Device for generating headlight control data, and vehicle control device

Abstract

Device for generating headlight control data, which generates headlight control data of a vehicle (10) having a headlight (20), wherein the headlight control data are data used in an automatic mode in the vehicle (10), which selects between a manual mode in which a user manually performs an actuation to switch the headlight (20) from one ON state or one OFF state to the other, and the automatic mode in which a process of automatically performing the switching actuation and / or a process of outputting a recommendation message signal prompting the switching actuation to a message device (52) is performed, wherein the device for generating headlight control data comprises: an electronic control unit (30) designed to then, when manual mode is selected, set data including a detection value of an illuminance (LM) around the vehicle (10) and ON and OFF data indicating that the headlight (20) is in the ON state or the OFF state at the time the detection value is requested, and to generate the headlight control data based on the retrieved set data, wherein: the headlight control data include data relating to information about the probability of a headlight (20) being in the ON state; and where the information about a probability that a state of the headlight (20) is ON is information in which the probability that the state of the headlight (20) is ON increases when the illuminance of the headlight is within a specified range, when a proportion of the set data indicating that the state of the headlight (20) is ON is relatively high compared to when the proportion of the set data indicating that the illuminance (LM) detection value is within a specified range is relatively low.

Description

BACKGROUND OF THE INVENTION 1. Field of the invention

[0001] The present invention relates to a device for generating headlight control data, which generates control data for a headlight of a vehicle, and a vehicle control device which uses the control data. 2. Description of the related technology

[0002] For example, JP 2012-171485A describes a device for automatically learning a threshold in an automatic mode, in which a user performs an operation to automatically switch the headlights from one ON or OFF state to the other, in a vehicle in which a manual mode, in which the user manually performs a switching operation from the ON or OFF state of the headlights to the other, or the automatic mode can be selected. When a switch for learning a threshold for switching the headlights from the OFF state to the ON state is operated, which is a learning switch different from a switch normally used in manual mode, the device learns an illuminance level present at that time as the threshold for switching the headlights from the OFF state to the ON state in automatic mode.

[0003] DE 10 2007 000 144 A1 discloses a front light control device and an associated method. According to DE 10 2007 000 144 A1, if a light beam state of a front light is manually switched in a manual control operating mode, threshold values ​​of the vehicle speed, steering angle, or the like are learned to determine the switch to a low beam or high beam state based on the detection results of a speed sensor, a light sensor, a steering angle sensor, and a weather sensor, and these values ​​are stored in a non-volatile memory.In automatic control mode, a CPU determines whether to switch to low beam or high beam based on driving conditions determined by the light sensor, using the switching thresholds learned in manual control mode. The CPU then automatically switches the headlight beam state based on this determination. SUMMARY OF THE INVENTION

[0004] However, in the aforementioned case, learning cannot be performed until a condition suitable for actuating the learning switch occurs. For example, if the headlight is switched to the ON position using a normal switch immediately before starting the vehicle, and the headlight is switched to the OFF position using the normal switch immediately before a user exits the vehicle, learning cannot be performed.

[0005] The present invention discloses: a device for generating headlight control data that is capable of learning a switching time without a user recording a time at which a headlight in a manual mode is switched from one state ON or one state OFF to the other, and a vehicle control device that uses headlight control data.

[0006] A first aspect of the present invention relates to a device for generating headlight control data, which generates headlight control data for a vehicle equipped with a headlight. The headlight control data is data used in an automatic mode in the vehicle, which selects between a manual mode, in which a user manually performs an actuation to switch the headlight from one ON or OFF state to the other, and the automatic mode, in which a process of automatically performing the switching actuation and / or a process of outputting a recommendation message signal prompting the switching actuation to a communication device is carried out. The device for generating headlight control data includes an electronic control unit.The electronic control unit is designed, when manual mode is selected, to sequentially retrieve set data, including a measured illuminance value around the vehicle and ON / OFF data indicating whether the headlight is in the ON or OFF state at the time the measured value is queried, and to generate the headlight control data based on this retrieved set data. The headlight control data includes information about the probability of a headlight being in the ON state.The information about the probability of a headlight state ON is information in which, if the headlight illuminance is within a specified range, the probability of the headlight state ON increases when, of the set data indicating that the illuminance detection value is within the specified range, a proportion of set data indicating that the headlight state is ON is relatively high, compared to when the proportion of set data is relatively low.

[0007] According to the first aspect of the present invention, the electronic control unit generates data, including information about the probability of the headlight being in the ON state, based on the set data of the illuminance and the headlight's state at that time, regardless of whether the user switches the headlight ON or OFF. That is, for example, within a specific illuminance range, the electronic control unit generates data containing information about the probability of the headlight being in the ON state, even if the user does not switch the headlight ON or OFF. This probability is determined by whether the headlight is continuously in the ON or OFF state. A tendency for the user to switch the headlight ON or OFF is reflected in the probability information.Since the probability information can be used to determine the illuminance at the time the user performs the ON and OFF switching operation of the headlight in manual mode, it is therefore possible to learn a switching time without a user recording a time at which the headlight is switched from the ON state or the OFF state to the other in manual mode.

[0008] In the device for generating headlight control data according to the first aspect of the present invention, an ON switching threshold, i.e., an illuminance threshold for switching from the ON state to the OFF state in automatic mode, and an OFF switching threshold, which is an illuminance threshold for switching from the OFF state to the ON state in automatic mode, can be set independently of one another. The headlight control data can include data relating to ON switching probability information for generating the ON switching threshold and data relating to OFF switching probability information for generating the OFF switching threshold.The electronic control unit can also be designed to determine, based on information about the illuminance around the vehicle, a percentage contribution of the retrieved set data to the ON switching probability information and a percentage contribution of the retrieved set data to the OFF switching probability information.

[0009] According to the first aspect of the present invention, by determining a percentage contribution to the ON switching probability information and a percentage contribution to the OFF switching probability information, it can be determined whether the majority of retrieved set data should be used to generate data relating to ON switching probability information, to generate data relating to OFF switching probability information, or to generate both. Therefore, it is possible to generate both the ON switching probability information and the OFF switching probability information based on the set data.The ON switching threshold can be set based on data relating to ON switching probability information, and the OFF switching threshold can be set based on data relating to OFF switching probability information.

[0010] In the device for generating headlight control data according to the first aspect of the present invention, the information on the illuminance can be information indicating whether the illuminance tends to decrease or whether the illuminance tends to increase, and the electronic control unit can be designed to further determine whether the illuminance tends to decrease or tends to increase.

[0011] According to the first aspect of the present invention, the user tends to switch the headlight from the ON state to the OFF state in manual mode when the illuminance tends to increase. Conversely, when the illuminance tends to decrease, the user tends to switch the headlight from the OFF state to the ON state in manual mode. Therefore, based on the illuminance information indicating whether the illuminance tends to decrease or increase, a percentage contribution to the ON switching probability information and a percentage contribution to the OFF switching probability information can be determined.

[0012] In the device for generating headlight control data according to the first aspect of the present invention, the electronic control unit can be configured to determine, based on a time zone, whether the illuminance tends to decrease or increase. Generally, the illuminance tends to increase from morning until midday and to decrease from evening onwards. According to the first aspect of the present invention, given the tendency described above, it is possible to easily retrieve the illuminance information as time zone information.

[0013] In the device for generating headlight control data according to the first aspect of the present invention, the electronic control unit can be designed to calculate an a-posteriori distribution of an adaptation parameter of an a-posteriori probability of an ON-switching class based on the set data, the a-posteriori probability of the ON-switching class and an a-priori distribution of the ON-switching of the headlight 20, and to calculate an a-posteriori distribution of an adaptation parameter of an a-posteriori probability of an OFF-switching class based on the set data, the a-posteriori probability of the OFF-switching class and an a-priori distribution of the OFF-switching.The a posteriori probability of the ON-switching class can be a function where a detected illuminance value is an independent variable and a probability of the headlight being in the ON or OFF state is a dependent variable, with the function representing the switching from the OFF state to the ON state. The a priori distribution of the ON-switching can be a probability distribution of the fitting parameter for adjusting a value of the dependent variable with respect to the value of the independent variable of the a posteriori probability of the ON-switching class.The a posteriori probability of the OFF switching class can be a function where a detected illuminance value is an independent variable and a probability of the headlight being in the ON or OFF state is a dependent variable, with the function being a function for switching from the ON state to the OFF state. The a priori distribution of the OFF switching can be a probability distribution of the fitting parameter for adjusting a value of the dependent variable with respect to the value of the independent variable of the a posteriori probability of an OFF switching class. The data relating to ON switching probability information can be data on the a posteriori distribution of the fitting parameter in the a posteriori probability of the ON switching class.The data relating to OFF-switching probability information can be data about the A-posteriori distribution of the fitting parameter in the A-posteriori probability of the OFF-switching class.

[0014] According to the first aspect of the present invention, the fitting parameter is caused to follow the a posterior probability of the ON-switching class of the probability distribution, and the probability distribution is updated through learning. Furthermore, the fitting parameter is caused to follow the a posterior probability of the OFF-switching class of the probability distribution, and the probability distribution is updated through learning. Thus, compared to a case where the value of the fitting parameter is adjusted directly from the set data, it is possible to suppress excessive, inappropriate learning based on a random value of the set data.

[0015] In the device for generating headlight control data according to the first aspect of the present invention, the electronic control unit can be configured to further determine whether the vehicle's speed is equal to or below a threshold value. The electronic control unit can be configured to use the retrieved set data to generate the headlight control data, provided that it is not determined whether the vehicle's speed is equal to or below the threshold value.

[0016] For example, a phenomenon may occur where the headlight switches to an OFF state due to the presence of an oncoming vehicle when the vehicle is stopped at an intersection or similar location. If learning is performed using the set data when such a phenomenon occurs, it is likely that a tendency will be learned that differs from the user's original tendency to switch the headlight ON and OFF. According to the first aspect of the present invention, the occurrence of such a situation is suppressed by not using the set data for learning when the vehicle speed is equal to or lower than a threshold value.

[0017] In the device for generating headlight control data according to the first aspect of the present invention, the electronic control unit can be configured to further determine whether or not a vehicle is driving through a tunnel. The electronic control unit can be configured to use the retrieved set data to generate the headlight control data under the condition that it is not determined that the vehicle is driving through a tunnel.

[0018] Since the illuminance around the vehicle changes rapidly when the vehicle travels through the tunnel, the user tends to perform the headlight ON / OFF operation with a pattern that differs from the original ON / OFF operating pattern. According to the first aspect of the present invention, the learning of a pattern that differs from the original ON / OFF operating pattern is suppressed by not using the set data for learning when the vehicle travels through the tunnel.

[0019] In the device for generating headlight control data according to the first aspect of the present invention, the electronic control unit can be configured to further determine whether or not a predetermined building is present within a predetermined distance of the vehicle. The electronic control unit can be configured to use the retrieved set data to generate the headlight control data under the condition that the electronic control unit does not determine that the predetermined building is present.

[0020] For example, if the vehicle is driving close to a relatively tall building, the illuminance can decrease rapidly due to the building's shadow. Furthermore, the illuminance around the vehicle can increase rapidly if the vehicle is driving close to a building such as an industrial plant that uses multiple lights. Therefore, when the illuminance around the vehicle changes rapidly, the user tends to operate the headlights on and off with a tendency that differs from their original on / off operating pattern.According to the first aspect of the present invention, the learning of a tendency that differs from an original tendency of the ON and OFF switching operation of the headlight is suppressed by not using the set data for learning when a distance to a predetermined housing is equal to or less than a predetermined distance.

[0021] A second aspect of the present invention relates to a vehicle control device which includes the device for generating headlight control data of the first aspect of the present invention. The electronic control unit is designed to set the ON switching threshold and the OFF switching threshold based on the headlight control data. In automatic mode, the electronic control unit switches the headlight from the OFF state to the ON state if the illuminance detection value is less than the ON switching threshold when the headlight is in the OFF state, and in automatic mode, it switches the headlight from the ON state to the OFF state if the illuminance detection value is greater than the OFF switching threshold when the headlight is in the ON state.

[0022] According to the second aspect of the present invention, it is possible to suppress user confusion in automatic mode by executing the automatic mode based on the threshold value, which reflects a tendency of a user to switch the headlight from the ON state to the OFF state.

[0023] A third aspect of the present invention relates to a device for generating headlight control data, which generates headlight control data for a vehicle equipped with a headlight. The headlight control data is data used in an automatic mode in the vehicle, which selects between a manual mode, in which a user manually performs an actuation to switch the headlight from one ON or OFF state to the other, and the automatic mode, in which a process of automatically executing the switching actuation and / or a process of outputting a recommendation message signal prompting the switching actuation to a communication device is performed. The device for generating headlight control data includes an electronic control unit.The electronic control unit is designed, when manual mode is selected, to sequentially retrieve set data, including an illuminance measurement around the vehicle and ON / OFF data indicating whether the headlight is in the ON or OFF state at the time the measurement is taken, and to generate the headlight control data based on this retrieved set data. The headlight control data includes a threshold for activation in automatic mode, and the electronic control unit is designed to set this threshold within an illuminance range where both the ON and OFF set data are present simultaneously.

[0024] For example, it can be assumed that if the headlight is continuously in the ON state within a certain illuminance range, the user will very likely not be inclined to switch the headlight from the ON state or from the OFF state to the other within that illuminance range. Furthermore, it can be assumed that if the headlight is continuously in the OFF state within another illuminance range, the user will very likely not be inclined to switch the headlight from the ON state or from the OFF state to the other within that illuminance range.According to the third aspect of the present invention, the electronic control unit retrieves the set data regarding the illumination range and the headlight state at any given time, independently of the user's ON / OFF operation of the headlight, and sets a threshold value with respect to an illuminance range in which the headlight is simultaneously in both an ON and an OFF state. It can be assumed that in this illuminance range, where the headlight is simultaneously in both an ON and an OFF state, the user has a relatively strong tendency to switch the headlight from one state to the other. Therefore, it is possible to learn a switching time without the user having to detect a specific time when the headlight is switched from one state to the other in manual mode.

[0025] A fourth aspect of the present invention relates to a vehicle control device which includes the device for generating headlight control data of the third aspect of the present invention. The electronic control unit is designed to switch the headlight between the ON and OFF states based on a comparison of the illuminance measurement and the threshold value when the headlight is in automatic mode in either the ON or OFF state.

[0026] According to the fourth aspect of the present invention, it is possible to suppress user confusion in automatic mode by executing the automatic mode based on the threshold value, which reflects a tendency of a user to switch the headlight from the ON state to the OFF state. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Features, advantages and the technical and industrial significance of exemplary embodiments of the invention are described below with reference to the accompanying drawings, in which the same reference numerals denote the same elements: Fig. Figure 1 is a sketch depicting the construction of a vehicle according to a first embodiment; Fig. 2 is a flowchart that depicts the sequence of a process in an automatic mode and its preprocessing according to the first embodiment; Fig. Figure 3 is a flowchart that illustrates the sequence of a process for classifying set data according to the first embodiment; Fig. Figure 4 is a flowchart that illustrates the sequence of a process for updating parameters according to the first embodiment; Fig. Figure 5 is a flowchart illustrating the sequence of a procedure for updating parameters according to the first embodiment; Fig. Figure 6 is a flowchart illustrating the sequence of a process for updating a switching threshold according to the first embodiment; Fig. Figure 7 is a flowchart that illustrates the sequence of a process for classifying set data according to a second embodiment; Fig. Figure 8 is a flowchart illustrating the sequence of a process for classifying set data according to a third embodiment; and Fig. Figure 9 is a flowchart that depicts the sequence of a process in an automatic mode and its preprocessing according to a fourth embodiment. DETAILED DESCRIPTION OF EXECUTION FORMS First embodiment

[0028] The following describes a first embodiment of a device for generating headlight control data and a vehicle control device with reference to the drawings.

[0029] A in Fig. The vehicle 10 shown in Figure 1 is equipped with a headlight 20 that illuminates the side in front of the vehicle 10 in one direction of travel. Furthermore, the vehicle 10 is equipped with an electronic control unit (ECU 30). The ECU 30 sets the headlight 20 as the control target and executes a control operation to switch the headlight 20 from the ON state or from the OFF state to the other. The ECU 30 comprises a CPU 32, a ROM 34 that stores programs executed by the CPU 32, a memory device 36, and a communication device 38. The memory device 36 is a non-volatile memory device that can electrically overwrite stored content.

[0030] The ECU 30 receives an output signal from an ON / OFF toggle switch 42, which the user uses to command the headlight 20 to switch from the ON or OFF state to the other. The ECU 30 also receives an output signal from a mode switch 44, which the user uses to command a switch between a manual mode, in which the user operates the ON / OFF toggle switch 42, and an automatic mode, in which the ECU 30 automatically switches the headlight 20 from the ON or OFF state to the other, regardless of the user's operation. Furthermore, the ECU 30 receives an illuminance LM around the vehicle 10, detected by an illuminance sensor 46, and a vehicle speed V, detected by a vehicle speed sensor 48.Furthermore, the ECU 30 can communicate with a navigation system 50 via a network in the vehicle and output a voice signal to a speaker 52. The ECU 30 can also communicate with a user's mobile device 60 using the communication device 38.

[0031] Fig. Figure 2 represents the sequence of a process in automatic mode. The one in Fig. The process shown in Figure 2 is executed by CPU 32, which repeatedly runs the program stored in ROM 34 within a predetermined time period. The step numbers are subsequently represented as digits beginning with "S".

[0032] In a series of processes that took place in Fig. As shown in Figure 2, the CPU 32 first determines whether or not flag F is set to "1" (S10). Flag F is "1" if the user ID has been retrieved and is "0" if the user ID has not been retrieved. If the CPU 32 determines that flag F is "0" (S10: NO), the CPU 32 retrieves the user ID by communicating with the mobile device 60 using the communication device 38 (S12). This can be achieved, for example, by installing a dedicated application program that causes the vehicle 10 to retrieve the user ID from the mobile device 60. In this embodiment, it is assumed that multiple users utilize the vehicle 10 and that the users have different user IDs.

[0033] When CPU 32 retrieves the user ID, it extracts a process specific to that user ID (S14). This process generates the data for switching the headlight 20 from the ON or OFF state to the other in automatic mode. Fig. 2 concerns, and data (described below) for generating the data are set to be specific to the user ID. CPU 32 sets flag F to "1" (S16).

[0034] When the process in S16 is complete, or if the determination in step S10 is positive, CPU 32 determines whether or not automatic mode is selected (S18). If CPU 32 determines that automatic mode is selected (S18: YES), it then retrieves the illuminance LM (S20). The illuminance is quantified in lux or a unit of measurement proportional to lux. CPU 32 then determines whether or not the state of headlight 20 is OFF (S22). If CPU 32 determines that headlight 20 is OFF (S22: YES), it determines whether the log value ln(LM) of the illuminance LM is less than the ON switching threshold xthfn (S24). This process determines whether or not headlight 20 should be switched on automatically. The use of the log value of the illuminance LM takes into account that a person's vision does not respond linearly to the illuminance LM.If CPU 32 determines that the log value ln(LM) is less than the ON switch-on threshold xthfn (S24: YES), then CPU 32 switches on the headlight 20 (S26).

[0035] If, on the other hand, CPU 32 determines that the headlight is in a state ON (S22: NO), then CPU 32 determines whether the log value ln(LM) is greater than an OFF switching threshold xthnf (S28). If CPU 32 determines that the log value ln(LM) is less than the OFF switching threshold xthnf (S28: YES), then CPU 32 switches off headlight 20 (S30).

[0036] When the process of step S30 is completed, or when a negative determination is made in S18, S24, and S28, CPU 32 terminates the sequence of processes that began in Fig. Figure 2 is shown for this time. Now, the calculation of the ON switching threshold xthfn and the OFF switching threshold xthnf is described. The ON switching threshold xthfn and the OFF switching threshold xthnf are calculated based on set data that includes a label variable t, which indicates whether the headlight 20 is in the ON or OFF state, and the log value of the illuminance LM (hereinafter: an illuminance x). The label variable t is "1" if the headlight 20 is ON, and "0" if the headlight 20 is OFF. In this embodiment, the set data is associated with the user ID via process S12. The set data used to calculate the ON switching threshold xthfn and the OFF switching threshold xthnf corresponding to a specific user ID are limited to the set data for the case where the user with user ID is using vehicle 10.

[0037] Fig. Figure 3 shows the process of classifying set data as ON learning data for calculating the ON switching threshold xthfn and as OFF learning data for calculating the OFF switching threshold xthnf. The in Fig. The process shown in Figure 3 is repeatedly carried out by the CPU 32 by executing the program stored in ROM 34 within a specified time period.

[0038] In a series of processes that took place in Fig. As shown in Figure 3, CPU 32 first determines whether or not it is in manual mode (S40). This process determines whether or not it is time to retrieve the ON and OFF learning data. In this embodiment, the ON threshold xthfn and the OFF threshold xthnf are learned by supervised learning, where the illuminance data set x is an input vector when manual mode is in use, and the label variable t is a target vector. Therefore, manual mode is an execution condition for the process of retrieving the ON and OFF learning data.

[0039] When CPU 32 determines that the mode is manual mode (S40: YES), CPU 32 retrieves the label variable t, which indicates the ON or OFF state of the spotlight 20, and the illuminance x (S42). CPU 32 then determines whether the current time is "0:00 to 12:00" (S44). This process determines whether the current time is a period during which the illuminance x tends to increase. If the illuminance x tends to increase, the user generally tends to switch the spotlight 20 from the ON state to the OFF state. Therefore, in this embodiment, the set data for the period during which the illuminance x tends to increase are set as OFF learning data.

[0040] If CPU 32 determines that the current time is "0:00 to 12:00" (S44: YES), CPU 32 sets the set data retrieved in process S42 as the OFF learning data for this instance (S46). CPU 32 determines whether the set data contains an anomalous value based on an OFF switching probability function pnf(x), which is calculated using a process described below (S48). This process determines whether the newly retrieved set data from process S42 can be adopted as learning data. Specifically, this process determines that the set data is anomalous if it deviates significantly from a probability prediction based on the OFF switching probability function pnf(x). More precisely, −ln⌊pnf(x)t⋅{1−pnf(x)}1−t⌋ the CPU 32 in the embodiment such that the set data are anomalous if a negative log likelihood defined in equation (c1) below is greater than the threshold lnth.

[0041] If the set data indicates that the headlight state is ON, the label variable t becomes "1", and therefore the independent variable of the log function in equation (c1) above becomes the OFF switching probability function pnf(x). If the value of the dependent variable of the OFF switching probability function pnf(x), where the illuminance x of the set data is an independent variable, is relatively small, meaning that the probability of the headlight state being ON is relatively small, then a negative log likelihood will be a relatively large value. Conversely, if the set data indicates that the headlight state is OFF, the label variable t becomes "0", and therefore the independent variable of the log function in equation (c1) above becomes "1 - pnf(x)".If the value of the dependent variable of the OFF switching probability function pnf(x), where the illuminance x in the set data is an independent variable, is relatively large, meaning that the probability of the headlight being OFF is relatively small, then a negative log-likelihood will be a relatively large value. If CPU 32 determines that the set data does not have an anomalous value (S48: NO), then CPU 32 sets the set data as the OFF learning data (S50).

[0042] If, on the other hand, CPU 32 determines that the current time is "12:00 to 0:00" (S44: NO), CPU 32 executes a process that corresponds to the processes from S46 to S50 within the processes from S52 to S56. That is, CPU 32 first provisionally sets the set data, which was retrieved in process S42, as ON learning data (S52). Based on an ON switching probability function pfn(x), which is calculated using a process described below (S48), CPU 32 determines whether or not the set data exhibits an anomalous value. Here, the CPU 32 determines, based on this, whether or not a negative log-likelihood, in which the OFF switching probability function pnf(x) in equation (c1) is replaced by the ON switching probability function pfn(x), is equal to or greater than the threshold lnth.If CPU 32 determines that the set data does not show an anomalous value (S54: NO), then CPU 32 sets the set data as the ON learning data (S56).

[0043] When the processes of S50 and S56 are complete, CPU 32 also terminates the process in Fig. 3. Sequence of processes shown for this time, if the negative determination is made in S40 or if the positive determination is made in S48 and S54.

[0044] When the ON learning data is retrieved using the process described above, the ON switching probability function pfn(x), which is a function of the probability that the headlight state 20 is ON, is learned in relation to the illuminance x based on the ON learning data. In this embodiment, the a posterior probability pfn(t = 1 | x) of the ON switching class, that is, the a posterior probability that the label variable t becomes "1", is represented by a sigmoid function in which a value of a linear function of the illuminance x is an independent variable.In particular, the a posteriori probability pfn(t = 1 | x) of the ON-switch class is a function represented by the equation (c2) below, in which a value obtained by multiplying a scalar product of a fitting parameter vector Wfn = (wfn0, wfn1) and a vector (1, x) by “-1” is an independent variable. pfn(t=1|x)=11+e−(wfn0+wfn1⋅x)

[0045] In this embodiment, the probability distribution of the fitting parameter vector Wfn is updated through learning. That is, the probability distribution of the fitting parameter vector Wfn is updated by calculating the posterior distribution, assuming that the prior distribution of the fitting parameter vector Wfn is a Gaussian distribution N(Wfn | Mfn0, Sfn0). The Gaussian distribution N(Wfn | Mfn0, Sfn0) is defined by the mean Mfn0 and the variance-covariance matrix Sfn0. If the learning has not yet been performed, a value derived from big data, for example, is set as the initial value for the mean Mfn0 and the variance-covariance matrix Sfn0. The following describes the process.

[0046] Fig. Figure 4 illustrates the calculation of the ON switching probability function pfn(x) using the ON learning data. The in Fig. The process shown in Figure 4 is achieved by the CPU 32 repeatedly executing the program stored in ROM 34 within a specified period of time.

[0047] In a series of processes that took place in Fig. As shown in Figure 4, CPU 32 first determines whether the number of queries of the ON learning data is n (S60). If CPU 32 determines that the number of queries is n (S60: YES), CPU 32 reads a prior distribution pfn(Wfn) of ON switching, which is a Gaussian distribution, from memory device 36 (S62). CPU 32 computes a MAP solution Wfnmap, which is a value of the fitting parameter vector Wfn for maximizing a posterior distribution pfn(Wfn | T) of ON switching, based on an input vector X = {x(1), x(2), ... , x(n)} including n set data elements and a label vector T = {t(1), t(2), ..., t(n)} (S64).

[0048] Here, a log value of the A posteriori distribution pfn(Wfn | T) of the ON switching is proportional according to Bayes' theorem to a log value lnp(Wfn | T) of an A posteriori distribution represented by the equation (c3) below. ln p(Wfn|T)=ln⌊N(Wfn|Mfn0,Sfn0)∏k=1npfn{t=1|x(k)}t(k)[1−pfn{t=1|x(k)}]t(k)⌋

[0049] CPU 32 calculates the fitting parameter vector Wfn to maximize the log value lnp(Wfn | T) of the a posteriori distribution as the MAP solution. This can be done, for example, using a conjugate gradient descent method.

[0050] When CPU 32 computes the MAP solution Wfnmap, it assumes that the A posteriori distribution pfn(Wfn | T) of the ON-distribution is also a Gaussian distribution and computes a mean Mfn and a variance-covariance matrix Sfn (S66). The mean Mfn becomes the MAP solution Wfnmap. Furthermore, the variance-covariance matrix is ​​expressed by the following equation (c4). However, yfn(k) Sfn−1←Sfn0−1+∑k=1nyfn(k)⋅{1−yfn(k)}⋅Wfn(k)Wfn(k)T a value of the dependent variable, where the fitting parameter vector Wfn of the sigmoid function is the MAP solution Wfnmap and an independent variable of the sigmoid function is the illuminance x(k).

[0051] Then, instead of the mean Mfn0 of the prior distribution stored in memory device 36, the CPU 32 sets the current mean Mfn and, instead of the variance-covariance matrix Sfn0 of the prior distribution stored in memory device 36, sets the current variance-covariance matrix Sfn (S68).

[0052] CPU 32 integrates a product of the a posterior probability pfn(t = 1 | x) of the ON class and the a posterior distribution pfn(Wfn | T) of the ON with respect to the fitting parameter vector Wfn to compute the ON-switching probability function pnf(x) (S70). CPU 32 then deletes the n query values ​​(S72).

[0053] When the process in S72 is complete, or when a negative determination is made in S60, CPU 32 terminates the sequence of processes running in S72. Fig. 4 are shown, for this time. Fig. Figure 5 illustrates the calculation of the OFF switching probability function pnf(x) using the OFF learning data. The in Fig. The process shown in Figure 5 is executed by CPU 32, which repeatedly runs the program stored in ROM 34 within a predetermined time period. The processes S60a to S72a in Fig. 5 correspond to the processes of S60 to S72 in Fig. 4 and are part of the processes of Fig. 4 are the same, except that the data that needs to be processed is for switching off.

[0054] Fig. Figure 6 describes the process of updating the ON switching threshold xthfn and the OFF switching threshold xthnf. The in Fig. The process shown in Figure 6 is carried out by the CPU 32, which repeatedly executes the program stored in ROM 34 within a specified period of time.

[0055] In a series of processes that took place in Fig. As shown in Figure 6, the CPU 32 first reads the variance-covariance matrices Sfn0 and Snf0 with respect to the prior distributions of both the ON and OFF states from the memory device 36 (S80). The CPU 32 determines whether or not a logical product of: an absolute value of a determinant of the variance-covariance matrix Sfn0 for the ON state is less than a threshold Sth AND an absolute value of the determinant of the variance-covariance matrix Sfn0 for the OFF state is less than the threshold Sth is true (S82). This process is intended to determine whether or not the ON state threshold xthfn and the OFF state threshold xthnf should be updated. That is, since it can be assumed that learning has converged when the absolute value of the determinant is less than the threshold Sth, this is set as a condition for updating the ON switching threshold xthfn and the OFF switching threshold xthnf.Here, it is assumed that the variance-covariance matrices Sfn0, Snf0, which are previously taken as a default setting before the learning process, are set to the negative value determined in S82.

[0056] When CPU 32 determines that the logical product is true (S82: YES), CPU 32 updates the ON switching threshold xthfn and the OFF switching threshold xthnf (S84). In this embodiment, the ON switching threshold xthfn is the illuminance x that occurs when the ON switching probability function pfn(x) is '1 / 2', and the OFF switching threshold xthnf is the illuminance x that occurs when the OFF switching probability function pnf(x) is '1 / 2'. In this embodiment, in processes S70 and S70a, the ON switching probability function pfn(x) and the OFF switching probability function pnf(x) approximate the sigmoid function. In this case, an independent variable of the sigmoid function can be expressed as cWfnX, cWnfX using a constant c (see, for example, Chapter 4 in “Pattern Recognition and Machine Learning: CM Bishop”).Therefore, the illuminance x given when the probability becomes “1 / 2” can be obtained as the illuminance x where a scalar product of the MAP solutions Wfnmap, Wnfmap calculated in S64 and S64a, and (1, x) is “0”.

[0057] Furthermore, CPU 32 terminates the sequence of processes that are in Fig. Figure 7 shows the following for this time: when the process of S84 is complete or when a negative determination is made in S82. If CPU 32 makes the negative determination in S82, CPU 32 can use the default values ​​as the ON threshold xthfn and the OFF threshold xthnf. The default values ​​can, for example, be pre-stored in ROM 34.

[0058] This section describes one mode of operation of the embodiment. When the user enters the vehicle 10 and selects manual mode, the CPU 32 performs a query using whether the headlight 20 is in the ON or OFF state as a label variable t at each periodic query time, which is set independently of the operation of the ON / OFF toggle switch 42, and queries the illuminance x at that time. In doing so, the CPU 32, at a time when it queries n set data elements from "0:00 to 12:00", calculates the MAP solution Wfnmap of the log lnp (Wfn | T) of the A posteriori distribution, which is represented by equation (c3) above.

[0059] In a case where a relatively high proportion of the set data elements, where the headlight 20 is in the ON state within a given illuminance range, is present, the a posteriori probability pfn(t = 1|x) of the ON-switching class increases the probability that the headlight 20 switches to the ON state within the given illuminance range, compared to a case where the proportion of the set data elements where the headlight 20 is in the ON state is relatively low. Accordingly, a log lnp (Wfn | T) of the a posteriori distribution is increased. Therefore, the MAP solution Wfnmap is a value by which the a posteriori probability pfn(t = 1 | x) of the ON-switching class increases the probability that the headlight 20 switches to the ON state within the given illuminance range.

[0060] Since the a posteriori distribution pfn(Wfn | T) of the ON switching is obtained by averaging the MAP solution Wfnmap calculated in this way, the ON switching probability function pfn(x) is a function in which the distribution of the fitting parameters is particularly high near the MAP solution Wfnmap. Thus, in a case where a relatively high proportion of the set data elements in which the headlight 20 is in the ON state within a given illuminance range is relatively high, the ON switching probability fpfn(x) tends to increase the probability that the headlight 20 will switch to the ON state within the given illuminance range, compared to a case where the relatively low proportion of the set data elements in which the headlight 20 is in the ON state is relatively low.Therefore, the user's tendency to switch the headlight 20 from the OFF state to the ON state is reflected in the ON switching probability function pfn(x). Likewise, the user's tendency to switch the headlight 20 from the ON state to the OFF state is reflected in the OFF switching probability function pnf(x). Therefore, the switching time can be learned without recording the time at which the user switches the headlight from the ON state or from the OFF state to the other in manual mode.

[0061] Furthermore, in the illuminance range where only data elements from the n set data points indicate that the spotlight 20 is in the ON state, the log lnp (Wfn | T) of the a posteriori distribution is increased by the a posteriori probability pfn(t = 1 | x) of the ON-switching class, which increases the probability that the spotlight 20 transitions to the ON state. Therefore, the MAP solution Wfnmap is a value such that an a posteriori probability pfn(t = 1 | x) of the ON-switching class increases the probability that the spotlight 20 transitions to the ON state in the illuminance range.In contrast, in the illuminance range where the only data from the n set data elements is that the spotlight 20 is in the OFF state, the log lnp (Wfn | T) of the A posteriori distribution is increased by the A posteriori probability pfn(t = 1 | x) of the ON-switching class, which increases the probability that the spotlight 20 transitions to the OFF state. Therefore, the MAP solution Wfnmap is a value by which the A posteriori probability pfn(t = 1 | x) of the ON-switching class increases the probability that the spotlight 20 goes into the OFF state in the illuminance range.In this case, within an illuminance range where the set data elements where headlight 20 is in the ON state and the set data elements where headlight 20 is in the OFF state are both present simultaneously, the probability that headlight 20 transitions to the ON state is calculated based on the a posteriori probability pfn(t = 1 | x) of the ON-switching class as a mean value. Therefore, in a case where the illuminance x, which is given when the value of the ON-switching probability function pfn(x) is "1 / 2", is used as the ON-switching threshold xthfn, there is a tendency for the ON-switching threshold xthfn to be a value within a range of illuminance x where the set data elements where headlight 20 is in the ON state and the set data elements where headlight 20 is in the OFF state are both present simultaneously.Similarly, the OFF switching threshold xthnf tends to be a value in a range of illuminance x in which the set data where the headlight 20 is in the ON state and the set data where the headlight 20 is OFF state are present simultaneously.

[0062] According to the embodiment described above, the following effects are also achieved. (1) The set data are classified as ON learning data and OFF learning data based on a time zone. Accordingly, it is possible to easily determine whether the individual set data units contribute to the ON switching probability function pfn(x) or to the OFF switching probability function pnf(x).

[0063] (2) The CPU 32 determines whether or not the set data exhibits an anomalous value, based on the ON switching probability function pfn(x) or the OFF switching probability function pnf(x), and prevents the set data from being used for learning if it exhibits an anomalous value. This makes it possible to suppress situations where learning is performed using unsuitable data that does not reflect a user's preferences. Second embodiment

[0064] The following mainly describes the differences between a second embodiment and the first embodiment with reference to the drawings.

[0065] For example, in a case where an oncoming vehicle is present, when vehicle 10 is stopped at an intersection, some users might perform a toggle operation of the headlight 20 from the ON state to the OFF state because of the oncoming vehicle. This toggle operation differs from the normal tendency with which the user switches the headlight 20 from the ON state to the OFF state. Therefore, in the embodiment, the use of the set data, in which such an operation was reflected for learning purposes, was suppressed by the following process.

[0066] Fig. Figure 7 illustrates the process of classifying set data as ON learning data and OFF learning data. The in Fig. The process shown in Figure 7 is implemented by the CPU 32, which repeatedly executes the program stored in ROM 34 within a predetermined time period. Fig. 7 processes will be those that are in Fig. The 3 processes shown correspond to each other, for the sake of simplicity they are designated with the same step numbers, and their description is omitted.

[0067] In a series of processes that took place in Fig. As shown in Figure 7, CPU 32 determines whether the vehicle speed V is higher than a threshold value Vth when the process of S42 is completed (S90). This process is intended to determine that the vehicle 10 has stopped, and the threshold value Vth is set to a very low speed value. CPU 32 terminates the sequence of processes that are shown in Figure 7. Fig. 7 are shown, for this time when CPU 32 determines that the vehicle speed V is equal to or lower than the threshold Vth (S90 NO), and proceeds to the process of S44 when CPU 32 determines that the vehicle speed V is higher than the threshold Vth (S90: YES). Third embodiment

[0068] The following mainly describes the differences between the third embodiment and the first embodiment with reference to the drawings.

[0069] For example, when vehicle 10 enters a tunnel, a phenomenon may occur in which the headlight 20 switches ON. If learning is performed using the set data when such a phenomenon occurs, it is likely that a tendency will be learned that differs from the original tendency of the user to switch the headlight ON and OFF.

[0070] Furthermore, if the vehicle 10 is driving near a relatively tall building, for example, the illuminance can decrease rapidly due to the building's shadow. Conversely, the illuminance around the vehicle 10 can increase rapidly if the vehicle 10 is driving near a building such as an industrial plant that uses multiple lights. Therefore, if the illuminance around the vehicle 10 changes rapidly, the user is more likely to perform an ON / OFF operation of the headlight with a tendency that differs from the original ON / OFF operating tendency of the headlight 20.

[0071] In this embodiment, the use of data in which such activity was reflected for learning purposes is suppressed by the following process. Fig. Figure 8 illustrates the process of classifying set data as ON learning data and OFF learning data. The section in Fig. The process shown in Figure 8 is implemented by the CPU 32, which repeatedly executes the program stored in ROM 34 within a predetermined time period. Fig. 8 processes will be those that are in Fig. The 3 processes shown correspond to each other, for the sake of simplicity they are designated with the same step numbers, and their description is omitted.

[0072] In a series of processes that took place in Fig. As shown in Figure 8, once the process of S42 is complete, CPU 32 communicates with the navigation system 50 to retrieve map information (S92). Based on this map information, CPU 32 determines whether or not vehicle 10 is driving through a tunnel (S94). To do this, CPU 32 sets a starting point, located a predetermined distance from the tunnel entrance, and an endpoint, located a predetermined distance from the tunnel exit. It then determines that vehicle 10 is driving through the tunnel if it is traveling within a segment between these starting and endpoints. If CPU 32 determines that the vehicle is not driving through the tunnel (S94: NO), it uses the map data to determine whether or not the vehicle is driving in an urban area (S96).This process is intended to determine whether or not vehicle 10 is driving in a location where the illuminance around vehicle 10 is likely to change abruptly. If CPU 32 determines that the vehicle is not driving in an urban area (S94: NO), CPU 32 proceeds to process S44. If the processes of S94 and S96 reach a positive determination, CPU 32 terminates the sequence of processes that are in S44. Fig. 8 is shown, for this time. Fourth embodiment

[0073] The following mainly describes the differences between a fourth embodiment and the first embodiment with reference to the drawings.

[0074] In this embodiment, a process is carried out in automatic mode which recommends to the user to operate the ON and OFF changeover switch 42 himself. Fig. Figure 9 represents the sequence of a process in automatic mode. The one in Fig. The process shown in Figure 9 is implemented by the CPU 32, which repeatedly executes the program stored in ROM 34 within a predetermined time period. Fig. 9 processes that are in Fig. The two processes shown correspond to each other, for the sake of simplicity they are designated with the same step numbers, and their description is omitted.

[0075] In a series of processes that took place in Fig. As shown in Figure 9, if the CPU 32 makes a positive determination in S24, it outputs a voice signal to the loudspeaker 52 recommending that the user switch the headlight 20 to the ON state (S26a). Accordingly, an audio signal "Switch on headlight? It's getting dark" is output by the loudspeaker 52 and communicated to the user. Furthermore, if the CPU 32 makes a positive determination in S28, it outputs a voice signal to the loudspeaker 52 recommending that the user switch the headlight 20 to the OFF state (S30a). Accordingly, an audio signal "Switch off?" is output by the loudspeaker 52 and communicated to the user.

[0076] The processes of steps S40 and S42 represent the retrieval, and the processes from S44 to S56 in Fig. 3 or the processes of Fig. Steps 4 to 6 represent the generation of the tax data. The process from S22 to S30 in Fig. 2 or the processes of S22, S24, S26a, S28 and S30a in Fig. 9 represents the automatic mode. The ON switching threshold xthfn, the OFF switching threshold xthnf, and the variance-covariance matrices Sfn0 and Snf0 represent control data. The loudspeaker 52 represents the communication device. The CPU 32 and the ROM 34 represent the electronic control unit. The ECU 30 represents the device for generating headlight control data. The variance-covariance matrix Sfn0 and the mean Mfn0 represent the data relating to the ON switching probability, and the variance-covariance matrix Snf0 and the mean Mnf0 represent the data relating to the OFF switching probability. The process of S90 represents the determination of whether the vehicle's speed is equal to or lower than the threshold. The processes of S92 and S94 determine whether or not the vehicle passes through the tunnel.Processes S92 and S96 determine whether a given building exists within a specified distance of the vehicle. ECU 30 represents the vehicle control unit. Processes S40 and S42 represent the retrieval, and processes S44 to S56 in... Fig. 3 or the processes of Fig. Processes 4 to 6 represent the generation of the tax data. The processes from S22 to S30 in Fig. 2 represent the automatic mode, or the processes of S22, S24, S26a, S28 and S30a in Fig. 9 represents the automatic mode. CPU 32 and ROM 34 represent the electronic control unit. ECU 30 represents the device for generating headlight control data. ECU 30 represents the vehicle control device. Other embodiments

[0077] At least one of the respective points of the above embodiment can be modified as follows. Derivation of the ON switching threshold xthfn, the OFF switching threshold xthnf based on the probability function

[0078] In the embodiment above, the ON switching threshold xthfn was the illuminance x that occurs when the ON switching probability function pfn(x) becomes "1 / 2", but the present invention is not limited thereto. For example, the ON switching threshold xthfn can be an illuminance x that occurs when the ON switching probability function pfn(x) becomes "2 / 3". In this case, the tendency is for the ON switching threshold xthfn to be a value in a range of illuminance x where the set data at which the headlight 20 is in the ON state and the set data at which the headlight 20 is in the OFF state are both present simultaneously.

[0079] In the embodiment above, the OFF switching threshold xthnf was the illuminance x that occurs when the OFF switching probability function pnf(x) becomes "1 / 2", but the present invention is not limited thereto. For example, the OFF switching threshold xthnf can be an illuminance x that occurs when the OFF switching probability function pnf(x) becomes "2 / 3". In this case, the tendency is for the OFF switching threshold xthnf to be a value in a range of illuminance x where the set data at which the headlight 20 is in the ON state and the set data at which the headlight 20 is OFF state are present simultaneously. A-posteriori probability of the ON-switching class and A-posteriori probability of the OFF-switching class

[0080] In the embodiment described above, the a posteriori probability pfn(t = 1 | x) of the ON-class can be represented by a sigmoid function, and the independent variable of the sigmoid function is a linear function of the illuminance x. However, the present invention is not limited to this, and the independent variable can be a linear function of the illuminance LM. Furthermore, the a posteriori probability pfn(t = 1 | x) of the ON-class is not limited to the illuminance being the measured value that serves as the parameter for determining the independent variable. For example, weather information, road type information, or illuminance information within the vehicle can be included in the parameter.

[0081] In the embodiment described above, the a posteriori probability pnf(t = 1 | x) of the off-class can be represented by a sigmoid function, and the independent variable of the sigmoid function is a linear function of the illuminance x. However, the present invention is not limited to this, and the independent variable can be a linear function of the illuminance LM. Furthermore, the a posteriori probability pnf(t = 1 | x) of the off-class is not limited to a measured value that serves as the parameter for determining the independent variable being the illuminance. For example, weather information, road type information, or illuminance information within the vehicle can be included in the parameter. ON switching probability function and OFF switching probability function

[0082] In the embodiment above, the ON switching probability function pfn(x) was calculated based on the a posteriori distribution pfn(Wfn | T) of the ON switching class in process S70; this process can be omitted. In this case, for example, a sigmoid function in which the fitting parameter vector Wfn of the sigmoid function has been replaced by the MAP solution Wfnmap calculated in S66 can be replaced by the ON switching probability function pfn(x) in process S54. Furthermore, process S54 can, for example, be omitted.

[0083] Furthermore, the a posteriori probability pfn(t = 1 | x) of the ON-switching class is not used; the ON-switching probability function pfn(x) can be set as a sigmoid function using the fitting parameter vector Wfn, and the fitting parameter vector Wfn can be fitted directly based on the set data. This can be achieved by replacing "wfn0" with the current value "c0·{t-pfn(x)}" using a coefficient c0 and replacing "wfn1" with the current value "c1·{t-pfn(x)}·x" using a coefficient c1.

[0084] In the embodiment above, the OFF switching probability function pnf(x) was calculated based on the a posterior distribution pnf(Wnf | T) of the OFF switching in process S70a; this process can be omitted. In this case, for example, a sigmoid function in which the fitting parameter vector Wnf of the sigmoid function has been replaced by the MAP solution Wnfmap calculated in S66a can be replaced by the OFF switching probability function pnf(x) in process S48. Furthermore, process S48 can, for example, be omitted.

[0085] Furthermore, the a posteriori probability of the OFF switching class pnf(t=1 | x) is not used; the OFF switching probability function pnf(x) can be set as a sigmoid function using the fitting parameter vector Wnf, and the fitting parameter vector Wnf can be fitted directly based on the set data. This can be achieved by replacing "wnf0" with the current value "c0·{t-pnf(x)}" using the coefficient c0 and replacing "wnf1" with the current value "c1·{t-pnf(x)}·x" using the coefficient c1. Posterior distribution of ON switching and posterior distribution of OFF switching

[0086] In the embodiment above, the prior distribution of switching ON and switching OFF were updated by updating the means Mnf0, Mfn0 and the variance-covariance matrices Sfn0, Snf0 by the processes of S68 and S68a, but the present invention is not limited thereto. For example, the processes of S72 and S72a are omitted, and the prior distribution of switching ON and switching OFF is calculated using the n set data elements. Then, in a case where the n set data elements are recovered again when the A posteriori distribution of switching ON and the A posteriori distribution of switching OFF are calculated based on the 2n set data elements, it is not necessary to update the A posteriori distribution of switching ON and the A priori distribution of switching OFF. Building designation

[0087] In the third embodiment, the CPU 32 determines whether or not the vehicle 10 has entered the urban area, defined as a case where the vehicle 10 is within a predetermined distance of a housing that causes a rapid change in illuminance around the vehicle 10. However, the present invention is not limited to this. If detailed map information can be retrieved, it can be determined, for example, by retrieving position information about a building that has a height equal to or greater than a predetermined value, whether or not a distance from the building is within a predetermined distance. Furthermore, if the detailed map information can be retrieved, it can be determined, for example, by obtaining position information about the commercial building, whether or not a distance from a commercial building is within a predetermined distance. Set data used for learning

[0088] In processes S48 and S54, it is not essential to use the log likelihood. For example, CPU 32 can determine that the set data shows the anomalous value if the absolute value of a difference between the label variable t and the OFF switching probability function pnf(x) is relatively larger in process S48.

[0089] For example, in processes S48 and S54, if the negative log-likelihood is smaller than a predefined value that is smaller than the threshold lnth, the impact on the change in the ON switching probability function pfn(x) and the OFF switching probability function pnf(x) is relatively small, and the set data does not need to be used for learning. Thus, it is possible to reduce the computational load on CPU 32.

[0090] Furthermore, for example, a method of executing the process of S90, the process of S94 and the process of S96 is not limited to that which is depicted in the second or the third embodiment, and in short, the method can be a method of executing at least one of the three processes. Illuminance information indicating whether the illuminance is tending to increase or decrease.

[0091] In the embodiment described above, if the current time is "0:00 to 12:00", the set data is classified as OFF learning data; otherwise, the set data is classified as ON learning data. However, the present invention is not limited to this. For example, if the current time is "4:00 to 12:00", the set data is classified as OFF learning data; if the current time is "15:00 to 20:00", the set data is classified as ON learning data; and the set data in the other time zone can belong to both the OFF and ON learning data. In other words, the percentage contribution of the set data in the other time zone to the A posteriori distribution pnf (Wnf | T) of switching OFF and the percentage contribution of the set data to the A posteriori distribution pfn(Wnf | T) of switching ON can both be 50%.

[0092] The information about illuminance, indicating whether the illuminance tends to increase or decrease, is not limited to information about the time zone in which the set data is retrieved. For example, weather information can be included in addition to the time zone. Accordingly, in a time zone such as "10:00 to 15:00," where a change in illuminance is expected to be relatively small if the weather is good, then if the weather information, for example, indicates that the weather is rapidly deteriorating from good, and the set data retrieved during this time period can be classified as the EIN learning data, then the weather information indicates that the illuminance is decreasing. Illuminance information

[0093] The illuminance information required to determine the percentage contribution of the set data to the data relating to the ON-probability information, and the percentage contribution of the set data to the data relating to the OFF-probability information, is not limited to information indicating increases or decreases in illuminance. For example, the illuminance information could be the illuminance x assigned to the label variable t. This can be done, for example, using a blended model.

[0094] A probability function p(x) for the headlight entering an ON state, with respect to the illuminance x, is defined as "p(x) = [π1·pfn(x) + π2·pnf(x)" as the mixed model of the ON switching probability function pfn(x) and the OFF switching probability function pnf(x) using weighting coefficients π1 and π2. However, "π1 + π2 = 1". Here, the ON switching probability function pfn(x) and the OFF switching probability function pnf(x) can be a sigmoid function using the fitting parameter vectors Wfn and Wnf. A likelihood function P of the following equation (c5) is defined.

[0095] By maximizing the likelihood function P using a known P=∏k=1n(π1⋅pfn{x(k)}t(k)[1−pfn{x(k)}]1−t(k)+π2⋅pnf{x(k)}t(k)[1−pnf{x(k)}]1−t(k)) The EM algorithm can calculate both the weighting coefficients π1 and π2 as well as the fitting parameter vectors Wfn and Wnf. Given the set data (x, t), the EM algorithm then determines the percentage contribution of the set data to the ON switching probability function pfn(x) and the percentage contribution of the set data to the OFF switching probability function pnf(x), depending on whether the set data is fitted using either of these two functions. A mixing expert model can be constructed by setting the weighting coefficients π1 and π2 as a function of the illuminance x.

[0096] In the mixed model or the expert mixed model, information about whether the illuminance increases or decreases can be added. When adding this information to the expert mixed model, it is more effective if the weighting coefficients π1 and π2 are a function of time rather than illuminance x. However, the weighting coefficients π1 and π2 can also be a function of both time and illuminance. Update of the ON switching threshold xthfn and the OFF switching threshold xthnf

[0097] In the embodiment described above, a determination as to whether or not an update condition is satisfied is made based on the absolute value of the variance-covariance matrices Sfn0, Snf0, but the present invention is not limited to this. For example, the determination as to whether or not the update condition is satisfied can be made by determining a degree of variation in the fitting parameter based on an eigenvalue and an eigenvector of the variance-covariance matrices Sfn0, Snf0. Generation of tax data

[0098] In this embodiment, the a posteriori probability of the class of transition of the headlight to the ON state is represented by the sigmoid function, but the present invention is not limited thereto, and the a posteriori probability of the class of transition of the headlight to the OFF state can also be represented by the sigmoid function. Furthermore, the present invention is not limited to the use of the sigmoid function, and, for example, a probit function can be used.

[0099] Furthermore, the present invention is not limited to the use of the function that represents the probability of the headlight transitioning to the ON or OFF state. For example, a range of values ​​that can be assumed by the illuminance x can be divided into a plurality of ranges, the number of set data points in each range at which the headlight transitions to the ON state can be recorded, and a graph representing the probability of the headlight transitioning to the ON state can be generated using a ratio of the number of set data points at which the headlight transitions to the ON state to a total number of set data points. In this case, the total number of set data points used to generate the graph increases with the learning progress.Two graphs can be created: one based on set data obtained when the illuminance tends to decrease, and another based on set data obtained when the illuminance tends to increase. The former can be used as the ON-switching graph, and the latter as the OFF-switching graph. In this case, the ON-switching threshold xthfn, generated from the ON-switching graph, is assumed to be a value within the illuminance range where both the set data indicating that headlight 20 is ON and the set data indicating that headlight 20 is OFF are simultaneously present.It can also be assumed that the OFF switching threshold value xthnf, which is generated on the basis of the diagram for switching OFF, is a value in a range of illuminance x in which the set data where the headlight 20 is in the ON state and the set data where the headlight 20 is in the OFF state are present simultaneously.

[0100] However, the present invention is not limited to the process of generating the ON switching probability information and the OFF switching probability information. For example, data with a single piece of information indicating the probability that the headlight will switch to the ON state or the OFF state can be generated according to the illuminance x. In the data, a value obtained by adding a predetermined amount Δ to the illuminance x0, which is present when the probability of the headlight switching to the ON state becomes "1 / 2", can be used as the ON switching threshold xthfn, and a value obtained by subtracting the predetermined amount Δ from the illuminance x0 can be used as the OFF switching threshold xthnf.

[0101] Furthermore, the present invention is not limited to generating the probability information, and a process can be carried out in which a range in which set data at which the headlight 20 transitions to the ON state and set data at which the headlight 20 transitions to the OFF state are simultaneously present are specified by dividing a range of values ​​that can be assumed by the illuminance x into a plurality of ranges and recording the number of data elements with which the headlight 20 transitions to the ON state and the number of data elements with which the headlight 20 transitions to the OFF state in each range. That is, in this case, for example, the ON switching threshold xthfn and the OFF switching threshold xthnf can be set in the middle of a range in which both set data are simultaneously present.More precisely, the set data can be classified as ON learning data and OFF learning data, whereby the number of data elements that belong independently to the respective areas can be recorded in the ON learning data and the OFF learning data, the ON switching threshold xthfn can be generated from the ON learning data and the OFF switching threshold xthnf can be generated from the OFF learning data.However, the classification as ON learning data and as OFF learning data is not performed, and, for example, in the area where both the set data at which the headlight 20 switches to the ON state and the set data at which the headlight 20 switches to the OFF state are present simultaneously, an illuminance higher than that in its center can be used as the ON switching threshold xthfn, and in the area where both the set data at which the headlight 20 switches to the ON state and the set data at which the headlight 20 switches to the OFF state are present simultaneously, an illuminance lower than that in its center can be used as the OFF switching threshold xthnf. Automatic mode

[0102] In the Fig. In example 9, the language recommendation was implemented as a recommendation process, but the present invention is not limited to this. For example, visual information representing the content of the recommendation can be displayed as a virtual image in front of a windshield using a windshield display.

[0103] Furthermore, the present invention is not limited to carrying out either the automatic switching or the recommendation process. For example, if the CPU 32 is in S24 of Fig. 2 If a positive determination is made, the CPU 32 can first output an audio signal “May the spotlight 20 be switched on?” from the speaker 52 and automatically switch the spotlight 20 from the OFF state to the ON state if the user gives a positive answer, for example using a speech recognition function. Vehicle control device

[0104] In the embodiments described above, the device for generating headlight control data and the device that executes the automatic mode share the same ECU 30, but the present invention is not limited to this. For example, a device that executes the automatic mode can be another ECU in the vehicle. Furthermore, the device for generating headlight control data and the device that executes the automatic mode need not necessarily be installed in the vehicle 10. For example, the device for generating headlight control data can be an external device that receives data from the vehicle 10 by communicating with it and generates the headlight control data based on the set data. In this case, the headlight control data output by the external device to the vehicle 10 can be the ON switching threshold xthfn and the OFF switching threshold xthnf.

[0105] Furthermore, the external device can be a large computer that collects and analyzes big data, which consists of the set data of a large number of vehicles, but it can also be, for example, a mobile device. This can be achieved, for example, by installing an application program that causes a computer to execute a process according to the one described in Fig. The process shown in steps 3 to 6 is implemented in the mobile device 60. Electronic control unit

[0106] In the embodiment above, the ROM is represented as a memory that is part of the electronic control unit, and the type of ROM is not specified. For example, the ROM could be a read-only memory or an electrically rewritable non-volatile memory.

[0107] The electronic control unit is not limited to an electronic control unit designed to include a program storage device, such as a ROM that stores a program, and a CPU that executes the program. For example, the electronic control unit is not limited to a software processing circuit that includes the CPU and the program storage device, and may, for example, be a dedicated hardware circuit that executes a predefined process, such as an ASIC. Furthermore, the electronic control unit may include both the software circuit and the dedicated hardware processing circuit. storage device

[0108] A data storage device with respect to the a priori distribution is not limited to an electrically rewritable non-volatile storage device and can, for example, be a backup RAM that maintains power supply even when a main power supply of the ECU 30 transitions to the OFF state. Furthermore, the storage device can include a RAM that receives power only when a main power supply transitions to the ON state and the non-volatile storage device. In this case, for example, the data that is stored in the RAM before the main power supply transitions to the OFF state can be stored in the non-volatile storage device. Other

[0109] In this embodiment, when a plurality of users use one and the same vehicle 10, the ON switching threshold xthfn and the OFF switching threshold xthnf are generated for each user, but the present invention is not limited to this, and in one and the same vehicle 11 a single ON switching threshold xthfn and a single OFF switching threshold xthnf could also be used when the user changes.

Claims

[1] Device for generating headlight control data, which generates headlight control data of a vehicle (10) having a headlight (20), wherein the headlight control data are data used in an automatic mode in the vehicle (10), which selects between a manual mode in which a user manually performs an actuation to switch the headlight (20) from one ON state or one OFF state to the other, and the automatic mode in which a process of automatically performing the switching actuation and / or a process of outputting a recommendation message signal prompting the switching actuation to a message device (52) is performed, wherein the device for generating headlight control data comprises: an electronic control unit (30) designed to then, when manual mode is selected, set data including a detection value of an illuminance (LM) around the vehicle (10) and ON and OFF data indicating that the headlight (20) is in the ON state or the OFF state at the time the detection value is requested, and to generate the headlight control data based on the retrieved set data, wherein: the headlight control data include data relating to information about the probability of a headlight (20) being in the ON state; and where the information about a probability that a state of the headlight (20) is ON is information in which the probability that the state of the headlight (20) is ON increases when the illuminance of the headlight is within a specified range, when a proportion of the set data indicating that the state of the headlight (20) is ON is relatively high compared to when the proportion of the set data indicating that the illuminance (LM) detection value is within a specified range is relatively low. [2] Device for generating headlight control data according to claim 1, wherein an ON switching threshold (xthfn), which is an illuminance threshold for switching from the ON state to the OFF state in automatic mode, and an OFF switching threshold, which is an illuminance threshold for switching from the OFF state to the ON state in automatic mode, can be set independently of each other; the headlight control data include data that includes information about the probability that the headlight (20) is ON, data relating to the ON switching probability information for generating the ON switching threshold (xthfn), and data relating to the OFF switching probability information for generating the OFF switching threshold (xthnf); and the electronic control unit (30) is designed to further determine, based on information about the illuminance around the vehicle (10), a percentage contribution of the retrieved set data to the ON switching probability information and a percentage contribution of the retrieved set data to the OFF switching probability information. [3] Device for generating headlight control data according to claim 2, wherein: The information about illuminance is information that indicates whether the illuminance (LM) tends to decrease or whether the illuminance (LM) tends to increase; and the electronic control unit (30) is designed to further determine whether the illuminance (LM) tends to decrease or tends to increase. [4] Device for generating headlight control data according to claim 3, wherein the electronic control unit (30) is designed to determine, on the basis of a time zone, whether the illuminance (LM) tends to increase or tends to decrease. [5] Device for generating headlight control data according to claim 3 or 4, wherein the electronic control unit (30) is designed to to calculate an a posteriori distribution of a fitting parameter, an a posteriori probability of an ON-switching class, based on the set data, the a posteriori probability of the ON-switching class, and an a priori distribution of the ON-switching, and to calculate an a posterior distribution of an adjustment parameter, an a posterior probability of an off-class based on the set data, the a posterior probability of the off-class and an a priori distribution of the off-class; the a posteriori probability of the ON-switching class is a function in which a detection value of the illuminance (LM) is an independent variable and a probability of the ON state or the OFF state of the headlight (20) is a dependent variable, wherein the function is a function for switching from the OFF state to the ON state; The A priori distribution of the ON switching is a probability distribution of the adjustment parameter for adjusting a value of the dependent variable with respect to the value of the independent variable of the A posteriori probability of the ON switching class; the a posteriori probability of the off-class is a function in which a detection value of the illuminance (LM) is an independent variable and a probability of the state ON or the state OFF of the headlight (20) is a dependent variable, wherein the function is a function for switching from the state ON to the state OFF; The A priori distribution of the switching-off is a probability distribution of the adjustment parameter for adjusting a value of the dependent variable with respect to the value of the independent variable of the A posteriori probability of the switching-off class; the data relating to ON switching probability information are data on the a posteriori distribution of the fitting parameter in the a posteriori probability of the ON switching class; and The data relating to the OFF-switching probability information are data on the A-posteriori distribution of the fitting parameter in the A-posteriori probability of the OFF-switching class. [6] Device for generating headlight control data according to any one of claims 1 to 5, wherein: the electronic control unit (30) is designed to further determine whether a vehicle speed (10) is equal to or lower than a threshold value; and the electronic control unit (30) is designed to use the retrieved set data to generate the headlight control data, provided that it is not determined that the vehicle speed (10) is equal to or below the threshold. [7] Device for generating headlight control data according to any one of claims 1 to 6, wherein: the electronic control unit (30) is designed to further determine whether or not the vehicle (10) is driving through a tunnel; and the electronic control unit (30) is designed to use the retrieved set data to generate the headlight control data under the condition that that it is not determined that the vehicle will travel through a tunnel. [8] Device for generating headlight control data according to any one of claims 1 to 7, wherein: the electronic control unit (30) is designed to further determine whether or not a given building is present within a given distance from the vehicle (10); and the electronic control unit (30) is designed to use the retrieved set data to generate the headlight control data under the condition that that the electronic control unit (30) does not determine that the specified building exists. [9] Vehicle control device comprising the device for generating headlight control data according to any one of claims 2 to 5, wherein the electronic control unit (30) is designed to set the ON switching threshold (xthfn) and the OFF switching threshold (xthnf) based on the headlight control data, and is designed to perform a switching from the OFF state to the ON state of the headlight (20) in automatic mode under the condition that the illuminance (LM) detection value is less than the ON switching threshold (xthfn) when the headlight (20) is in OFF mode, and to perform a switching from the ON state to the OFF state of the headlight (20) in automatic mode under the condition that the illuminance (LM) detection value is greater than the OFF switching threshold (xthnf) when the headlight (20) is in ON mode. [10] Device for generating headlight control data, which generates headlight control data of a vehicle (10) having a headlight (20), wherein the headlight control data are data that are generated in an automatic mode in the The vehicle (10) is used, which selects between a manual mode in which a user manually performs an actuation to switch the headlight (20) from one state ON or one state OFF to the other, and the automatic mode in which a process of automatically performing the switching actuation and / or a process of issuing a recommendation message signal prompting the switching actuation to a communication device (52) is carried out, the device comprising: an electronic control unit (30) designed to then, when manual mode is selected, set data including a detection value of an illuminance (LM) around the vehicle (10) and ON and OFF data indicating that the headlight (20) is in the ON state or the OFF state at the time the detection value is requested, and to generate the headlight control data based on the retrieved set data, wherein: The headlight control data includes a threshold for switching in automatic mode; and the electronic control unit (30) is designed to set the threshold in a range of illuminance (LM) in which both the set data where the state of the headlight (20) is ON and the set data where the state of the headlight (20) is OFF are present simultaneously. [11] Vehicle control device comprising the device for generating headlight control data according to claim 10, wherein the electronic control unit (30) is designed to perform a switching to the other state from ON or from the state OFF of the headlight (20) based on a size comparison between the detected illuminance (LM) value and the threshold value when the headlight (20) is in automatic mode in the state ON or in the state OFF.

Images