. 7, the scores of acceleration waveforms L1 to L3 belonging to group A are all corrected to the same score of 2 by correction unit 300d. Moreover, the scores of acceleration waveforms L4 to L6 belonging to group B are all corrected to the same score of 4 by correction unit 300d. As a result, the scores of the acceleration waveforms belonging to each group are all corrected to the same score, and the variation in scores among the groups is eliminated.

[0045] In this way, the correction unit 300d can correct teacher data that varies within a group, and can unify all teacher data within a group to the same score. If the correction unit 300d cannot correct the scores of the acceleration waveforms L1 to L9 according to the first correction rule, it corrects the scores of the acceleration waveforms L1 to L9 according to the second correction rule.

[0046] For example, when the maximum number of acceleration waveforms evaluated to the same score among the grouped acceleration waveforms is the same, the correction unit 300d corrects the scores of the acceleration waveforms L1 to L9 according to a second correction rule. The second correction rule is to correct the scores to a predetermined score according to the magnitude of the maximum amplitude of the grouped acceleration waveforms.

[0047] 8 is a graph showing acceleration waveforms L10 to L13 of group D, which is different from groups A, B, and C classified by grouping unit 300c. The acceleration waveforms L10 to L13 are acceleration waveforms classified into group D by grouping unit 300c as acceleration waveforms having similar shapes.

[0048] 8, of the grouped acceleration waveforms L10 to L13, the acceleration waveforms L10 and L11 are evaluated as a score of 2, and the acceleration waveforms L12 and L13 are evaluated as a score of 3. In this way, in group D, of the grouped acceleration waveforms L10 to L13, the maximum numbers of acceleration waveforms evaluated as the same score are the same.

[0049] In this case, the correction unit 300d determines that the teacher data, which are the scores of the acceleration waveforms L10 to L13, cannot be corrected according to the first correction rule, and corrects the teacher data, which are the scores of the acceleration waveforms L10 to L13, according to the second correction rule.

[0050] 8, the correction unit 300d determines whether the maximum amplitudes of the grouped acceleration waveforms L10 to L13 are equal to or greater than a first threshold value according to the second correction rule. If the maximum amplitudes are equal to or greater than the first threshold value, the correction unit 300d corrects the scores of the acceleration waveforms L10 to L13 of group D to all be equal to score 2.

[0051] On the other hand, if the maximum amplitude is less than the first threshold, the correction unit 300d determines whether the maximum amplitude of the grouped acceleration waveforms L10 to L13 is equal to or greater than a second threshold. Here, the second threshold is a value smaller than the first threshold. The first threshold and the second threshold are values ​​that are set in advance by experiments or sensory evaluation.

[0052] If the maximum amplitude is less than the first threshold and greater than or equal to the second threshold, the correction unit 300d corrects the scores of the acceleration waveforms L10 to L13 of group D to all be unified to a score of 3. On the other hand, if the maximum amplitude is less than the second threshold, the correction unit 300d corrects the scores of the acceleration waveforms L10 to L13 of group D to all be unified to a score of 4.

[0053] In this way, the correction unit 300d of the present embodiment performs a correction process to correct the variation in the teacher data in the learning phase when the teacher data has variation. This correction process can eliminate the variation in the teacher data, allowing the AI ​​model to learn correctly.

[0054] In addition, when there is only one acceleration waveform grouped without having a similar shape to other acceleration waveforms, the score of the single acceleration waveform is not corrected by the correction unit 300d, and the score is maintained. However, without being limited to this, the correction unit 300d may correct the score of the single acceleration waveform in the group based on the second correction rule.

[0055] The learning unit 300e trains the AI ​​model in the learning phase based on the teacher data of the scores corrected by the correction unit 300d. In this way, the learning unit 300e trains the AI ​​model based on the teacher data with the variance corrected, so that the AI ​​model can be trained correctly, and a trained model that has been trained correctly is constructed.

[0056] In the prediction phase, the predicted score acquisition unit 300f acquires a predicted score by inputting the acceleration waveform acquired by the acceleration waveform acquisition unit 300a into the trained model constructed by the learning unit 300e.

[0057] As described above, according to the machine learning device 300 of the present embodiment, the input data and the teacher data in which the variation has been corrected are input to the AI ​​model in the learning phase, so that the AI ​​model can be correctly trained, and a correctly trained model can be constructed. As a result, the acceleration waveform can be input to the trained model as input data in the prediction phase, so that a score appropriately predicted by the trained model can be output. As a result, the trained model can perform a sensory evaluation during vehicle driving instead of an experienced driver, and the number of steps required for setting the adaptation constant can be reduced.

[0058] Fig. 9 is a flowchart showing a process flow of the machine learning device 300 according to this embodiment. As shown in Fig. 9, the acceleration waveform acquisition unit 300a acquires an acceleration waveform of the vehicle 200 based on acceleration data obtained when the vehicle 200 is traveling (step S100).

[0059] The teacher data acquiring unit 300b acquires teacher data that is a score obtained by evaluating the acquired acceleration waveform (step S200). The grouping unit 300c groups the acquired acceleration waveform based on the shape of the acceleration waveform (step S300).

[0060] The correction unit 300d corrects the scores of the grouped acceleration waveforms based on the first correction rule or the second correction rule (step S400). The learning unit 300e trains an AI model based on the teacher data of the corrected scores (step S500). The predicted score acquisition unit 300f inputs the acceleration waveform when the vehicle 200 is traveling to the learned AI model to acquire a predicted score (step S600).

[0061] Although the preferred embodiment of the present invention has been described above with reference to the accompanying drawings, it goes without saying that the present invention is not limited to such an embodiment. It is clear that a person skilled in the art can think of various modifications or alterations within the scope of the claims, and it is understood that these also naturally belong to the technical scope of the present invention.

[0062] The series of processes performed by the machine learning device 300 according to the present embodiment described above may be realized using any of software, hardware, or a combination of software and hardware. The programs constituting the software are stored in advance, for example, in a non-transitory storage medium provided inside or outside each device. The programs are then read from the non-transitory storage medium (for example, ROM) to a transitory storage medium (for example, RAM) and executed by a processor such as a CPU.

[0063] A program for implementing each function of each of the above devices can be created and installed in the computer of each of the above devices. The processor executes the program stored in the memory, thereby executing each of the above functions. At this time, the program may be shared and executed by multiple processors, or the program may be executed by one processor. Also, each of the above functions of each of the above devices may be implemented by cloud computing using multiple computers connected to each other by a communication network. The program may be provided to the computer of each device by distribution from an external device through a communication network, and installed therein. [Explanation of symbols]

[0064] 100 Machine Learning Systems 200 vehicles 210 Acceleration Sensor 300 Machine Learning Device 300a Acceleration waveform acquisition section 300b Teacher data acquisition section 300c Grouping section 300d correction section 300e Learning Department 300f Predicted score acquisition section 310 I / F section 320 Data storage unit 330 System Bus 340 Processor 350 Memory

Claims

1. one or more processors; one or more memories coupled to the processor; Equipped with The processor operates in conjunction with a program contained in the memory; Obtaining a plurality of types of acceleration waveforms when a vehicle is traveling; Obtaining a plurality of types of teacher data, which are scores obtained by evaluating each of the plurality of types of acceleration waveforms; grouping the plurality of types of acceleration waveforms based on shapes of the acceleration waveforms; correcting the scores of the grouped acceleration waveforms based on a correction rule; Perform a process that includes Machine learning device.

2. the correction rule is to use as a standard the score of the acceleration waveform having the largest number of acceleration waveforms evaluated to the same score among the grouped acceleration waveforms; The machine learning device according to claim 1 .

3. The correction rule is to correct the grouped acceleration waveforms to a predetermined score according to the magnitude of the maximum amplitude of the grouped acceleration waveforms. The machine learning device according to claim 1 .

4. The process of grouping the plurality of types of acceleration waveforms includes: The plurality of types of acceleration waveforms are grouped using a K-Shape method; The machine learning device according to any one of claims 1 to 3.

5. The processor, training an AI model based on the teacher data of the corrected score; Obtaining a predicted score by inputting an acceleration waveform when the vehicle is traveling into the learned AI model; and performing further processing including: The machine learning device according to any one of claims 1 to 3.

Citation Information

Patent Citations

  • Information processing device, information processing system, information processing method, and program

    WO2020035996A1