Vehicle ride comfort prediction device, method, and program

Abstract

To provide a vehicle ride comfort prediction device, a method, and a program capable of more appropriately predicting the ride comfort of passengers.SOLUTION: A vehicle ride comfort prediction device 1000 according to the present invention predicts the ride comfort of passengers in a vehicle and includes: a vibration acquisition unit 1 which acquires vibration data representing the vibrations experienced by the passenger; a biological signal prediction unit 22 which utilizes a first machine learning model that has been machine-learned to predict a biological signal of the passenger on the basis of the vibration data acquired by the vibration acquisition unit 1; and a ride comfort prediction unit 23 which utilizes a second machine learning model that has been machine-learned to predict the ride comfort of the passenger on the basis of the biological signal of the passenger predicted by the biological signal prediction unit 22.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a vehicle ride comfort prediction device, a vehicle ride comfort prediction method, and a vehicle ride comfort prediction program for predicting the ride comfort of a vehicle occupant. [Background technology]

[0002] With the recent development of autonomous driving technology, optimization of steering and acceleration / deceleration, including the ride comfort felt by passengers, is being considered in addition to route selection and safety. Among these, quantifying the subjective feeling of ride comfort and building a prediction model is one major challenge, and a ride comfort prediction model has been proposed, for example, in Non-Patent Document 1. The ride comfort prediction model disclosed in Non-Patent Document 1 is a machine learning model that predicts ride comfort from vehicle vibration data using an emotion structure model based on Wundt's three-way emotion theory. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Yoichi Takagishi, Koji Kamimura, Takamasa Tsunominami, Tomoyuki Kobayashi, Atsuya Yamamoto, "Development of a Sensory Evaluation Model for Ride Comfort on Thai Public Roads," Society of Automotive Engineers of Japan 2021 Autumn Meeting Academic Lecture Summary of the Invention [Problem to be solved by the invention]

[0004] In the ride comfort prediction model disclosed in the above-mentioned non-patent document 1, there is a variation in the emotion structure vector for each subject, and there is a fluctuation in the evaluation value, so there is room for improvement.

[0005] The present invention has been made in consideration of the above-mentioned circumstances, and its object is to provide a vehicle ride comfort prediction device, a vehicle ride comfort prediction method, and a vehicle ride comfort prediction program that can more appropriately predict the ride comfort of passengers. [Means for solving the problem]

[0006] As a result of various studies, the inventors of the present invention have found that the above object can be achieved by the present invention described below. That is, a vehicle ride comfort prediction device according to one aspect of the present invention is a device for predicting the ride comfort of a vehicle occupant, and includes a vibration acquisition unit that acquires vibration data representing the vibration of the occupant, a biosignal prediction unit that predicts the biosignal of the occupant with a first machine learning model that has been machine-learned based on the vibration data acquired by the vibration acquisition unit, and a ride comfort prediction unit that predicts the ride comfort of the occupant with a second machine learning model that has been machine-learned based on the biosignal of the occupant predicted by the biosignal prediction unit.

[0007] It is believed that ride comfort affects biosignals. The vehicle ride comfort prediction device predicts ride comfort from vibration data via biosignals, and therefore can more appropriately predict the ride comfort of the occupants.

[0008] In another aspect, in the above-mentioned vehicle ride comfort prediction device, the vibration data is three-axis acceleration data, and the biological signal is any one of an electrocardiogram, an R-R interval, which is the interval between R waves in the electrocardiogram, and a pulse wave.

[0009] According to this, it is possible to provide a vehicle ride comfort prediction device that uses three-axis acceleration data as vibration data and any one of an electrocardiogram, an RR interval, and a pulse wave as a biological signal.

[0010] In another aspect, in the above-mentioned vehicle ride comfort prediction device, the explanatory variables of the first machine learning model are first and second principal components obtained by principal component analysis of each transformation result obtained by wavelet transforming each acceleration data of each axis in the three-axis acceleration data.

[0011] Such a vehicle ride comfort prediction device reduces the dimensionality of explanatory variables using principal component analysis, thereby simplifying the first machine learning model.

[0012] In another aspect, in the vehicle ride comfort prediction device described above, the ride comfort is represented by an emotion vector that represents a center of gravity of the evaluation values ​​of all items in Russell's circular model.

[0013] This makes it possible to provide a vehicle ride comfort prediction device that expresses the ride comfort as an emotion vector that indicates the center of gravity of the evaluation values ​​of all items in the Russell ring model.

[0014] A vehicle ride comfort prediction method according to another aspect of the present invention is a method for predicting ride comfort of a vehicle occupant, the method comprising: a vibration acquisition step of acquiring vibration data representing the vibration of the occupant; a biosignal prediction step of predicting a biosignal of the occupant with a first machine learning model that has been machine-learned based on the vibration data acquired in the vibration acquisition step; and a ride comfort prediction step of predicting the ride comfort of the occupant with a second machine learning model that has been machine-learned based on the biosignal of the occupant predicted in the biosignal prediction step. A vehicle ride comfort prediction program according to another aspect of the present invention is a program for predicting ride comfort of a vehicle occupant, the program causing a computer to function as a vibration acquisition unit that acquires vibration data representing the vibration of the occupant, a biosignal prediction unit that predicts the biosignal of the occupant with a first machine learning model that has been machine-learned based on the vibration data acquired by the vibration acquisition unit, and a ride comfort prediction unit that predicts the ride comfort of the occupant with a second machine learning model that has been machine-learned based on the biosignal of the occupant predicted by the biosignal prediction unit.

[0015] Such a vehicle ride comfort prediction method and vehicle ride comfort prediction program predicts the ride comfort from vibration data via biological signals, and therefore can more appropriately predict the ride comfort of the occupants. Effect of the Invention

[0016] The vehicle ride comfort prediction device, the vehicle ride comfort prediction method, and the vehicle ride comfort prediction program according to the present invention can more appropriately predict the ride comfort of the occupants. [Brief description of the drawings]

[0017] [Figure 1] 1 is a block diagram showing a configuration of a vehicle ride comfort prediction device according to an embodiment. [Diagram 2] FIG. 2 is a diagram for explaining a test track as an example. [Diagram 3] FIG. 13 is a diagram showing a biological signal as an example. [Figure 4] FIG. 13 is a diagram showing three-axis acceleration data as an example. [Diagram 5] FIG. 1 is a diagram for explaining Russell's ring model. [Figure 6] FIG. 13 is a diagram showing the wavelet transform result of an electrocardiogram as an example. [Figure 7] 13 is a diagram showing, as an example, the results of wavelet transform of each acceleration data of each axis. [Figure 8] FIG. 13 is a diagram showing, as an example, the cumulative contribution ratio and the first and second principal components of a principal component analysis. [Figure 9] FIG. 13 is a diagram showing, as an example, eigenvectors of the first and second principal components. [Figure 10] As an example, this figure shows the regression characteristics of a first machine learning model of a random forest. [Figure 11] FIG. 13 is a diagram for explaining the correlation between a biosignal and an emotion vector, as an example. [Figure 12] 4 is a flowchart showing an operation of the vehicle ride comfort prediction device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] Hereinafter, one or more embodiments of the present invention will be described with reference to the drawings. However, the scope of the invention is not limited to the disclosed embodiments. In addition, configurations with the same reference numerals in each drawing indicate that they are the same configurations, and their description will be omitted as appropriate. In this specification, when referring to a general term, a reference numeral without a subscript is used, and when referring to an individual configuration, a reference numeral with a subscript is used.

[0019] The vehicle ride comfort device in the embodiment is a device that predicts the ride comfort of a vehicle occupant. This vehicle ride comfort prediction device includes a vibration acquisition unit, a biosignal prediction unit, and a ride comfort prediction unit. The vibration acquisition unit acquires vibration data representing the vibration of the occupant. The biosignal prediction unit predicts the biosignal of the occupant using a first machine learning model that has been machine-learned based on the vibration data acquired by the vibration acquisition unit. The ride comfort prediction unit predicts the ride comfort of the occupant using a second machine learning model that has been machine-learned based on the biosignal of the occupant predicted by the biosignal prediction unit. Hereinafter, such a vehicle ride comfort prediction device, as well as a vehicle ride comfort prediction method and a vehicle ride comfort prediction program implemented therein will be described in more detail.

[0020] FIG. 1 is a block diagram showing a configuration of a vehicle ride comfort prediction device in an embodiment. FIG. 2 is a diagram for explaining a test run as an example. FIG. 2A shows a run where bumps are in phase, FIG. 2B shows a run where bumps are in opposite phase, and FIG. 2C is a diagram for explaining the shape of the bumps. FIG. 3 is a diagram showing a biosignal as an example. FIG. 3A shows a pulse wave, the horizontal axis of which is time (elapsed time) and the vertical axis of which is magnitude. FIG. 3B shows an electrocardiogram, the horizontal axis of which is time (elapsed time) and the vertical axis of which is cardiac potential. FIG. 3C shows an RR interval, the horizontal axis of which is time (elapsed time) and the vertical axis of which is magnitude. FIG. 4 is a diagram showing three-axis acceleration data as an example. FIG. 5 is a diagram for explaining Russell's circular ring model. The horizontal axis of FIG. 5 is comfort / unpleasantness, and the vertical axis of FIG. 5 is awakening / calmness. FIG. 6 is a diagram showing a wavelet transform result of an electrocardiogram as an example. The horizontal axis of FIG. 6 is time (elapsed time), and the vertical axis is frequency. FIG. 7 is a diagram showing, as an example, each wavelet transform result of each acceleration data of each axis. FIG. 7A shows the wavelet transform result of the acceleration data of the X axis, with the horizontal axis being time (elapsed time), and the vertical axis being frequency. FIG. 7B shows the wavelet transform result of the acceleration data of the Y axis, with the horizontal axis being time (elapsed time), and the vertical axis being frequency. FIG. 7C shows the wavelet transform result of the acceleration data of the Z axis, with the horizontal axis being time (elapsed time), and the vertical axis being frequency. FIG. 8 is a diagram showing, as an example, the cumulative contribution ratio of the principal component analysis and the first and second principal components. FIG. 8A shows the cumulative contribution ratio, with the horizontal axis being each principal component from the first principal component, and the vertical axis being the cumulative contribution ratio. FIG. 8B shows the first and second principal components for the target vehicle speed, with the horizontal axis being the first principal component and the vertical axis being the second principal component. FIG. 8C shows the first and second principal components for the road pattern, with the horizontal axis being the first principal component and the vertical axis being the second principal component. FIG. 9 is a diagram showing, as an example, the eigenvectors of the first and second principal components. FIG. 9A shows the eigenvector of the first principal component in the X-axis acceleration data, with the horizontal axis being the frequency and the vertical axis being the magnitude.FIG. 9B is an eigenvector of the first principal component in the Y-axis acceleration data, the horizontal axis of which is frequency, and the vertical axis of which is magnitude. FIG. 9C is an eigenvector of the first principal component in the Z-axis acceleration data, the horizontal axis of which is frequency, and the vertical axis of which is magnitude. FIG. 9D is an eigenvector of the second principal component in the X-axis acceleration data, the horizontal axis of which is frequency, and the vertical axis of which is magnitude. FIG. 9E is an eigenvector of the second principal component in the Y-axis acceleration data, the horizontal axis of which is frequency, and the vertical axis of which is magnitude. FIG. 9F is an eigenvector of the second principal component in the Z-axis acceleration data, the horizontal axis of which is frequency, and the vertical axis of which is magnitude. FIG. 10 is a diagram showing the regression characteristics in the first machine learning model of a random forest, as an example. FIG. 10A shows the regression characteristic when the wavelet transform result of the electrocardiogram is used for the biosignal, FIG. 10B shows the regression characteristic when the RR interval is used for the biosignal, and FIG. 10C shows the regression characteristic when the pulse wave is used for the biosignal. The horizontal axis of each of FIG. 10A to FIG. 10C is the evaluation value, and the vertical axis is the predicted value. FIG. 11 is a diagram for explaining the correlation between the biosignal and the emotion vector as an example. FIG. 11A shows the regression characteristic when the wavelet transform result of the electrocardiogram is used for the biosignal, the horizontal axis is the evaluation value, and the vertical axis is the predicted value. FIG. 11B shows the relationship with the pleasant / unpleasant / excited / calm axis when the wavelet transform result of the electrocardiogram is used for the biosignal, the horizontal axis is the pleasant / unpleasant / excited / calm axis, and the vertical axis is the importance distribution. FIG. 11C shows the regression characteristics when the RR interval is used as the biosignal, with the horizontal axis being the evaluation value and the vertical axis being the predicted value. FIG. 11D shows the relationship with the comfort / discomfort / excitement / calm axis when the RR interval is used as the biosignal, with the horizontal axis being the comfort / discomfort / excitement / calm axis and the vertical axis being the importance distribution. FIG. 11E shows the regression characteristics when the pulse wave is used as the biosignal, with the horizontal axis being the evaluation value and the vertical axis being the predicted value. FIG. 11F shows the relationship with the comfort / discomfort / excitement / calm axis when the pulse wave is used as the biosignal, with the horizontal axis being the comfort / discomfort / excitement / calm axis and the vertical axis being the importance distribution.

[0021] A vehicle ride comfort prediction device 1000 in the embodiment includes a vibration acquisition unit 1, a control processing unit 2, an input unit 3, an output unit 4, an interface unit (IF unit) 5, and a storage unit 6, as shown in FIG.

[0022] The vibration acquisition unit 1 is connected to the control processing unit 2 by wire or wirelessly, and is a device that acquires vibration data representing the vibration of the occupant under the control of the control processing unit 2. The vibration acquisition unit 1 is, for example, a three-axis acceleration sensor attached to the chest of the occupant, and measures the vibration of the occupant on mutually orthogonal X-axis, Y-axis, and Z-axis, acquires the X-axis acceleration, Y-axis acceleration, and Z-axis acceleration, and outputs these to the control processing unit 2. For example, the X-axis is set in the up-down direction, the Y-axis is set in the left-right direction, and the Z-axis is set in the front-rear direction.

[0023] The vibration acquiring unit 1 may be configured to include an interface circuit for inputting and outputting data to and from an external device. The external device is, for example, a storage medium such as a USB (Universal Serial Bus) memory or an SD card (registered trademark) that stores three-axis acceleration data representing the vibration of an occupant as a time-series three-axis acceleration. The three-axis acceleration data includes time-series X-axis acceleration data (X-axis acceleration data (X-axis vibration data)) generated by sampling the X-axis acceleration at each sampling timing, time-series Y-axis acceleration data (Y-axis acceleration data (Y-axis vibration data)) generated by sampling the Y-axis acceleration at each sampling timing, and time-series Z-axis acceleration data (Z-axis acceleration data (Z-axis vibration data)) generated by sampling the Z-axis acceleration at each sampling timing. Alternatively, the external device is a drive device that reads data from a recording medium, such as a CD-ROM (Compact Disc Read Only Memory), a CD-R (Compact Disc Recordable), a DVD-ROM (Digital Versatile Disc Read Only Memory), or a DVD-R (Digital Versatile Disc Recordable) on which the three-axis acceleration data is recorded. Alternatively, the vibration acquisition unit 1 is, for example, a communication interface circuit that transmits and receives communication signals to and from an external device, and the external device is a server device that is connected to the communication interface circuit via a network (WAN (Wide Area Network, including a public communication network)) or LAN (Local Area Network) and manages the three-axis acceleration data. Here, when the vibration acquisition unit 1 is an interface circuit or a communication interface circuit, the vibration acquisition unit 1 may also be used as the IF unit 5 (i.e., the IF unit 5 may be used as the vibration acquisition unit 1).

[0024] The input unit 3 is connected to the control processing unit 2 and is a device for inputting various commands, such as a command to start predicting ride comfort, and various data required for operating the vehicle ride comfort prediction device 1000, such as the names of occupants, to the vehicle ride comfort prediction device 1000, and is, for example, a keyboard, a mouse, and a plurality of input switches to which predetermined functions are assigned. The output unit 4 is connected to the control processing unit 2 and is a device for outputting the commands and data input from the input unit 3 and the predicted ride comfort, etc., under the control of the control processing unit 2, and is, for example, a display device such as a CRT display, an LCD (liquid crystal display device), an organic EL display, or a printing device such as a printer.

[0025] The input unit 3 and the output unit 4 may be configured with a touch panel. In the case of configuring this touch panel, the input unit 3 is a position input device that detects and inputs an operation position, for example, a resistive film type or a capacitive type, and the output unit 4 is a display device. In this touch panel, a position input device is provided on the display surface of the display device, and one or more input content candidates that can be input to the display device are displayed. When a user touches a display position that displays the input content that the user wants to input, the position is detected by the position input device, and the display content displayed at the detected position is input to the vehicle ride comfort prediction device 1000 as the user's operation input content. In such a touch panel, the user can easily intuitively understand the input operation, so that a vehicle ride comfort prediction device 1000 that is easy for the user to handle is provided.

[0026] The IF unit 5 is connected to the control processing unit 2 and is a circuit that inputs and outputs data to and from, for example, an external device under the control of the control processing unit 2, and is, for example, an interface circuit of RS-232C, which is a serial communication method, an interface circuit using the Bluetooth (registered trademark) standard, an interface circuit using the USB standard, etc. The IF unit 5 may also be, for example, a communication interface circuit that transmits and receives communication signals to and from an external device, such as a data communication card or a communication interface circuit conforming to the IEEE802.11 standard, etc.

[0027] The storage unit 6 is connected to the control processing unit 2 and is a circuit that stores various predetermined programs and various predetermined data according to the control of the control processing unit 2. The various predetermined programs include, for example, a control processing program, and the control processing program includes, for example, a control program, a biosignal prediction program, and a ride comfort prediction program. The control program is a program that controls each of the units 1, 3 to 6 of the vehicle ride comfort prediction device 1000 according to the function of each unit. The biosignal prediction program is a program that predicts the biosignal of the occupant using a first machine learning model that has been machine-learned based on the vibration data acquired by the vibration acquisition unit 1. The ride comfort prediction program is a program that predicts the ride comfort of the occupant using a second machine learning model that has been machine-learned based on the biosignal of the occupant predicted by the biosignal prediction program. The various predetermined data include, for example, data necessary for executing each of these programs, such as the vibration data acquired by the vibration acquisition unit 1 and the ride comfort of the predicted result.

[0028] Such a storage unit 6 includes, for example, a ROM (Read Only Memory) which is a nonvolatile storage element, an EEPROM (Electrically Erasable Programmable Read Only Memory) which is a rewritable nonvolatile storage element, etc. The storage unit 6 also includes a RAM (Random Access Memory) which serves as a so-called working memory of the control processing unit 2 for storing data generated during execution of the predetermined program, etc. The storage unit 6 may also be configured to include a hard disk device with a relatively large storage capacity.

[0029] The control processing unit 2 is a circuit for controlling each of the units 1, 3 to 6 of the vehicle ride comfort prediction device 1000 according to the function of each unit, and predicting the ride comfort via a biosignal from the vibration data acquired by the vibration acquisition unit 1. The control processing unit 2 is configured, for example, with a CPU (Central Processing Unit) and its peripheral circuits. In the control processing unit 2, a control unit 21, a biosignal prediction unit 22, and a ride comfort prediction unit 23 are functionally configured by executing the control processing program.

[0030] The control unit 21 controls each of the units 1, 3 to 6 of the vehicle ride comfort prediction device 1000 in accordance with the function of each unit, and is in charge of controlling the vehicle ride comfort prediction device 1000 as a whole.

[0031] The biosignal prediction unit 22 predicts the biosignal of the occupant based on the vibration data acquired by the vibration acquisition unit 1 using a first machine learning model that has been machine-learned.

[0032] The ride comfort prediction unit 23 predicts the ride comfort of the occupant based on the biological signal of the occupant predicted by the biological signal prediction unit 22, using a second machine learning model that has been machine-learned.

[0033] In this embodiment, for example, the vibration data is three-axis acceleration data, and the biological signal is any one of an electrocardiogram, an R-R interval which is the interval between R waves in the electrocardiogram, and a pulse wave.

[0034] The explanatory variables of the first machine learning model are the first and second principal components obtained by principal component analysis of the wavelet transform of each acceleration data of each axis in the three-axis acceleration data, and the objective variable is a biosignal. The explanatory variables of the second machine learning model are the biosignal, and the objective variable is ride comfort. This ride comfort is represented by a vector (emotion vector) representing the center of gravity of each evaluation value of all items in Russell's ring model.

[0035] As the machine learning model, various models used in machine learning, such as a neural network, a support vector machine (SVM), and a random forest, can be used.

[0036] The generation of such first and second machine learning models will now be described in more detail.

[0037] In order to generate a learning data set for creating the first and second machine learning models, a test was conducted in which a vehicle carrying an occupant (subject) was driven under various driving conditions. The subject was fitted with a sensor on the chest for measuring triaxial acceleration and further measuring an electrocardiogram and a pulse wave. For example, a Silmee Bar type Lite manufactured by TDK Corporation was used as such a sensor. Silmee is a registered trademark. The triaxial acceleration data, electrocardiogram, and pulse wave are measured synchronously with each other.

[0038] As shown in FIG. 2, three bumps (crossing jigs) were provided on the road at a predetermined interval in the traveling direction, each of which was a rectangular columnar member with a trapezoidal cross section and extended in a direction perpendicular to the traveling direction of the vehicle. As shown in Table 1, four bump arrangement patterns, first to fourth patterns PN1 to PN4, were set. In the first and second patterns PN1 and PN2, the bump height was set to 15 mm, and in the third and fourth patterns PN3 and PN4, the bump height was set to 25 mm. In the first and fourth patterns PN1 and PN4, the bumps were arranged so that both the left and right wheels would cross the bumps simultaneously (Parallel), and in the second and third patterns PN2 and PN3, the bumps were arranged so that both the left and right wheels would cross the bumps alternately with a phase difference of 180 degrees (Alternate). The subjects sat in the rear right seat of the vehicle, and the test was conducted with and without blindfolds. Tests were conducted at vehicle running speeds (target vehicle speeds) of 10 [km / h], 20 [km / h], 30 [km / h], and 40 [km / h]. Thus, tests were conducted on one subject with each of the first to fourth patterns PN1 to PN4, blindfolded, unblinded, and running speeds of 10 [km / h], 20 [km / h], 30 [km / h], and 40 [km / h]. During the test, the subject's three-axis acceleration was sampled at a predetermined sampling interval, three-axis acceleration data was generated, and a pulse wave and an electrocardiogram were measured. An example of a pulse wave is shown in FIG. 3A, and an example of an electrocardiogram is shown in FIG. 3B. FIG. 3C shows the RR interval, which is the interval between R waves in an electrocardiogram, generated from the measured electrocardiogram. The R wave was extracted by Christov's algorithm. Since the RR interval is discretized at the R wave position, spline interpolation was performed to match the number of data points with other data. An example of the acceleration data of each axis in the three-axis acceleration data is shown in FIG.

[0039] [Table 1]

[0040] After the driving, the subject and the tester conducted a sensory evaluation of the ride comfort. The Russell cyclotomic model was used for the sensory evaluation index, and the index value was generated by scoring the ride comfort in a stationary state of the vehicle on a 5-point Likert scale. In the Russell cyclotomic model, as shown in FIG. 5, a total of 28 items of emotions and feelings are adopted, and each item is mapped (plotted) on a coordinate system with a horizontal axis of pleasant / unpleasant (Positive Valence / Negative Valence) and a vertical axis of arousal / calmness (High Arousal / Low Arousal), and the strength and direction of the emotion are expressed as a vector. Then, the center of gravity of each point of all these items (each point of each evaluation value) was obtained, and the vector representing this center of gravity was obtained as an emotion vector indicating the ride comfort of the passenger. This quantifies the ride comfort.

[0041] The vibration data of the three-axis acceleration data generated in this manner and any one of the biological signals of the pulse wave, electrocardiogram, and RR interval may be used as learning data (first learning data) for a first machine learning model, and a plurality of such first learning data may be included to form a learning dataset (first learning dataset) for the first machine learning model, and any one of the biological signals of the pulse wave, electrocardiogram, and RR interval and the emotion vector may be used as learning data (second learning data) for a second machine learning model, and a plurality of such second learning data may be included to form a learning dataset (second learning dataset) for the second machine learning model, but in this embodiment, the electrocardiogram and three-axis acceleration data have been subjected to a predetermined pre-processing.

[0042] More specifically, the electrocardiogram was filtered with a bandpass filter having a passband of 3 to 45 [Hz] to remove noise, and then wavelet transformed using the mother wavelet Φ(t) of the following formula 1. The wavelet transform is an analysis method for simultaneously obtaining time information and frequency information contained in the analysis target, and generates a plurality of wavelets different from each other by expanding and contracting the mother wavelet in the time axis direction, and obtains the frequency spectrum of the analysis target for each of the plurality of wavelets. An example is shown in FIG. 6. The value (magnitude) obtained by the wavelet transform at each frequency is obtained by calculating the sum of each frequency spectrum at that frequency, and is expressed as brightness in FIG. 6, with the brighter the value, the larger the value.

[0043]

number

[0044] Each acceleration data of each axis in the three-axis acceleration data was wavelet transformed using the mother wavelet ψ(t) of the following equation 2. An example is shown in Figure 7. In Figure 7, as in Figure 6, each value is expressed as brightness, with the brighter the value, the larger it is.

[0045]

number

[0046] The explanatory variables of the first machine learning model are the first and second principal components obtained by principal component analysis of the transformation results obtained by performing dimensional compression and wavelet transformation on each acceleration data of each axis in the three-axis acceleration data. As shown in FIG. 8A, the cumulative contribution rate is about 70[%] when the number of dimensions N=1, and about 90[%] when the number of dimensions N=3. Also, the first and second principal components at each target vehicle speed are shown in FIG. 8B, and the eigenvectors of the first and second principal components on each axis are shown in FIG. 9A to FIG. 9F. In both cases, acceleration of 10[Hz] or less in the X direction (vertical direction) is dominant, and the lower the frequency, the higher the contribution tends to be. In the first principal component, the frequency tends to be high around about 3.8[Hz], and in the second principal component, the frequency tends to be high around about 1.3[Hz] and about 3[Hz]. This is considered to correspond to the "fluffy feeling" and "rough feeling", respectively, and it has been shown that these have a large effect on the ride comfort in machine learning as well. It can be seen that separation is roughly achieved in the contracted space, and the first and second principal components can be used as explanatory variables.

[0047] In this embodiment, the first and second principal components are used as explanatory variables of the first machine learning model, but the first to third principal components may be used as explanatory variables of the first machine learning model.

[0048] In this embodiment, when the biological signal is a wavelet transform result of an electrocardiogram, the first learning data includes the first and second principal components of the triaxial acceleration data, and the wavelet transform result of the electrocardiogram. The first learning data set includes a plurality of the first learning data. Alternatively, when the biological signal is an R-R interval, the first learning data includes the first and second principal components of the triaxial acceleration data, and the R-R interval. The first learning data set includes a plurality of the first learning data. Alternatively, when the biological signal is a pulse wave, the first learning data includes the first and second principal components of the triaxial acceleration data, and the pulse wave. The first learning data set includes a plurality of the first learning data.

[0049] When the biological signal is a wavelet transform result of an electrocardiogram, the second learning data includes the wavelet transform result of the electrocardiogram and an emotion vector. The second learning data set includes a plurality of the second learning data. Alternatively, when the biological signal is an RR interval, the second learning data includes an RR interval and an emotion vector. The second learning data set includes a plurality of the second learning data. Alternatively, when the biological signal is a pulse wave, the second learning data includes a pulse wave and an emotion vector. The second learning data set includes a plurality of the second learning data.

[0050] The same type of biological signal is used for the first learning data and the second learning data, and the same type of biological signal is used for the first machine learning model and the second machine learning model.

[0051] Then, by using such a first learning dataset, a first machine learning model is trained by machine learning, a first machine learning model that has been machine-learned is generated, and this first machine learning model that has been machine-learned is stored in the storage unit 6. Similarly, by using such a second learning dataset, a second machine learning model is trained by machine learning, a second machine learning model that has been machine-learned is generated, and this second machine learning model that has been machine-learned is stored in the storage unit 6.

[0052] Such a machine-learned first machine learning model was generated by random forest, and its regression characteristics are shown in Figure 10. When the biological signal is the wavelet transform result of the electrocardiogram, the coefficient of determination R 2 is 0.571, and when the biological signal is the R-R interval, the coefficient of determination R 2 is 0.997, and when the biological signal is a pulse wave, the coefficient of determination R 2 The correlation coefficient was 0.996. The tendency was generally reproduced for all types of biosignals.

[0053] On the other hand, an example of the correlation between a biosignal and an emotion vector is shown in Figure 11. Random forest regression was used for the correlation between a biosignal and an emotion vector. When the biosignal is the result of wavelet transformation of an electrocardiogram, the coefficient of determination R 2 is 0.691, and when the biological signal is the R-R interval, the coefficient of determination R 2 is 0.035, and when the biological signal is a pulse wave, the coefficient of determination R 2 was 0.017. For the first and second machine learning models, an electrocardiogram is preferably used as the biological signal.

[0054] The control processing unit 2, the input unit 3, the output unit 4, the IF unit 5, and the storage unit 6 can be configured by, for example, a desktop computer, a notebook computer, etc. As described above, when the vibration acquiring unit 1 is an interface circuit or a communication interface circuit, the vibration acquiring unit 1 can also be used as the IF unit 5, and therefore the vibration acquiring unit 1 as well can be configured by a computer.

[0055] Next, the operation of this embodiment will be described with reference to a flowchart shown in FIG 12.

[0056] When the vehicle ride comfort prediction device 1000 having such a configuration is powered on, it initializes the necessary parts and starts its operation. In the control processing unit 2, a control unit 21, a biological signal prediction unit 22, and a ride comfort prediction unit 23 are functionally configured by executing the control processing program.

[0057] In FIG. 12, at a sampling timing, the vehicle ride comfort prediction device 1000 causes the control unit 21 of the control processing unit 2 to acquire the three-axis acceleration of the occupant at the current sampling timing from the vibration acquisition unit 1 and store it in the storage unit 6 (S1).

[0058] Next, the vehicle ride comfort prediction device 1000 causes the biosignal prediction unit 22 of the control processing unit 2 to predict the biosignal of the occupant using a machine-learned first machine learning model based on the three-axis acceleration data acquired by the vibration acquisition unit 1, and stores the prediction result in the storage unit 6 (S2). More specifically, the biosignal prediction unit 22 generates three-axis acceleration data using the three-axis acceleration acquired at a sampling timing a predetermined time in the past from the three-axis acceleration acquired at the current sampling timing, performs wavelet transform on the acceleration data of each axis in the generated three-axis acceleration data, inputs the first and second principal components to the machine-learned first machine learning model, and predicts the biosignal.

[0059] Next, the vehicle ride comfort prediction device 1000 predicts the ride comfort of the occupant using a second machine learning model that has been trained by machine learning, based on the occupant's bio-signals predicted by the bio-signal prediction unit 22 in the process S2, using the ride comfort prediction unit 23 of the control processing unit 2, and stores the prediction result in the memory unit 6 (S3).

[0060] Next, the vehicle ride comfort prediction device 1000 causes the control unit 21 of the control processing unit 2 to output the ride comfort of the occupant predicted by the ride comfort prediction unit 23 in the process S3 to the output unit 4 (S4). Note that the control unit 21 may output the ride comfort of the occupant to an external device via the IF unit 5 as necessary.

[0061] Then, the vehicle ride comfort prediction device 1000 judges whether or not the prediction is to be ended by the control unit 21 of the control processing unit 2 (S5). If the result of this judgment indicates that the prediction is to be ended (Yes), for example, by receiving an input of a command instructing the prediction to be ended at the input unit 3 or by receiving an unillustrated power switch being turned off, the vehicle ride comfort prediction device 1000 ends this process, whereas if the result of the judgment indicates that the prediction is not to be ended (No), the vehicle ride comfort prediction device 1000 returns the process to step S1.

[0062] By operating in this manner, the vehicle ride comfort prediction device 1000 predicts the ride comfort of the occupants at predetermined sampling intervals.

[0063] As described above, the vehicle ride comfort prediction device 1000 in the embodiment and the vehicle ride comfort prediction method and vehicle ride comfort prediction program implemented therein predict the ride comfort from the vibration data (three-axis acceleration data in the above example) via the bio-signals, since it is believed that the ride comfort affects the bio-signals, and therefore, it is possible to more appropriately predict the ride comfort of the occupants.

[0064] The vehicle ride comfort prediction device 1000, the vehicle ride comfort prediction method, and the vehicle ride comfort prediction program reduce the dimensions of explanatory variables into the first and second principal components by principal component analysis, thereby simplifying the first machine learning model.

[0065] According to the embodiments, a vehicle ride comfort prediction device 1000, a vehicle ride comfort prediction method, and a vehicle ride comfort prediction program can be provided that use three-axis acceleration data as vibration data and any one of an electrocardiogram, an RR interval, and a pulse wave as a biological signal, and that express ride comfort as an emotion vector that represents the center of gravity of each evaluation value of all items in Russell's ring model.

[0066] The vehicle ride comfort prediction device 1000, the vehicle ride comfort prediction method, and the vehicle ride comfort prediction program can quantify the ride comfort with an emotion vector.

[0067] In order to express the present invention, the present invention has been described adequately and sufficiently through the embodiments with reference to the drawings in the above description, but it should be recognized that those skilled in the art can easily change and / or improve the above-mentioned embodiments. Therefore, unless the changes or improvements made by those skilled in the art are at a level that departs from the scope of the claims described in the claims, the changes or improvements are interpreted as being included in the scope of the claims. [Explanation of symbols]

[0068] 1000 Vehicle ride comfort prediction device 1 Vibration acquisition section 2. Control processing section 3. Input section 4 Output section 5 Interface section (IF section) 6 Memory section 21 Control section 22 Biosignal Prediction Department 23 Ride Comfort Prediction Department

Claims

1. A vehicle ride comfort prediction device for predicting ride comfort of a vehicle occupant, comprising: a vibration acquisition unit that acquires vibration data representing the vibration of the occupant; a biosignal prediction unit that predicts a biosignal of the occupant using a first machine learning model that has been machine-learned based on the vibration data acquired by the vibration acquisition unit; and a ride comfort prediction unit that predicts the ride comfort of the occupant using a second machine learning model that has been machine-learned based on the occupant's biosignal predicted by the biosignal prediction unit. Vehicle ride comfort prediction device.

2. the vibration data is three-axis acceleration data, The biological signal is any one of an electrocardiogram, an R-R interval which is an interval between R waves in the electrocardiogram, and a pulse wave. The vehicle ride comfort prediction device according to claim 1 .

3. The explanatory variables of the first machine learning model are first and second principal components obtained by principal component analysis of each transformation result obtained by wavelet transforming each acceleration data of each axis in the three-axis acceleration data, The vehicle ride comfort prediction device according to claim 2.

4. The ride comfort is expressed by an emotion vector that represents the center of gravity of each evaluation value of all items in the Russell Circular Model. The vehicle ride comfort prediction device according to claim 1 .

5. A vehicle ride comfort prediction method for predicting a ride comfort of a vehicle occupant, comprising: a vibration acquiring step of acquiring vibration data representative of the vibration of the occupant; a biosignal prediction step of predicting a biosignal of the occupant using a first machine learning model that has been machine-learned based on the vibration data acquired in the vibration acquisition step; and a ride comfort prediction step of predicting the ride comfort of the occupant using a second machine learning model that has been machine-learned based on the occupant's biosignal predicted in the biosignal prediction step. A method for predicting vehicle ride comfort.

6. A vehicle ride comfort prediction program for predicting a ride comfort of a vehicle occupant, comprising: Computer, a vibration acquisition unit for acquiring vibration data representing the vibration of the occupant; a biosignal prediction unit that predicts a biosignal of the occupant using a first machine learning model that has been machine-learned based on the vibration data acquired by the vibration acquisition unit; and a ride comfort prediction unit that predicts the ride comfort of the occupant using a second machine learning model that has been machine-learned based on the occupant's biosignal predicted by the biosignal prediction unit; A vehicle ride comfort prediction program to function as a

Images