JP2007137278A - Driving intension estimating device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a driving intension estimating device for suppressing deterioration in estimation precision of driving intension when a vehicle makes cornering. <P>SOLUTION: A driving intension estimating device estimating driving intension on the basis of at least either a value of acceleration in an anteroposterior direction or a value of lateral acceleration to be operated to the vehicle estimates the driving intension on the basis of values differing in influence degree of the value of the acceleration in the anteroposterior direction and the value of the lateral acceleration when the vehicle makes cornering and does not make cornering (S5, S8, S10, S13). <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a driving orientation estimation device that estimates driving orientation, and more particularly to a driving orientation estimation device that can suppress a decrease in estimation accuracy of driving orientation when a vehicle turns.

  Japanese Laid-Open Patent Publication No. 9-242863 (Patent Document 1) is a driving direction estimation device for estimating driving direction of a vehicle, and includes an output operation amount at the time of starting the vehicle, a maximum change rate of the output operation amount, Calculated by the driving operation related variable calculating means for calculating at least one driving operation related variable selected from the maximum deceleration during braking operation, the coasting traveling time of the vehicle, and the constant traveling speed of the vehicle, and the driving operation related variable calculating means There is disclosed a driving orientation estimation device for a vehicle that includes a neural network to which a driving operation related variable is input and includes driving orientation estimation means for estimating the driving orientation of the vehicle based on the output of the neural network.

JP-A-9-242863 JP 2005-98364 A

  For example, a technique for estimating the driving direction of a driver by inputting information indicating the amount of operation of the driver and the amount of state of the vehicle into an information processing mechanism using an intelligent system such as a neural network is known. Even if the driver's driving direction has not actually changed, the amount of change in the accelerator operation may be small while the vehicle is turning (during cornering). Regardless of the driving orientation, a value indicating that the driving orientation has changed is output from the information processing mechanism.

  Here, when the driving direction does not actually change as described above and the accelerator operation change amount becomes small during the turning of the vehicle, a value indicating that the driving direction has not changed is output. It is conceivable to create an information processing mechanism. However, in that case, it is necessary to increase the items to be input to the information processing mechanism, which causes a problem that the calculation load increases. In addition, there is a problem that man-hours increase when such an information processing mechanism is newly created. On the other hand, since the driving direction may actually change during the turning of the vehicle, it is desired that the driving direction during the turning of the vehicle can be estimated with higher accuracy.

An object of the present invention is to provide a driving orientation estimation apparatus capable of suppressing a decrease in driving orientation estimation accuracy when a vehicle turns.
Another object of the present invention is to provide a driving orientation estimation device capable of suppressing a reduction in estimation accuracy of driving orientation when a vehicle turns without increasing a calculation load or man-hours.

  The driving orientation estimation device of the present invention is a driving orientation estimation device that estimates driving orientation based on at least one of a value corresponding to longitudinal acceleration acting on a vehicle and a value corresponding to lateral acceleration, When the vehicle is turning and when the vehicle is not turning, the driving direction is estimated based on values having different influence levels of the value corresponding to the longitudinal acceleration and the value corresponding to the lateral acceleration.

  In the driving orientation estimation apparatus of the present invention, a predetermined relationship is set in which a relationship between a value corresponding to the longitudinal acceleration and a value corresponding to the lateral acceleration is set in advance when the vehicle is not turning and when the vehicle is turning. When the relationship is, the driving direction is estimated based on values having different degrees of influence of the value corresponding to the longitudinal acceleration and the value corresponding to the lateral acceleration.

  In the driving orientation estimation device of the present invention, when the relationship between the value corresponding to the longitudinal acceleration and the value corresponding to the lateral acceleration is a predetermined relationship set in advance, the value corresponding to the lateral acceleration is The present invention is characterized in that it relates to a magnitude relationship between a product of a predetermined value set in advance and a value corresponding to the acceleration in the longitudinal direction.

  In the driving orientation estimation apparatus of the present invention, a predetermined relationship is set in which a relationship between a value corresponding to the longitudinal acceleration and a value corresponding to the lateral acceleration is set in advance when the vehicle is not turning and when the vehicle is turning. When the value corresponding to the acceleration in the longitudinal direction is a value equal to or greater than 0, the degree of influence of the value corresponding to the acceleration in the forward direction and the value corresponding to the lateral acceleration in the longitudinal direction is The driving direction is estimated based on different values.

  In the driving orientation estimation apparatus of the present invention, a predetermined relationship is set in which a relationship between a value corresponding to the longitudinal acceleration and a value corresponding to the lateral acceleration is set in advance when the vehicle is not turning and when the vehicle is turning. When the value corresponding to the acceleration in the longitudinal direction is less than 0, the degree of influence of the value corresponding to the acceleration in the backward direction and the value corresponding to the lateral acceleration in the longitudinal direction is The driving direction is estimated based on different values.

  In the driving orientation estimation apparatus of the present invention, the predetermined value set in advance is set to a different value based on at least one of a road environment and a traveling environment.

  In the driving orientation estimation device of the present invention, when the driver can easily perform an operation of returning the accelerator relative to at least one of the road environment and the traveling environment, the predetermined value set in advance is It is characterized by being set to a relatively large value.

  In the driving orientation estimation device of the present invention, when a relatively large lateral acceleration is likely to act on the vehicle in relation to at least one of the road environment and the traveling environment, the predetermined value set in advance is a relative value. It is characterized by being set to a small value.

  In the driving orientation estimation apparatus according to the present invention, the road environment and the traveling environment include at least one of a road surface gradient, a road surface state, a degree of road curvature, curve continuity, weather, and brightness.

  According to the driving orientation estimation device of the present invention, it is possible to suppress a decrease in driving orientation estimation accuracy when the vehicle turns.

  Hereinafter, an embodiment of a driving orientation estimation apparatus according to the present invention will be described in detail with reference to the drawings.

  The driving orientation estimation apparatus according to the present embodiment uses the driver's operation amount and vehicle state quantity (including longitudinal acceleration G) as input information of an information processing mechanism using an intelligent system such as a neural network, for example. A driving orientation estimation device for estimating the acceleration operation change amount when the vehicle is turning, even if the driving orientation of the driver has not changed. Acceleration / deceleration (back-and-forth acceleration G) decreases. For this reason, the input terms of the information processing mechanism using an intelligent system such as a neural network, the acceleration input term and the deceleration input term are reduced, and as a result, the driver's driving orientation has not changed. The output value (driver directivity) will decrease.

  In order to suppress the above problems, it is conceivable to newly add signals indicating characteristics during turning such as lateral G (turning acceleration G) and steering angle to items to be input to the neural network. Therefore, there is a large amount of teacher data, and as a result, there is a possibility that the number of matching man-hours (man-hours required for learning) increases and the calculation load during calculation by the neural network increases.

  Therefore, in the present embodiment, the lateral G is input instead of the longitudinal acceleration G as the input value of the neural network under a predetermined condition. The horizontal G is a value reflecting the driving orientation during cornering. When the driving orientation is sports driving orientation, even if the accelerator operation change amount is small, the lateral G is large, so the decrease in the output of the neural network is suppressed, and the driving orientation estimation result and the actual driver Can be reduced.

  Input values related to the accelerator, such as the amount of accelerator operation and the accelerator opening that have decreased during cornering, are input to the neural network as usual, but the lateral G is input instead of the longitudinal acceleration G under predetermined conditions. As a result, a decrease in the output of the neural network is not a problem.

  In this embodiment, when “turn determination is established and | lateral G | × constant> front / rear acceleration G> 0”, | lateral G | × constant is input to the acceleration input term instead of the longitudinal acceleration G (normalization) In the case of “when turning determination is established and | lateral G | × constant> | longitudinal acceleration G | (longitudinal acceleration G <0)”, instead of the longitudinal acceleration G, the | lateral G | × constant is set. Input to deceleration input term (normalized value is changed).

  Thus, in the present embodiment, instead of newly adding the lateral G as an item to be input to the neural network, a value corresponding to the lateral G is substituted instead of the longitudinal acceleration G under a predetermined condition. This eliminates the need to give a new lateral G as teacher data when learning a neural network, thereby enabling collection of teacher data and shortening of the learning time. In addition, without inputting the lateral G information as a new input term into the neural network, it is possible to avoid an increase in the number of input values for the neural network by inputting from the existing acceleration input term and deceleration input term. Therefore, it is possible to suppress an increase in calculation load without reducing the driving-oriented estimation accuracy.

  In the present embodiment, as described above, the acceleration term or the deceleration term is prevented from becoming excessive by multiplying by a constant (≦ 1) in preparation for the case where | lateral G | is too large. In the above, “normalized value is changed” means that a value normalized to, for example, a value between −1 and +1 or a value between 0 and 1 is input to the acceleration input term or the deceleration input term. This means that when normalization is performed, the value obtained by dividing the actual longitudinal acceleration G value (normalized value) is different from the value obtained by dividing | lateral G | × constant value (normalized value). .

  FIG. 2 shows a schematic configuration diagram of a control device for an automatic transmission to which the driving orientation estimation device of the present embodiment is applied. In FIG. 2, reference numeral 10 is a stepped automatic transmission, and 40 is an engine. The automatic transmission 10 is capable of five-speed shifting by controlling the hydraulic pressure by energization / non-energization of the solenoid valves 121a, 121b, and 121c. In FIG. 2, three electromagnetic valves 121a, 121b, and 121c are illustrated, but the number of electromagnetic valves is not limited to three. The solenoid valves 121a, 121b, and 121c are driven by a signal from the control circuit 130.

  The throttle opening sensor 114 detects the opening of the throttle valve 43 disposed in the intake passage 41 of the engine 40. The engine speed sensor 116 detects the speed of the engine 40. The vehicle speed sensor 122 detects the rotation speed of the output shaft 120c of the automatic transmission 10 that is proportional to the vehicle speed. The shift position sensor 123 detects the shift position. The pattern select switch 117 is used when instructing a shift pattern. The acceleration sensor 90 detects the longitudinal acceleration and turning acceleration (lateral G) of the vehicle. The road surface μ detection / estimation unit 112 detects or estimates the friction coefficient μ of the road surface or the slipperiness of the road surface. The accelerator opening detector 113 detects the accelerator opening. The brake operation amount detection unit 111 detects the operation amount of the brake.

  The road surface μ detection / estimation unit 112 detects or estimates the ease of slipping of the road surface (whether the road surface is a low μ road) represented by the road surface friction coefficient μ. Here, the low μ road includes a bad road (including a case where the road surface has large unevenness or a step on the road surface). That is, the road surface μ detection / estimation unit 112 calculates the friction coefficient μ of the traveling road surface, and determines whether the road is a low μ road depending on whether the calculated friction coefficient μ exceeds a predetermined threshold value. Is determined.

  In the road surface μ detection / estimation unit 112, instead of the above, the front wheel (not shown) detected by various conditions, for example, a front wheel speed sensor (not shown), without obtaining a specific numerical value of the friction coefficient μ by calculation. Whether or not the road surface is a low μ road based on the difference between the rotation speed (driven wheel speed) and the rotation speed (drive wheel speed) of the rear wheels (not shown) detected by the vehicle speed sensor 122. Can be detected.

  Here, the specific method of detecting / estimating whether or not the road surface μ detecting / estimating unit 112 is a low μ road is not particularly limited, and a known method can be appropriately employed. For example, in addition to the wheel speed difference before and after the above, the rate of change of wheel speed, ABS (anti-lock brake system), TRS (traction control system) and VSC (vehicle stability control) operation It is possible to detect / estimate whether the road is a low μ road by using at least one of the relationship between the history, the acceleration of the vehicle, and the wheel slip ratio.

  An example of such a detection / estimation method for determining whether the road is a low μ road is disclosed in Japanese Patent Laid-Open Nos. 5-223157, 8-121582, 10-94110, and 2000-79834. Japanese Patent No. 2780390, Japanese Patent Laid-Open No. 5-346394, and Japanese Patent Laid-Open No. 6-115417.

  The road surface μ detection / estimation unit 112 predicts whether or not the road surface is a low μ road, based on information (navigation information or the like) about a road surface scheduled to travel in the future. Here, the navigation information includes information on road surfaces (for example, non-paved roads) recorded in advance on a storage medium (DVD, HD, etc.) as in the navigation system device 95, as well as past actual driving and other information. Information (including information indicating road conditions and information indicating weather conditions) obtained through communication (including vehicle-to-vehicle communication and road-to-vehicle communication) with other vehicles and communication centers. Such communications include road traffic information communication systems (VICS) and so-called telematics.

  The navigation system device 95 has a basic function of guiding the host vehicle to a predetermined destination, and includes an arithmetic processing device and information (map, straight road, curve, uphill / downhill, highway) necessary for traveling the vehicle. Etc.), a first information detection device including a geomagnetic sensor, a gyrocompass, and a steering sensor, and a current position of the vehicle by radio navigation. It is for detecting a position, road conditions, etc., and is provided with a second information detection device including a GPS antenna and a GPS receiver.

  The control circuit 130 is a signal indicating the detection results of the throttle opening sensor 114, the engine speed sensor 116, the vehicle speed sensor 122, the shift position sensor 123, the acceleration sensor 90, the accelerator opening detection unit 113, and the brake operation amount detection unit 111. , A signal indicating the switching state of the pattern select switch 117, a signal from the navigation system device 95, and a signal indicating the detection or estimation result by the road surface μ detection / estimation unit 112. input.

  The control circuit 130 is configured by a known microcomputer and includes a CPU 131, a RAM 132, a ROM 133, an input port 134, an output port 135, and a common bus 136. The input port 134 has signals from the sensors 114, 116, 122, 123, 90 described above, a signal from the accelerator opening detection unit 113, a signal from the brake operation amount detection unit 111, and a signal from the switch 117 described above. The signal from the navigation system device 95 and the signal from the road surface μ detection / estimation unit 112 are input. Solenoid valve driving units 138a, 138b, and 138c are connected to the output port 135.

  The road gradient measurement / estimation unit 118 can be provided as a part of the CPU 131. Here, the specific method of measuring / estimating the road gradient by the road gradient measuring / estimating unit 118 is not particularly limited, and a known method can be appropriately adopted. For example, the road gradient measurement / estimation unit 118 can measure or estimate the road gradient based on the acceleration detected by the acceleration sensor 90. Further, the road gradient measuring / estimating unit 118 may store the acceleration on the flat road in the ROM 133 in advance and obtain the road gradient by comparing with the acceleration actually detected by the acceleration sensor 90. Further, the road gradient measuring / estimating unit 118 can obtain road gradient information from the navigation information.

  The driving orientation estimation unit 115 can be provided as a part of the CPU 131. The driving direction estimation unit 115 estimates the driving direction (sport driving direction or normal driving direction) of the driver based on the driving state of the driver and the driving state of the vehicle. Details of the driving orientation estimation unit 115 will be described later. Note that the configuration of the driving orientation estimation unit 115 is not limited to the content described later, and any configuration having various known configurations (for example, the configuration described in Patent Document 1 above) can be used as long as the driving orientation of the driver is estimated. ) Widely. Here, the term “sports driving orientation” refers to a direction that emphasizes power performance, an acceleration direction, or a preference for sports driving in which the response of the vehicle to the driver's operation is quick.

  In ROM 133, a program in which the operation (control step) shown in the flowchart of FIG. 1 is described in advance, data (not shown) of a plurality of shift diagrams that are switched according to the estimated driving orientation, accelerator opening, and engine A map of rotation speed is stored. The control circuit 130 shifts the automatic transmission 10 based on various input control conditions and a shift diagram.

Next, details of the driving orientation estimation unit 115 will be described.
The driving orientation estimation unit 115 includes a neural network NN in which the driving operation related variable is input and an estimation calculation is started every time one of a plurality of types of driving operation related variables is calculated, and based on the output of the neural network NN. To estimate the driving direction of the vehicle.

  For example, as shown in FIG. 3, the driving direction estimation unit 115 includes a signal reading unit 96, a preprocessing unit 98, and a driving direction estimation unit 100. The signal reading means 96 reads detection signals from the sensors / detectors 90, 111, 112, 113, 114, 116, 122, 123 and the like at a relatively short predetermined cycle. The detection signal read by the signal reading means 96 is output to the preprocessing means 98.

The pre-processing means 98 uses a plurality of types of driving operation-related variables closely related to the driving operation reflecting the driving direction from the signals sequentially read by the signal reading means 96, that is, the output operation amount (accelerator pedal operation) when starting the vehicle. Amount), that is, the throttle valve opening TA _ST when the vehicle _starts , the maximum change rate of the output operation amount during acceleration operation, that is, the maximum change rate A _CCMAX of the throttle valve opening, the maximum deceleration G _NMAX when braking the vehicle, _Coasting travel time T _COAST , constant vehicle speed travel time T _VCONST , section maximum value of signals input from each sensor within a predetermined section, maximum vehicle speed V _max after starting operation, vehicle acceleration term (acceleration input term), vehicle Is a driving operation related variable calculating means for calculating the deceleration term (deceleration input term) and the like. The driving orientation estimation unit 100 includes a neural network NN that performs the driving orientation estimation calculation by permitting the driving operation related variable every time the driving operation related variable is calculated by the preprocessing unit 98, and outputs the neural network NN. A certain driving direction estimation value is output.

The preprocessing means 98 in FIG. 3, the starting time of the output control input calculation means 98a for calculating the throttle valve opening TA _ST when the output operation amount i.e. vehicle starting when the vehicle start, the maximum change in the output operation amount when the acceleration operation rate i.e. accelerating operation when the output operation amount maximum change rate calculating means 98b for calculating the maximum change rate a _CCmax of the throttle valve opening, braking maximum deceleration calculating means for calculating the maximum deceleration G _NMAX during braking operation of the vehicle 98c , input signals from the sensors in the coasting time calculation means 98d, constant vehicle speed running time calculating means 98e for calculating the constant vehicle speed running time T _VCONST, for example 3 seconds to a predetermined section within which calculates the coasting time T _COAST vehicle maximum value periodically calculated to the input signal interval maximum value calculating means 98f of the maximum vehicle speed calculating means 98g for calculating the maximum vehicle speed V _max in operation since the start of the vehicle An acceleration input term calculating unit 98h for calculating an acceleration term, a deceleration input term calculating unit 98i for calculating a vehicle deceleration term, and the like are provided.

As the maximum value of the input signals in the predetermined section calculated by the input signal section maximum value calculating means 98f, the throttle valve opening TA _maxt , the vehicle speed V _maxt , and the engine speed N _Emaxt are used.

  The neural network NN provided in the driving orientation estimation means 100 of FIG. 3 can be configured by modeling a living nerve cell group by software based on a computer program or hardware consisting of a combination of electronic elements. For example, it is comprised so that it may be illustrated in the block of the driving | operation direction estimation means 100 of FIG.

In FIG. 3, the neural network NN includes an input layer composed of _r nerve cell elements (neurons) X _i (X _{1 to} X _r ) and s nerve cell elements Y _j (Y _{1 to} Y _s ). Is a three-layered hierarchical type composed of an intermediate layer composed of _t and an output layer composed of _t neuron elements Z _k (Z _{1 to} Z _t ). In order to transmit the state of the nerve cell element from the input layer to the output layer, the r nerve cell elements X _i and s nerve cell elements Y having a coupling coefficient (weight) W _Xij are provided. a transfer element D _Xij coupling the _j respectively, the coupling coefficient (weight) W _Yjk the have the s neuronal elements Y _j and t pieces of transmission elements D _Yjk of neuronal elements Z _k and the coupling respectively Is provided.

The neural network NN is its coupling coefficient (weight) W _Xij, pattern associative system that is made to learn the coupling coefficient (weight) W _Yjk called backpropagation learning algorithm. Learning, so are allowed to advance completed by running experiments in matching driving manner and the value of the driving operation related variables, during vehicle assembly, the coupling coefficient (weight) W _Xij, the coupling coefficient (weight) W _Yjk Is given a fixed value.

  In the above learning, sports-oriented driving and normal driving (normal) -oriented driving are carried out for a plurality of drivers on various roads such as highways, suburban roads, mountain roads, and city roads, respectively. With the directivity as a teacher signal, n indicators (input signals) obtained by pre-processing the teacher signal and the sensor signal are input to the neural network NN. The teacher signal is converted into a value from 0 to 1 for driving orientation. For example, normal driving orientation is 0 and sports driving orientation is 1. The input signal is used after being normalized to a value between -1 and +1 or between 0 and 1 (in this embodiment, it is used after normalizing to a value between 0 and 1). ).

  In the prior art, as shown in FIG. 4, in the case where “turn determination 211 is turned on at time t1 and the accelerator operation amount (not shown) is small”, the longitudinal acceleration G or As shown in the graph 213 of the turning acceleration G (lateral G), the longitudinal acceleration G221 decreases (see reference C1). Therefore, as shown in the | acceleration term | graph 214 and the | deceleration term | graph 215, the | front / rear acceleration G | 231 (the absolute value of the front / rear acceleration G221) also decreases. The values of | acceleration term | and | deceleration term | input (see symbols 98h and 98i in FIG. 3) input to (see symbol 100) are lowered. As a result, as shown in the driver directivity graph 216, the driver directivity 251 (the output of the reference numeral 100 in FIG. 3) decreases.

In the above, | front-rear acceleration G | 231 is described in each of graph 214 of | acceleration term | and graph 215 of | deceleration term |. This | back-and-forth acceleration G | 231 is represented by a locus of points in FIG. 4. From the point P ₄₀ at the time t0 to the point P ₄₈ at the time t8 in the graph 215 of the | deceleration term | corresponding terms P ₄₉ of time t9 graph 214 to a line connecting a to the point P ₅₂ of the time t12.

In the graph 213 of the longitudinal acceleration G or the turning acceleration G, as shown by the longitudinal acceleration G221, the deceleration G acts from the time t0 to the time t8, and the acceleration G acts after the time t9. Therefore, in the prior art, is input to the neural network | acceleration term | has a value, | acceleration term | t8 time from time t0 of the graph 214 is 0 as shown from the point P ₃₀ to a point P _38, t9 point forward, from the point P _49, as shown in P ₅₂ | is the same value as 231 | longitudinal acceleration G. Moreover, inputted into the neural network | deceleration section | values, | deceleration section | t8 time from time t0 of the graph 215, as shown from the point P ₄₀ to a point P ₄₈ | longitudinal acceleration G | 231 is the same value as, since time t9, a 0 as shown from the point P ₂₄₉ to P _252. Since the value input to the neural network is limited to a positive value, the absolute value of the deceleration is input to the deceleration term of the neural network.

  In contrast, in the present embodiment, as shown in FIGS. 1 and 4, “when the turning determination 211 is established (t1 in FIG. 4) and | longitudinal acceleration G | 231 <| lateral G | × constant 233”. 1 (step S3-Y in FIG. 1, t3 to t11 in FIG. 4), by substituting the value of | lateral G | × constant 233 for each of the acceleration term and deceleration term (step S5, step S8), Each decrease in the acceleration term and the deceleration term is suppressed (C2, C4 in FIG. 4), and as a result, the decrease in the driver directivity 216 is suppressed (C6).

  Hereinafter, the operation of the present embodiment will be described in detail with reference to FIGS. 1, 2, and 4.

[Step S1]
In step S1 of FIG. 1, the longitudinal acceleration G221 is calculated. The longitudinal acceleration G221 can be detected by the acceleration sensor 90. The longitudinal acceleration G221 is shown in the longitudinal acceleration G or turning acceleration G graph 213 of FIG. After step S1, the process proceeds to step S2.

[Step S2]
In step S2, the lateral G × constant value 223 is calculated. First, the lateral G222 is calculated, and the lateral G222 is multiplied by a constant, whereby the lateral G × constant value 223 is calculated. The lateral G222 can be detected by the acceleration sensor 90. The lateral G × constant value 223 is calculated by the acceleration input term calculation unit 98h and the deceleration input term calculation unit 98i of the driving direction estimation unit 115 of the control circuit 130. After step S2, the process proceeds to step S3.

[Step S3]
In step S3, the control circuit 130 determines whether or not | longitudinal acceleration G | 231 <| lateral G | × constant 233AND turning determination 211 = ON. If the result of the determination is affirmative, the process proceeds to step S4, and if not, the process proceeds to step S9.

  The | front / rear acceleration G | 231 is the absolute value of the front / rear acceleration G221 obtained in step S1. | Horizontal G | × constant 233 is the product of the absolute value of the lateral G222 obtained in step S2 and the constant, and corresponds to the absolute value of lateral G × constant value 223 obtained in step S2. The | lateral G | × constant 233 is shown in the | acceleration term | graph 214 and | deceleration term | graph 215.

  Of the conditions in step S3, the turning determination 211 is ON at time t1. The control circuit 130 can determine the turning determination 211 based on the size of the lateral G222 of the vehicle. Alternatively, the turning determination 211 can be determined based on data input from the navigation system device 95 and indicating the current position of the vehicle and the position of the entrance of the corner. Also, as shown in the | acceleration term | graph 214 and the | deceleration term | graph 215, the | lateral G | × constant 233 exceeds the | front-rear acceleration G | 231 at the intermediate point between t2 and t3. As shown in the graph 212 of the longitudinal acceleration G | <| lateral G | × constant, the determination of | longitudinal acceleration G | <| lateral G | × constant is ON at time t3. From these things, step S3 begins to be determined affirmatively from time t3 (from step t0 to t2, step S3 is negatively determined).

  On the other hand, the turning determination 211 is OFF at time t12. In addition, as shown in the | acceleration term | graph 214 and the | deceleration term | graph 215, the | lateral G | × constant 233 is equal to or less than the longitudinal acceleration G | 231 at an intermediate time between t10 and t11. As shown in the graph 212 of | front / rear acceleration G | <| lateral G | × constant, the determination of | front / rear acceleration G | <| lateral G | × constant is OFF at time t11. From these things, step S3 is negatively determined from the time t11.

[Step S4]
In step S4, the control circuit 130 determines whether or not the longitudinal acceleration G221 ≧ 0. The case where the determination in step S4 is positive is the range from t9 to t10 in which the longitudinal acceleration G221 is positive among the cases where the determination is positive in step S3 (range from t3 to t10). On the other hand, the case where the negative determination is made in step S4 is the range from t3 to t8 in which the longitudinal acceleration G221 is negative in the case where the positive determination is made in step S3 (range from t3 to t10). If the result of determination in step S4 is affirmative (range from t9 to t10), the process proceeds to step S5. Otherwise (range from t3 to t8), the process proceeds to step S7.

[Step S5]
In step S5, | lateral G | × constant 233 obtained in step S3 is substituted for the acceleration term by the acceleration input term calculating means 98h. That is, in the graph 214 of | acceleration term |, as indicated by reference numeral C2, instead of the values of points P ₄₉ and P ₅₀ that have been input in the acceleration term of the conventional neural network, the point of | lateral G | × constant 233 P ₁₄₉ and P ₁₅₀ are substituted respectively. Thus, by substituting the value of | lateral G | × constant 233, the decrease in the acceleration term is suppressed. After step S5, the process proceeds to step S6.

[Step S6]
In step S6, 0 is substituted for the deceleration term by the deceleration input term calculation means 98i. That is, in the graph 215 of the | deceleration term |, as indicated by a symbol C5, values of 0 are substituted for the points P ₂₅₁ and P ₂₅₂ . After step S6, the process proceeds to step S14.

[Step S7]
In step S7, 0 is substituted into the acceleration term by the acceleration input term calculation means 98h. That, | acceleration term | In the graph 214, as indicated by reference numeral C3, 0 as the value of P ₃₈ from the point P ₃₃ in the range from t3 to t8 is substituted. After step S7, the process proceeds to step S8.

[Step S8]
In step S8, | lateral G | × constant 233 obtained in step S3 is assigned to the deceleration term by the deceleration input term calculation means 98i. That is, in the graph 215 of | deceleration term |, as indicated by reference numeral C4, instead of the values of the points P ₄₃ to P ₄₈ that have been input to the deceleration term of the conventional neural network, | lateral G | × constant 233 Points P ₁₄₃ to P ₁₄₈ are respectively substituted. In this way, by substituting the value of | lateral G | × constant 233, the decrease in the deceleration term is suppressed. After step S8, the process proceeds to step S14.

[Step S9]
In step S9, the control circuit 130 determines whether or not the longitudinal acceleration G221 ≧ 0. The case where the determination in step S9 is affirmative is the time after t11 in which the longitudinal acceleration G221 is positive in the case where the determination is negative in step S3 (range from t0 to t2 and after t11). On the other hand, the case where the negative determination is made in step S9 is the case where the negative acceleration is determined in step S3 (the range from t0 to t2 and after t11), the range from t0 to t2 in which the longitudinal acceleration G221 is negative. It is. As a result of the determination in step S9, if the determination is affirmative (after t11), the process proceeds to step S10. If not (the range from t0 to t2), the process proceeds to step S12.

[Step S10]
In step S10, | front-rear acceleration G | 231 is substituted into the acceleration term by the acceleration input term calculating means 98h. That is, in the graph 214 of | acceleration term |, the points P ₅₁ and P ₅₂ of | the longitudinal acceleration G | 231 are substituted after t11. After step S10, the process proceeds to step S11.

[Step S11]
In step S11, 0 is substituted into the deceleration term by the deceleration input term calculation means 98i. That is, in the graph 215 of | deceleration term |, as indicated by reference numeral C5, 0 is substituted as the values of the points P ₂₅₁ and P ₂₅₂ after t11. After step S11, the process proceeds to step S14.

[Step S12]
In step S12, 0 is substituted into the acceleration term by the acceleration input term calculation means 98h. That is, in the graph 214 of | acceleration term |, 0 is substituted as the value of the points P ₃₀ to P _{32 in} the range from t 0 to t 2 as indicated by reference numeral C 3. After step S12, the process proceeds to step S13.

[Step S13]
In step S13, | longitudinal acceleration G | 231 is substituted into the deceleration term by the deceleration input term calculation means 98i. That is, in the | deceleration term | graph 215, the points P ₄₀ to P ₄₂ of the | front-rear acceleration G | 231 are substituted in the range from t0 to t2. After step S13, the process proceeds to step S14.

[Step S14]
In step S14, the acceleration input term is finally calculated by the acceleration input term calculation means 98h. Based on the results of steps S5, S7, S10, and S12, an acceleration input term is finally obtained. The value of the acceleration input term obtained in step S14 is input to the acceleration input term of the input layer of the neural network NN by the acceleration input term calculation means 98h.

[Step S15]
In step S15, the deceleration input term calculation means 98i finally calculates the deceleration input term. Based on the results of steps S6, S8, S11, and S13, the deceleration input term is finally obtained. The value of the deceleration input term obtained in step S15 is input to the deceleration input term of the input layer of the neural network NN by the deceleration input term calculation means 98i.

  As described above, the acceleration input term and the deceleration input term are input to the acceleration input term and the deceleration input term of the neural network NN from the acceleration input term calculation unit 98h and the deceleration input term calculation unit 98i, respectively, and preprocessing is performed. When a value is input to the input layer of the neural network NN from the other calculating means 98a to 98g of the means 98, the output value of the neural network NN is calculated. The driving direction is estimated based on the output value of the driving direction estimation means (neural network NN) 100, and the shift diagram is switched based on the estimated driving direction. As a result, the automatic transmission 10 is shifted to reflect the driving orientation.

As described above, according to this embodiment, the value to be input to the acceleration term of the neural network NN is a point P ₃₈ of the time t8 from point ₃₀ of the time t0, from one point P ₁₄₉ of the t9 time of t10 time and P _0.99, a value corresponding to the line connecting the point P ₅₂ of the time t12 from the point P ₅₁ of the time t11. On the other hand, the values input to the deceleration term of the neural network NN include the point P ₄₀ at the time t0 to the point P ₄₂ at the time t2, the point P ₁₄₃ at the time t3 to the point P ₁₄₈ at the time t8, and the point at the time t9. A value corresponding to a line connecting P ₂₄₉ to point P ₂₅₂ at time t12.

As described above, in the prior art, is input to the neural network | acceleration term | values, | acceleration term | t8 time from time t0 of the graph 214, as shown from the point P ₃₀ to a point P ₃₈ 0 , and the since time t9, as shown in P ₅₂ from the point P ₄₉ | were the same value as 231 | longitudinal acceleration G. Moreover, inputted into the neural network | deceleration section | values, | deceleration section | t8 time from time t0 of the graph 215, as shown from the point P ₄₀ to a point P ₄₈ | longitudinal acceleration G | 231 and it has the same value, since time t9, was 0 as shown from the point P ₂₄₉ to P _252.

From this, the value input to the deceleration section of the neural network, where the prior art was a value equivalent from the point P ₄₃ to a point P ₄₈ is, in the present embodiment corresponds from point P ₁₄₃ on the point P ₁₄₈ changes to the value, and the value to be input to the acceleration term of the neural network, where the prior art was a value equivalent from the point P ₄₉ to a point P ₅₀ is equivalent to the point P _0.99 from the point P ₁₄₉ in the present embodiment The value changes to Therefore, as indicated by reference numeral C6, the decrease in the driver directivity 216 output from the neural network is suppressed, and the value corresponds to the reference numeral 251 in the prior art, and thus corresponds to the reference numeral 252 in the present embodiment. Rise to value. Until time t8, the value input to the acceleration term is 0 as in the case of the prior art (step S7), but the value input to the deceleration term is larger than that in the case of the prior art. (Step S8), the value of the driver directivity 216 output from the neural network is increased.

  From the above, according to the present embodiment, when cornering is being performed and the accelerator operation amount is small, the output value of the neural network (driver directivity) is independent of the driver's actual driving orientation. The problem of the prior art of being reduced is suppressed, and the decrease in the accuracy of driving orientation is suppressed. That is, according to the present embodiment, the difference between the driver-oriented estimation result and the actual driver-oriented can be reduced.

  Further, as described above, when the accelerator operation change amount is small when the vehicle is turning, the driver directivity output from the neural network is lowered even though the driver's driving orientation is not changed. In order to suppress the point, it is conceivable to newly add a signal indicating characteristics at the time of turning such as lateral G (turning acceleration G) to items to be input to the neural network. As a result, the number of teacher data is large, and as a result, there is a possibility that the number of matching man-hours (the man-hours required for learning) increases and the calculation load during calculation by the neural network increases.

  In the present embodiment, as described above, the lateral G is not newly added as an item to be input to the neural network, but a value corresponding to the lateral G is substituted instead of the longitudinal acceleration G under a predetermined condition. Thus, when learning a neural network, there is no need to newly give a lateral G as teacher data, so that it is possible to collect teacher data and shorten the learning time. In addition, without inputting the lateral G information as a new input term into the neural network, it is possible to avoid an increase in the number of input values for the neural network by inputting from the existing acceleration input term and deceleration input term. Therefore, it is possible to suppress an increase in calculation load without reducing the driving-oriented estimation accuracy.

(Modification of the first embodiment)
In the first embodiment, when the vehicle is turning, unlike in the case of non-turning, the acceleration input layer and the deceleration input layer of the neural network NN are replaced with | Although the value of G | × constant is input (step S5, step S8), in this case, it can be modified as follows.

  That is, as in the first embodiment described above, instead of during non-turning, under predetermined conditions during turning, in step S5 and step S8, instead of | front-back acceleration G | Instead of inputting the value itself (| horizontal G | × value reflecting 100% of the constant value), in this modification, the | horizontal G | × constant value is relatively large (for example, 90%). In addition to the reflection, it is possible to input a value reflecting a relatively small value (for example, 10%) of the | front-rear acceleration G |.

  As described above, at the time of non-turning, the value of the | front / rear acceleration G | itself is reflected in each of the acceleration input layer and the deceleration input layer of the neural network NN (the value reflecting the value of the | front / rear acceleration G | = 100%). Horizontal G | × value reflecting the constant value (0%) is input (step S10, step S13). In this way, during turning and when turning, the ratio of | front / rear acceleration G | and | lateral G | × constant values (influence and reflection) is applied to each of the acceleration input layer and deceleration input layer of the neural network NN. You can enter values with different degrees.

  In the first embodiment, the input layer of the neural network NN has a technically equivalent value for the longitudinal acceleration (a value corresponding to the longitudinal acceleration) instead of the longitudinal acceleration itself. Further, a value that is technically equivalent to the lateral acceleration (a value corresponding to the lateral acceleration) can be input during turning.

  Furthermore, in step S14 and step S15 of FIG. 1, the result obtained by performing statistical processing on the acceleration term and the deceleration term, respectively, may be used as the final acceleration input term and deceleration input term.

(Second Embodiment)
Next, a second embodiment will be described with reference to FIG.
In the second embodiment, the description of the parts common to the first embodiment is omitted.

  The constants (≦ 1) of step S2, step S3, step S5, and step S8 in FIG. 1 in the first embodiment can be changed based on at least one of the road environment and the traveling environment. For example, as shown in FIG. Compared to the road surface gradient when the neural network NN is learned, the constant is smaller as the uphill gradient is larger, and the constant is larger as the downhill gradient is larger. Even if the driving direction has not changed, there is a tendency to step on the accelerator on the uphill road (thus, the vehicle speed tends to increase and the lateral G increases), and on the downhill road, the accelerator tends to return (the vehicle speed decreases to the side). Because G tends to be small), by changing the constant, correction is made to reduce the influence of the road gradient between the terms of the lateral G.

  From the same way of thinking, the constant is set to a small value when the degree of corner bending is large (corner R is small or corner curvature is large) as compared to the condition when the neural network NN is learned. This is because, even if the driving direction has not changed, a large lateral G tends to appear when the degree of corner bending is large.

  Furthermore, the constant can be set larger when the road surface μ is smaller than the condition when the neural network NN is learned. This is because the accelerator tends to return even if the driving orientation has not changed. In addition, the constant is set to be smaller in a winding path such as a continuous curve than the condition when the neural network NN is learned. This is because a large lateral G tends to appear on the winding road.

  Further, the constant is increased when the weather is bad, and the constant is set larger as the surrounding of the vehicle is darker than the condition when the neural network NN is learned. This is because, even in these cases, even if the driving direction has not changed, the accelerator tends to be returned.

  As described above, according to the present embodiment, the steps in FIG. 1 are performed based on at least one of the road environment (road surface gradient, corner bending degree, road surface μ, continuous curve) and the travel environment (weather, brightness). By changing the constants (≦ 1) of S2, Step S3, Step S5, and Step S8, it is possible to suppress a decrease in the accuracy of driving orientation estimation.

  Further, in the above description, the neural network NN is used as a means for estimating the driving orientation. However, the neural network NN is not limited to the neural network. For example, an (artificial) intelligent system such as a genetic algorithm (an optimization method, a soft computer) is used. An information processing mechanism using a singing) can be used. Even when such an information processing mechanism is used, a smaller number of input values used for information processing reduces the calculation load and is advantageous for solution convergence.

It is a flowchart which shows operation | movement of one Embodiment of the driving | operation direction estimation apparatus of this invention. It is a block diagram which shows schematic structure of the control apparatus of the automatic transmission to which one Embodiment of the driving | operation direction estimation apparatus of this invention was applied. It is a block diagram which shows schematic structure of the principal part of one Embodiment of the driving | operation direction estimation apparatus of this invention. It is a time chart which shows operation | movement of one Embodiment of the driving | operation direction estimation apparatus of this invention. It is a figure for demonstrating the map data used with other embodiment of the driving | operation direction estimation apparatus of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Automatic transmission 40 Engine 90 Acceleration sensor 95 Navigation system apparatus 96 Signal reading means 98h Acceleration input term calculation means 98i Deceleration input term calculation means 100 Driving direction estimation means 111 Brake operation amount detection part 112 Road surface μ detection / estimation part 113 Accelerator Opening detection unit 114 Throttle opening sensor 115 Driving direction estimation unit 116 Engine speed sensor 118 Road gradient measurement / estimation unit 122 Vehicle speed sensor 123 Shift position sensor 130 Control circuit 131 CPU
133 ROM
211 Turning judgment 212 | Front / back acceleration G | <| Horizontal G | × Constant graph 213 Graph of longitudinal acceleration G or turning acceleration G 214 | Graph of acceleration term | 215 | Graph of deceleration term | 216 Driver orientation 222 Horizontal G
223 lateral G × constant value 231 | longitudinal acceleration G |
232 | Horizontal G |
233 | lateral G | × constant 251 driver directivity 252 driver directivity NN neural network

Claims (9)

A driving orientation estimation device that estimates driving orientation based on at least one of a value corresponding to a longitudinal acceleration acting on a vehicle and a value corresponding to a lateral acceleration,
Driving direction is estimated when the vehicle is turning and when the vehicle is not turning based on values that are different in the degree of influence of the value corresponding to the longitudinal acceleration and the value corresponding to the lateral acceleration. Estimating device. The driving orientation estimation apparatus according to claim 1,
When the vehicle is not turning, and when the vehicle is turning, the relationship between the value corresponding to the longitudinal acceleration and the value corresponding to the lateral acceleration is a predetermined relationship set in advance. A driving direction estimation device, wherein driving direction is estimated based on values having different degrees of influence of a value corresponding to a longitudinal acceleration and a value corresponding to the lateral acceleration. In the driving orientation estimation device according to claim 2,
When the relationship between the value corresponding to the longitudinal acceleration and the value corresponding to the lateral acceleration is a predetermined relationship, a product of the value corresponding to the lateral acceleration and a predetermined value set in advance The driving orientation estimation apparatus according to claim 1, wherein the driving direction estimation apparatus relates to a magnitude relationship between values corresponding to the acceleration in the longitudinal direction. In the driving orientation estimation apparatus according to claim 2 or 3,
When the vehicle is not turning and when the vehicle is turning, the relationship between the value corresponding to the acceleration in the longitudinal direction and the value corresponding to the lateral acceleration is a predetermined relationship, and the acceleration in the longitudinal direction When the value corresponding to is a value equal to or greater than 0, the driving orientation is determined based on the influence value of the value corresponding to the acceleration in the forward direction and the value corresponding to the lateral acceleration in the front-rear direction different from each other. A driving orientation estimation device characterized by estimating. The driving orientation estimation apparatus according to any one of claims 2 to 4,
When the vehicle is not turning and when the vehicle is turning, the relationship between the value corresponding to the acceleration in the longitudinal direction and the value corresponding to the lateral acceleration is a predetermined relationship, and the acceleration in the longitudinal direction When the value corresponding to is a value less than 0, the direction of driving is determined based on the value of the influence corresponding to the value corresponding to the acceleration in the backward direction and the value corresponding to the lateral acceleration in the front-rear direction. A driving orientation estimation device characterized by estimating. In the driving orientation estimation apparatus according to claim 3,
The driving direction estimation apparatus, wherein the predetermined value set in advance is set to a different value based on at least one of a road environment and a traveling environment. The driving orientation estimation apparatus according to claim 6,
In relation to at least one of the road environment and the traveling environment, the predetermined value set in advance is set to a relatively large value when the driver can easily perform an operation of returning the accelerator. A driving orientation estimation apparatus characterized by that. The driving orientation estimation apparatus according to claim 6,
When a relatively large lateral acceleration is likely to act on the vehicle in relation to at least one of the road environment and the driving environment, the predetermined value set in advance is set to a relatively small value. A driving orientation estimation device characterized by the above. The driving orientation estimation apparatus according to any one of claims 6 to 8,
The driving environment estimation device characterized in that the road environment and the traveling environment include at least one of a road surface gradient, a road surface state, a road bending degree, a curve continuity, weather, and brightness.

Images