JP5665400B2 - Vehicle control device - Google Patents

Description

  The present invention relates to a vehicle control device, and more particularly to a vehicle control device that dynamically changes a vehicle control algorithm in accordance with a vehicle state, a vehicle exterior state, and an occupant state.

  Patent Document 1 below describes a method for efficiently optimizing a decision variable to be optimized whose objective function changes depending on conditions in a short time. Patent Document 2 below describes a method of recognizing a deviation from the driver's normal internal state by learning the relationship between the driving operation of the driver and the external environment as needed.

JP 2009-99051 A JP 2009-73465 A

  By the way, in the technique described in Patent Document 1, a method for efficiently optimizing an optimization symmetric decision variable whose objective variable changes depending on conditions is proposed in a short time, but the objective function is estimated from driving behavior. It is difficult to change autonomously.

  Further, in the technique described in Patent Document 2, a method has been proposed in which the relationship between the driving operation by the driver and the external environment is learned at any time, and the deviation from the usual internal state of the driver is recognized. It is difficult to change the model structure dynamically.

  As described above, in the conventional vehicle control device, when the vehicle encounters a situation that was not assumed at the design stage of the control algorithm, it is difficult to perform vehicle control corresponding to the situation.

  Therefore, the present invention provides a vehicle control device capable of autonomously generating a new control algorithm corresponding to a travel location when the vehicle encounters an unexpected travel location. It is aimed.

  The “running place” of the vehicle includes the situation, state, situation or scene that the vehicle is encountering, including the running state of the vehicle including the stop state, the situation of the outside environment, and the behavior and recognition of the driver of the vehicle ( For example, a sense of crisis) and intention. The “emergence” of the control algorithm includes generating a new control algorithm. In emergence, properties that are not just a simple sum of the properties of parts may appear as a whole. For example, a system that cannot be predicted from the behavior of individual elements may be configured by a complex organization of a plurality of local interactions.

In order to achieve the above object, a vehicle control apparatus according to the present invention is detected by a detection means for detecting at least one of a state of a driver driving a vehicle, an external vehicle environment state, and a vehicle state, and the detection means. Based on the detection state, a first memory that stores an estimation unit that estimates a traveling place that the vehicle encounters using an identity mapping model, and a known traveling place and a vehicle control algorithm that are associated with each other. And the estimated traveling place estimated by the estimating means is a known traveling place stored in the first storage means or an unknown traveling place that is not stored in the first storage means. A new control algorithm corresponding to the estimated travel location using genetic network programming when the determination means determines that the estimated travel location is an unknown travel location. A control algorithm corresponding to the estimated travel location obtained from the first storage means when the estimated travel location is determined to be a known travel location by the emergence means for generating a rhythm and the determination means is a control parameter of the vehicle And converting means for converting a new control algorithm created by the generating means into vehicle control parameters when the estimated driving place is determined to be an unknown driving place. Yes.

According to the present invention configured as described above, when the vehicle encounters an unexpected driving field, a vehicle control capable of autonomously generating a new control algorithm corresponding to the driving field. An apparatus can be provided.
In genetic network programming, a genotype is expressed using a directed graph (network), so that data with a structure such as mathematical formulas and program codes that were difficult to express with conventional genetic algorithms is expressed. Can do. For this reason, genetic network programming is suitable for generating new control algorithms for unknown driving fields.
The identity mapping model is a hierarchical neural network that presents the same data to the input layer and the output layer and learns the intermediate layer output as the identity map. In the identity mapping model, a variety of vague and incomplete information can be input and output, and the reduced information necessary for problem solving can be automatically extracted. In particular, since the identity mapping model does not require a teacher signal, it is suitable for use in estimating an unknown traveling place that has not been assumed in advance.

In the present invention, it is preferable that the control algorithm for the known traveling place stored in the first storage means is created using genetic network programming.
In genetic network programming, genotypes are expressed using directed graphs (networks), so it is possible to express structured data such as mathematical formulas and program codes that were difficult to express with conventional genetic algorithms. . For this reason, genetic network programming is suitable for use in generating appropriate control algorithms for known driving fields.

In the present invention, it is preferable to further include a third storage unit that stores a new control algorithm generated by the generating unit.
By accumulating new control algorithms corresponding to unexpected driving fields, it becomes possible to perform vehicle control corresponding to more diverse driving fields.

  According to the vehicle control device of the present invention, when a vehicle encounters a driving field that has not been assumed, a new control algorithm corresponding to the driving field can be autonomously generated.

It is a block diagram which shows the structure of the control apparatus for vehicles by embodiment of this invention. It is a flowchart which shows the action | operation of the control apparatus for vehicles by embodiment of this invention. It is a flowchart of the algorithm which estimates the traveling place of the own vehicle behavior. It is a graph which shows the result of having estimated the run field of the own vehicle behavior, (A) shows time change of vehicle speed, and (B) shows time change of acceleration. It is a graph which shows the relationship between relative distance and relative speed, and the estimated vehicle-related driving field. It is a flowchart of the algorithm which estimates the driving field of a driver's sense of crisis. It is a graph which shows the result of having estimated the driving field of a driver's crisis feeling, (A) shows time change of a brake depression angle, and (B) shows time change of a brake differential value. It is a flowchart which shows the flow of the basic algorithm of a genetic algorithm (GA). FIG. 3 is a schematic diagram of a group in genetic network programming (GNP) and a directed graph of one individual constituting the group. It is a graph which shows an example of the phenotype of a gene of an individual of genetic network programming (GNP), a genotype, and a node library. It is a schematic diagram which shows the example of the crossover of a genetic network programming (GNP). It is a schematic diagram which shows the example of the mutation of a genetic network programming (GNP). It is a schematic diagram of the vehicle system of a series system hybrid electric vehicle. It is a chart of a parameter of vehicle specifications. It is a graph which shows the relationship between an engine output and a fuel consumption rate. It is a schematic diagram of a battery structure. It is a flowchart which shows the flow of genetic network programming (GNP). It is a flowchart which shows the emergence process of the control algorithm using a correspondence table. It is a schematic diagram of a traveling place. It is a schematic diagram which shows the relationship between the estimated driving field and a corresponding | compatible table. It is a flowchart which shows the emergence process of the control algorithm with respect to an unexpected driving field. It is a schematic diagram of the directed graph structure of genetic network programming (GNP). It is a schematic diagram of the identity mapping model used for estimation of a driving field in 2nd Embodiment. It is a schematic diagram of the intermediate | middle layer output of the identity mapping model in 2nd Embodiment. (A) is a schematic diagram of the assumed range in 2nd Embodiment, (B) is a schematic diagram which shows the relationship between an assumed range and a corresponding | compatible table.

(1. First embodiment)
Hereinafter, an embodiment of a vehicle control device of the present invention will be described with reference to the accompanying drawings.
First, the configuration of the vehicle control device according to the first embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram illustrating a configuration of a vehicle control device according to an embodiment.

  As shown in FIG. 1, the vehicle control apparatus according to the embodiment includes a driver state monitor 11 that detects a state of a driver who drives the vehicle, an external environment monitor 12 that detects an external environment state, and a vehicle state monitor that detects the vehicle state. 13, a detection means 10 including 13, an estimation unit 20 that estimates a traveling place where the vehicle encounters based on a detection state detected by the detection means 10, a known traveling place based on past experience, and the vehicle The first storage unit 30 that stores the control algorithm in association with each other, and the estimated traveling place estimated by the estimating unit 11 is assumed to be assumed from the known traveling place stored in the first storage unit 30. When the determination unit 40 determines whether the estimated traveling place is an unexpected traveling place or not, the determination unit 40 determines whether the traveling place is an unexpected traveling place. A control algorithm corresponding to the estimated driving field acquired from the first storage means when the generating unit 50 for generating a corresponding control algorithm and the determining unit determine that the estimated driving field is an expected driving field Is converted into a vehicle control parameter, and when it is determined that the estimated driving field is an unexpected driving field, a conversion unit 60 is provided for converting the control algorithm generated by the generating means into the vehicle control parameter. doing.

  Furthermore, in the present embodiment, the second storage unit 70 that stores a correspondence table in which a running field that is not based on past experience and a vehicle control algorithm are associated with each other, and a new control algorithm that has been created by the emergence unit 50 are provided. And a third storage unit 80 for storing.

  The estimation unit 20, the determination unit 40, the emergence unit 50, and the conversion unit 60 correspond to, for example, a processing function in an in-vehicle ECU (electric control unit). These processing functions may be realized by executing a predetermined program in a computer, or may be realized by a microchip. The first storage unit 30, the second storage unit 70, and the third storage unit 80 may be configured by a storage device such as a semiconductor memory chip or a hard disk.

Here, with reference to the flowchart of FIG. 2, operation | movement of the control apparatus for vehicles by 1st Embodiment is demonstrated.
(1-1. State detection)
First, the detection means 10 detects the state of the driver who drives the vehicle, the environment state outside the vehicle, and the vehicle state vehicle situation (S21).

  Examples of the driver status monitor 11 that constitutes the detection unit 10 include a monitor camera and an image processing system that are arranged in a vehicle interior and that captures the driver's line of sight and facial movements, and a sensor that detects the amount of depression of the driver's brake Is mentioned. Moreover, as an example of the vehicle exterior environment monitor 12, the radar directed to the front of the vehicle for detecting the distance to the obstacle including the preceding vehicle and the relative speed, the radar directed to the side or the rear of the vehicle, or directed to the outside of the vehicle Examples include monitor cameras and image processing systems. Examples of the vehicle state monitor 13 include a vehicle speed sensor, an acceleration sensor, a brake depression amount sensor, an accelerator depression amount sensor, and a steering angle sensor.

  The detection unit 10 does not always need to be provided with the driver status monitor 11, the vehicle exterior environment monitor 12, and the vehicle status monitor 13, and may be provided with only one or two of them. Good.

(1-2. Estimation of driving field)
Next, the estimation unit 20 estimates the traveling place where the vehicle encounters based on the detection state detected by the detection means 10 (S22).

  Hereinafter, an example in which the behavior of the host vehicle is estimated based on the vehicle status, the vehicle relationship between the preceding vehicle and the host vehicle is estimated based on the environment outside the vehicle, and the driver's sense of crisis is estimated based on the driver state. Will be described.

(1-2-1. Estimating the driving field of own vehicle behavior)
Generally, a vehicle starts from a stopped state and travels. Then, acceleration and deceleration are repeated to finally stop. Thus, the behavior of the vehicle repeats the cycle of “stop”, “start”, and “travel”. For this reason, as shown in Table 1 below, three driving locations of “start”, “travel”, and “stop” are estimated as the travel locations for the behavior of the host vehicle.

  In estimating the traveling place of the own vehicle behavior, paying attention to the vehicle speed V and the acceleration α which is a time derivative thereof, the own vehicle starts to move from the state of “stop” at the vehicle speed V = 0, and the acceleration α> 0. And move to "start". Further, after “starting”, the vehicle speed reaches the maximum value, and the acceleration α <0, and the routine proceeds to “running”. Then, the vehicle decelerates from “travel”, and the vehicle speed V = 0 and shifts to “stop”. For this reason, the vehicle speed V and the acceleration α can be used as parameters for estimating the traveling place of the vehicle behavior.

  FIG. 3 shows a flowchart of an algorithm for estimating the traveling place of the own vehicle behavior from the vehicle speed V and the acceleration α. Until the input vehicle speed V> 0 (in the case of “no” in S32), the traveling place is estimated to be “stop” (S31). Next, when the vehicle speed V> 0 (in the case of “yes” in S32), the driving field is estimated to be “start” until the input acceleration α <0 (in the case of “no” in S34) (S33). ). Next, when acceleration <0 (in the case of “yes” in S34), the traveling place is estimated to be “traveling” until the input vehicle speed V = 0 (in the case of “no” in S36) (S35). . Then, when the vehicle speed V = 0 (in the case of “yes” in S36), the flowchart returns to the start, and the traveling place is estimated to be “stop”.

  FIG. 4 shows a result of estimating the traveling place of the own vehicle behavior by applying the estimation algorithm shown in FIG. 3 to the traveling pattern of the vehicle represented by the vehicle speed V and the acceleration α. A line I in the graph of FIG. 4A indicates a time change of the vehicle speed V, and a line II in the graph of FIG. 4B indicates a time change of the acceleration α.

  In the example shown in FIG. 4, since the vehicle speed V = 0 until time t0, the traveling place is estimated to be “stop”. Then, from the vehicle speed V> 0 at time t0 to the acceleration α <0 at time t1, the running field is estimated to be “start”. Further, the traveling place is estimated to be “traveling” from the time when acceleration α <0 at time t1 until the vehicle speed V = 0 at time t2. Then, after time t2, the vehicle speed V = 0, and the traveling place is estimated to be “stopped”. In this way, it is possible to estimate the traveling place of the vehicle behavior based on the vehicle speed V and the acceleration α.

(1-2-2. Estimation of the driving field of the inter-vehicle system)
As shown in Table 2 below, there are six areas of “following”, “free”, “shrinking”, “expanding”, “approaching”, and “collision”. Estimate the driving field.

  Here, the inter-vehicle distance L between the host vehicle and the preceding vehicle and the relative speed Vr of the preceding vehicle with respect to the host vehicle are used as parameters for estimating a vehicle-related traveling place. Specifically, as shown in FIG. 5, the inter-vehicle distance L is within a predetermined range (L1 ≦ L ≦ L2), and the relative speed Vr is within a predetermined range with Vr = 0 therebetween (−V1 ≦ Vr ≦ + V1), it is estimated that the driving field is “following”. Further, when the inter-vehicle distance L is longer than the upper limit inter-vehicle distance of “follow-up” (L2 <L), the traveling place is estimated as “free”. The inter-vehicle distance L is within a predetermined range of “following” (L1 ≦ L ≦ L2), but when the relative speed Vr is less than the lower limit value of “following” (Vr <−V1), the travel field is “shrinked”. ". The inter-vehicle distance L is within a predetermined range of “following” (L1 ≦ L ≦ L2), but when the relative speed Vr exceeds the upper limit value of “following” (+ V1 <Vr), the driving field is “expanded”. presume. Further, when the inter-vehicle distance L is less than the lower limit value of “following” and 0 or more (0 ≦ L <L1), the traveling place is estimated as “approaching”. When the inter-vehicle distance L is less than 0 (L <0), it is estimated as “collision”.

  FIG. 5 shows the relationship between the relative distance L and the relative speed Vr, and the estimated vehicle-related traveling place. The horizontal axis of the graph in FIG. 5 represents the inter-vehicle distance L between the preceding vehicle and the host vehicle, and the vertical axis represents the relative speed Vr of the preceding vehicle with respect to the speed of the host vehicle. As shown in FIG. 5, a vehicle-related traveling place is estimated from the relative distance L and the relative speed Vr.

(1-2-3. Estimating the driving field with a sense of driver crisis)
As shown in Table 3 below, three driving locations of “normal”, “precursor”, and “near hat” are estimated as driving locations for the driver's sense of crisis. Generally, a driver applies a brake suddenly when encountering an emergency such as jumping out or sudden interruption. Therefore, the brake depression angle (brake depression amount) B and the brake differential angle time differential value (hereinafter also referred to as “brake differential value”) β are used as parameters for estimating the driving field for the driver's sense of crisis. To do.

  FIG. 6 shows a flowchart of an algorithm for estimating a driver's critical driving field from the brake depression angle B and the brake differential value β. Until the input brake differential β exceeds a predetermined threshold βT and β> βT is satisfied (in the case of “no” in S62), the traveling field is estimated to be “normal” (S61). Next, when the brake differential value β> βT is satisfied (in the case of “yes” in S62), the input brake depression angle B exceeds the predetermined threshold value BT and becomes B> BT (in the case of “no” in S654). The traveling place is estimated to be “a precursor” (S63).

  Note that when the brake depression angle B becomes 0 (in the case of “yes” in S64), the flowchart returns to the start, and the traveling place is estimated to be “stop”. That is, not all “precursors” will shift to “near-miss”. This is because even if the brake depression amount B is small, only the time differential value β of the brake depression amount may be a large value.

  Next, when the brake depression angle B> BT is satisfied (in the case of “yes” in S65), the traveling place is “near-miss”, that is, the driver until the input brake depression angle B <BT (in the case of “yes” in S67). Is estimated to be in a state of crisis. Next, when the input brake depression angle B <BT is satisfied (in the case of “no” in S67), the traveling place is estimated again as “precursor” (S63).

  FIG. 7 shows the result of estimating the driving field by applying the driver's criticality estimation algorithm shown in FIG. 6 to the driving pattern represented by the brake depression angle B and the brake differential value β. A line I in the graph of FIG. 7A shows a time change of the brake depression angle B, and a line II in the graph of FIG. 7B shows a time change of the brake differential value β.

  In the example shown in FIG. 7, since the brake differential value β <βT until time t0, the traveling place is estimated to be “normal”. The traveling place is estimated to be a “precursor” from the time when the brake differential value β> βT at time t0 to the time when the brake depression angle B> BT at time t1. Further, the traveling place is estimated to be “near hat” until the brake depression angle B> BT at time t1 and the brake depression angle B <BT at time t2. Further, the traveling place is estimated to be “precursor” until the brake depression angle B <BT at time t2 and the brake depression angle B = 0 at time t3. The traveling field is estimated to be “normal” from when the brake depression angle B = 0 at time t3 until the brake differential value β> βT again at time t4.

  Further, the traveling place is estimated to be a “precursor” until the brake differential value β> β again at time t4 until the brake depression amount B = 0 at time t5. At time t5, the brake depression amount B becomes 0, and the traveling place is estimated to be “normal”. In this way, it is possible to estimate the driving field of the driver's sense of crisis based on the brake depression angle B and the brake differential value β.

  In addition to the above-described driving field regarding the vehicle behavior, vehicle relations, and driver's sense of crisis, a driving field related to the road shape may be estimated. In general, the road shape is represented by the curvature and slope of the road. For this reason, as the driving field related to the road shape, three driving fields of “straight line”, “right curve” and “left curve” are estimated for the curvature, and “flat”, “uphill” and “downhill” for the gradient. Estimate three driving fields.

  The driving field related to the road shape may be estimated by comparing the current position of the host vehicle obtained by using a global positioning system (GPS) with a map database of the car navigation system. Further, the road-shaped traveling place may be estimated as, for example, a gradient in a predetermined traveling section, a combination of gradients, or a combination of gradient and curvature. Further, for example, the traveling place may be estimated by combining any two or more of the own vehicle behavior, the vehicle relation, the driver's sense of crisis, and the hitting of the running place with respect to the road shape.

(1-3. Judgment of running place)
Next, the determination unit 40 determines whether the estimated traveling place estimated by the estimating unit 11 is an expected traveling place that is assumed from the known traveling places stored in the first storage unit 30 or an unexpected traveling. It is determined whether it is a place (S23 in FIG. 2).

(1-3-1. Control algorithm)
The first storage unit 30 stores a known driving field based on past experience and a vehicle control algorithm in association with each other. In the present embodiment, the control algorithm for the known driving field stored in the first storage unit 30 is created using genetic network programming (GNP).

  Hereinafter, first, genetic network programming will be described, and then an example of a control algorithm created by the genetic network programming will be described.

  Genetic network programming is an evolutionary calculation method that seeks an optimal solution or a sub-optimal solution by imitating the evolution process of a living organism, and is a genetic programming (GP) that extends a genetic algorithm (GA). : Genetic Programming).

(1-3-1-1. Genetic Algorithm (GA))
A genetic algorithm (GA) is a computational technique that mimics the selection, crossover and mutation of chromosomes found in the evolutionary process of organisms and is applied engineeringly. In the genetic algorithm, a design variable is regarded as a gene, and the gene is expressed by a bit string structure.

Here, the basic algorithm flow of the genetic algorithm will be described with reference to the flowchart of FIG.
First, an initial group is created (S81). The initial population is a set of individuals, and each individual is a solution candidate symbolized, each having one gene.

Next, an evaluation value for each individual is calculated (S82).
The evaluation value refers to an index that quantitatively indicates the effectiveness of each individual. For example, good fuel efficiency can be used as the evaluation value.

Next, an individual is selected based on the evaluation value (S83).
Individual selection refers to selecting individuals to be left in the next generation.

Next, the selected individual is crossed and mutated (S84).
Crossing means exchanging the genes of the parent generation and creating a new gene as a child generation. Mutation means that a part of a gene is randomly rewritten to a parent generation to become a child generation.

Next, the final generation is determined (S85).
The above steps S82 to S84 are repeated until the final generation.
In this way, an optimal solution or a sub-optimal solution is searched.

(1-3-1-2. Genetic programming (GP))
The flow of the algorithm of genetic programming (GP) is basically the same as that shown in FIG. 8, but in genetic programming (GP), a gene is expressed by a tree structure using a node function. Then, the node function is used to determine the condition based on an if-then statement, and the process is executed. Thereby, in genetic programming (GP), it is possible to handle a structural system that cannot be handled by the bit string structure of the genetic algorithm (GA).

(1-3-1-3. Genetic Network Programming (GNP))
The flow of genetic network programming (GNP) algorithm is basically the same as that shown in FIG. 8, but in genetic network programming (GNP), as shown in FIG. Have a directed graph (network) structure in which node functions are connected to each other by directed edges. The node function is composed of two types of nodes, a determination node and a processing node. The decision node is responsible for condition judgment and branching by an if-then sentence. On the other hand, the processing node has a role of executing a predetermined process (action). Further, the directional branch extends from the determination node by the number of branches. On the other hand, only one directional branch extends from the processing node.
In Genetic Network Programming (GNP), since the number of nodes is constant in any individual, bloat (solution expansion) does not occur.

  Further, FIG. 10 shows an example of an individual gene of genetic network programming (GNP). The gene is expressed in a network form like the phenotype shown in FIG. 10, and is described as the genotype shown in FIG. 10 on the program. The genotype “NIDi” is the ID of the node i, and there are as many as the set number of nodes. The genotype “NTi” represents a node function, and “NTi = 1” represents a determination node, and “NTi = 2” represents a processing node. The genotype “IDi” represents a node function ID, and indicates a label of a note library that describes the contents of a determination node and a processing node. Of the node contents described in the note library, the node contents to be executed are determined by “NTi” and “IDi”. The genotypes “Ci1” and “Ci2” are connection destination information of a directional branch extending from the node i, and a node ID to be connected is described. Since the processing node has only one connection destination, Ci2 does not exist in the processing node.

  Such an evaluation value of each individual is obtained by quantifying the achievement level and effectiveness of the individual, and the value is obtained by evaluation conversion. The calculation formula of the evaluation function is set by the designer depending on the problem. Since genetic manipulation is performed based on the evaluation value of the individual, optimization of genetic network programming (GNP) corresponds to optimization of the evaluation value.

  Then, before performing crossover or mutation, an individual with a high evaluation value is preferentially selected by selection. Thereby, the gene of an individual with a high evaluation value is inherited to the next generation. Examples of selection methods include roulette selection, tournament selection, and elite preservation selection.

In roulette selection, individuals to be subjected to crossover and mutation operations are selected, and therefore the offspring of each individual is selected with a probability proportional to the evaluation value. When the evaluation value of the i-th individual is fi, the probability pi that the i-th individual is selected is obtained by the following equation (1).
pi = fi / Σf (1)

  In the tournament selection, in order to select individuals to be subjected to crossover and mutation operations, individuals are randomly selected from the population of individuals determined in advance by the designer, and the individual with the highest fitness is selected. This is repeated until the required number of individuals is obtained.

  In the elite preservation selection, an operation is performed to leave a certain number of individuals with high evaluation values in the generation as they are to the next generation. Thereby, it is possible to prevent the evaluation of the optimum individual from deteriorating.

  Crossover is performed between the two selected individuals to generate two child individuals. At this time, all the gene information between the intersecting nodes is exchanged. Here, the crossover procedure will be described with reference to FIG.

  First, two parent individuals are arbitrarily selected from a group of parent individuals. In FIG. 11, “parent 1” and “parent 2” are selected. Next, in one “parent 1” of the selected parent individuals, a node to be crossed is selected based on the set crossing probability. In FIG. 11, the node “P1” of “parent 1” is selected. Next, all the genes are transferred between the node “P1” selected in one parent individual “parent 1” and the node “P2” having the same identification number as “P1” in the other parent individual “parent 2”. Exchange. The two individuals “child 1” and “child 2” generated in this way become the next generation individuals.

Mutation is performed within one parent individual to generate one child individual. There are two types of mutations: one that changes the connection destination node and one that changes the node contents. Here, the mutation will be described with reference to FIG.
First, in the mutation for changing the connection destination node (mutation A), one individual is arbitrarily selected, and the node to be mutated is selected based on the set mutation probability. Then, the connection destination of the selected node is randomly changed. In FIG. 12, in the mutation A, the connection destination node of the node P1 is switched from the upper right processing node to the upper left determination node.

Further, in the mutation (mutation B) for changing the node contents, one individual is arbitrarily selected, and the node to be mutated is selected based on the set mutation probability. Then, the label of the selected node is randomly changed. In FIG. 12, in the mutation B, the selected node P1 is changed to the node P2.
In this way, a next generation individual is generated.

(1-3-1-4. Usage example of genetic network programming (GNP))
An example of generating a vehicle control algorithm related to the sharing of an engine and a motor in a hybrid electric vehicle by using gene network programming (GNP) will be described below. Here, a vehicle control algorithm that improves fuel efficiency in accordance with the driving field is created.

First, FIG. 13 shows a vehicle system of a series-type hybrid electric vehicle, which is a premise of the control algorithm. Moreover, the parameter of a vehicle specification is shown in FIG.
Further, the gravitational acceleration g is set to 9.8 m / s ² and the air density ρ = 1.29 kg / m ³ is set.

The motor output W _M (W) is obtained by the following equation (2).
W _M = T _M · (V / r _D ) i (2)
Here, T _M [Nm] represents motor torque, and V [m / s] represents vehicle speed. From the relationship between the maximum motor output W _Mmax and the motor torque T _M , the maximum torque is 400 Nm up to the vehicle speed V = 18.75 m / s, and the maximum torque decreases in inverse proportion to the speed above this. . The motor torque is obtained from the product of the maximum torque and the normalized accelerator input amount.

As shown in FIG. 13, in a series-type hybrid electric vehicle, engine power is converted into electric power by a generator. The power generation amount W _G [W] is obtained from the following equation (3).
W _G [W] = η _G W _E (3)
However, W _E [W] is an engine output. Further, the fuel consumption rate F _E [g / kWh] is obtained from the graph of FIG. The graph of FIG. 15 shows the relationship between engine output and fuel consumption rate. Further, the fuel consumption amount F _C [g / s] per second is obtained by the following equation (4).
F _C = W _E · F _E / 1000/3600 (4)

The battery configuration is shown in FIG. Further, the battery internal voltage V _B = 400V is set, and the battery internal resistance R _B = 400 mV is set.

The remaining battery capacity E _B [Wh] is obtained from the following equation (5).
dE _B / dt = W _B − (W _B / V _B ) ² · R _B
W _B = W _G −W _M −W _A (5)
However, W _B is a battery input and output.

The equivalent fuel consumption F _B (g) is obtained from the following equation (6).
F _B = ΔSOC · E _Bmax F _E / η _M / 1000 (6)
The equivalent fuel consumption amount is a fuel consumption amount required for operating the engine after the travel is completed and returning the remaining battery charge SOC to the written remaining charge. In this case, the engine is operated at the maximum efficiency point F _E = 250 g / kWh.
However, ΔSOC (%) is the difference between the remaining battery charge SOC at the end of travel and the remaining battery charge SOC at the start of travel.

The brake includes a regenerative brake, a hydraulic brake, and an engine brake. The regenerative brake is used when the required brake torque requested by the driver is _not more than the maximum motor torque T _Mmax , that is, not more than 400 Nm. The magnitude of the regenerative energy obtained by the regenerative brake can be obtained from the above equation (2). The product of the regenerative energy and the power generation efficiency η _M of the motor is the power generation amount of the motor. At this time, the motor output W _M in the above equation (5) is negative.

The hydraulic brake is used when the required brake torque exceeds the torque obtained by the regenerative brake, that is, when it is 400 Nm or more, and compensates for the excess of the maximum motor torque T _Mmax . The required brake torque is obtained from the product of the maximum brake torque and the normalized brake input amount.

  The engine brake is used when neither the brake nor the accelerator is input. In a series-type hybrid electric vehicle, the drive shaft and the engine are not directly connected, and therefore there is essentially no engine brake. However, regenerative braking torque of 35 Nm is applied to reproduce it, and charging is performed.

Further, the behavior of the vehicle body is obtained from the following equation (7).
m (dV / dt) = { (T M iη t -T B) / r D} -R ··· (7)
R = R _i + R _r + R _l
R _i = mg sin θ
R _r = μ _r mg
R _l = (ρ / 2) C _D AV ²

However, m [kg] represents the weight of the vehicle body, T _M represents the motor torque [Nm], i represents the reduction ratio, η _t represents the transmission efficiency, and T _B represents the brake torque [Nm]. R (N) represents running resistance, Ri [N] represents gradient resistance, Rr [N] represents rolling resistance, Rl [N] represents air resistance, and θ [rad] represents road gradient. Further, mu _r is the rolling friction coefficient, C _D is the drag coefficient, A represents the frontal projected area of the vehicle.

  Genetic network programming (GNP) setting parameters are shown in Table 4 below, and determination node and processing node settings are shown in Table 5 below. The determination node and the processing node cut a range of input values at arbitrary intervals and generate a plurality of nodes having different threshold values. The determination and processing bonds are output in two directions “yes” or “No” depending on whether or not the input value is within the range of the threshold value.

Next, the flow of genetic network programming (GNP) under the above-described assumption will be described with reference to the flowchart of FIG.
First, as a random initial group, a network in which determination nodes and processing nodes are randomly combined is created by the number of individuals (S171).

Next, the accelerator depression angle and gradient for each traveling place are input (S172). The accelerator depression amount and the gradient are, for example, input data when the vehicle-related driving field is “free” and the road gradient driving field is “flat”, “uphill” or “downhill”. It is good to use. The slope when the road slope is “uphill” or “downhill” is 4.0%.
Next, the required motor output and brake torque are calculated from the input of the accelerator depression angle (S173).

  Next, a condition determination for maintaining the remaining battery charge SOC within a predetermined range (40 to 70%) is performed (S174). When the remaining battery SOC exceeds 70% (in the case of “yes” in S174), the engine is stopped (S175). Further, when the remaining battery charge SOC is less than 40% (in the case of “yes” in S176), the engine is operated (S177). When operating the engine, if the required engine output is 35 kW or less, the engine output is 35 kW, and if it is 35 kW or more, the requested value is output (S177).

  When the remaining battery SOC is within a predetermined range (40 to 70%) (“no” in S174 and “no” in S176), the engine output is determined by a control algorithm based on genetic network programming (GNP). (S178).

  Subsequently, the power generation amount is calculated, and the fuel consumption amount is calculated (S179). Further, the remaining battery charge SOC is calculated (S180), and the equation of motion is solved (S181). And the loop of said step S172-S181 is repeated until the driving | running | working mode in a predetermined driving field is complete | finished (it is "no" in S182).

  After the travel mode ends (in the case of “yes” in S182), an evaluation value of the network (one individual) is calculated (S183). The evaluation value is expressed as the sum of the fuel consumption and the equivalent fuel consumption. The loop of steps S172 to S183 is repeated for the number of individuals.

  Then, after the evaluation values of all the individuals are obtained (in the case of “yes” in S184), it is determined whether or not it is the last generation (S185). If it is not the final generation (in the case of “no” in S185), the individual population is evolved by applying genetic operations such as selection, crossover, and mutation according to the probability of the setting parameter shown in Table 4 (S186). If it is the final generation (in the case of “yes” in S185), the node transition network of the optimum individual is stored (S187), and the process ends.

  A control algorithm based on such a node transition network of the optimum individual is stored in the first storage unit 30 in association with the traveling place.

(1-4. When the estimated field is within the expected range)
Next, when it is determined by the determination unit 40 that the estimated traveling place is an expected traveling place assumed from the known traveling places stored in the first storage unit 30 ("yes" in S23 of FIG. 2). In this case, the conversion unit 60 reads out a control algorithm corresponding to the estimated traveling place from the first storage means (S24), and converts it into a vehicle control parameter (S25).

  For example, in the series-type hybrid electric vehicle of the system shown in FIG. 13, the vehicle-related driving field is “free” and the road gradient driving field is “flat”. In some cases, a control algorithm corresponding to the traveling place is read from the first storage unit 30. Then, according to the read control algorithm, the shared output between the engine and the motor is calculated as the vehicle control parameter.

  In addition to the engine and motor, actuators such as handles and brakes may be controlled according to the control parameters. For example, the steering wheel may be automatically operated according to the steering angle converted from the control algorithm, or the brake may be automatically operated according to the vehicle speed converted from the control algorithm.

  Further, according to the control parameter, a human machine interface (HMI) such as a speaker or a display may be controlled. For example, the driver may be guided by voice so that the vehicle travels according to the steering angle and the vehicle speed converted from the control algorithm.

(1-5. When the estimated field is unexpected)
On the other hand, when it is determined by the determination unit 40 that the estimated traveling place is an unexpected traveling place that is not an assumed traveling place assumed from the known traveling places stored in the first storage unit 30 (FIG. 2 is “no” in S23), the emergence unit 50 creates a control algorithm corresponding to the estimated traveling place (S26).

  With reference to the flowchart of FIG. 18, the emergence process of the control algorithm in the case where the estimated traveling place is unexpected is described. The flowchart in FIG. 18 corresponds to steps S23, S26, and S27 of the flowchart in FIG. Here, the case where the accident vehicle B is stopped as an obstacle in front of the traveling direction of the host vehicle C as shown in FIG.

First, it is determined whether or not the traveling place estimated by the estimating unit 11 is a traveling place within an assumed range (S23 in FIG. 18).
Here, as shown in FIG. 20, the traveling place is based on “width” left on the side of the accident vehicle and “deceleration” necessary for the own vehicle to stop before the accident vehicle. Presumed. The “width” is obtained from the front image captured by the external environment monitor 12. The “deceleration” is calculated from the distance from the current position of the host vehicle to the accident vehicle obtained from the front image captured by the outside environment monitor 12 and the current vehicle speed measured by the vehicle state monitor 10. Desired.

  On the other hand, the known driving field stored in the first storage unit 30 is also specified by two parameters of “width” and “deceleration”. FIG. 20 schematically shows some of the known running places with white circles. A white circle A in FIG. 20 represents a traveling place where “width” is sufficient and the “deceleration” is small because the distance to the accident vehicle is long. A control algorithm for largely avoiding an accident vehicle by a steering wheel operation is stored for such a driving field. Also, a white circle B in FIG. 20 represents a traveling place where “width” is sufficient and the “deceleration” is large because the distance to the accident vehicle is short. A control algorithm for largely avoiding an accident vehicle caused by a steering wheel operation is stored even for such a driving field. A white circle C in FIG. 20 indicates a traveling place where “width” is not enough and “deceleration” is between the white circle A and the white circle B. A control algorithm for applying a brake is stored for such a traveling place because the vehicle cannot travel beside the accident vehicle. In this case, it is also preferable to store a control algorithm that instructs the driver by voice to hold the steering wheel in preparation for a collision with the accident vehicle.

  In addition, a curve I surrounding all the white circles in FIG. 20 represents an expected range of the driving field. The range of the expected driving ranges includes not only the known driving points themselves indicated by the white circles but also the driving points between the known driving points, that is, the driving points that can be interpolated between the known driving points. included. Therefore, when the estimated traveling place can be interpolated between known traveling places stored in the first storage unit 30, it is determined as an assumed traveling place, and extrapolation from the known traveling place is necessary. In the case, it is determined as an unexpected traveling place.

  Further, in FIG. 20, the traveling place estimated by the estimation unit 11 is indicated by a black circle D. A black circle D represents a traveling place where the “width” is between the white circle A and the white circle C and the “deceleration” is large. As shown in FIG. 20, the estimated traveling place is not included in the assumed range surrounded by the curve I. For this reason, it is determined that the estimated traveling place indicated by the black circle D is an unexpected traveling place (S23 in FIG. 18).

When it is determined that the estimated traveling place is unexpected, the correspondence table is read from the second storage unit 70 (S261). The correspondence table stores a driving field that is not based on past experience and a vehicle control algorithm in association with each other.
Here, an example of the correspondence table is shown in Table 6 below.

Next, from the correspondence table, an approximate traveling place that most closely approximates the estimated traveling place is selected, and a control algorithm corresponding to the approximate traveling place is read out (S262).
Here, the assumption shown in Table 6 above is that the “front collision prevention distance” is “none”, the “rear collision prevention distance” is “present”, and the “side space” is “ The “existing” case is set as a construction extrapolation point (S262).

Next, an approximate traveling place that is most approximate to the estimated traveling place indicated by the black circle D is selected from the known traveling places stored in the first storage means, and a control algorithm corresponding to this approximate traveling place is read ( S263).
Here, the traveling place indicated by the white circle E in FIG. 20 is selected as the approximate traveling place.

  Then, an intermediate driving field between the most approximate driving field of the white circle E selected from the assumed driving field and the closest driving field selected from the correspondence table is selected as the estimation field (S264), and the driving of the white circle E is performed. A control algorithm intermediate between the field control algorithm and the driving field control algorithm selected from the correspondence table is generated (S265).

Here, the emergence of the control algorithm will be described with reference to FIG.
First, when there is a margin in the width (in the case of “Yes” in S211), for example, in the case of the white circle B that is assumed in FIG. 20, a control algorithm for avoiding an obstacle by a steering wheel operation is read (S212). . Further, the driver is instructed to drive slowly (S213).

  Further, when there is no margin in the width (in the case of “None” in S211), for example, in the case of the driving circle of the white circle C that is assumed in FIG. 20, the damage caused by the collision of the host vehicle with the obstacle is reduced. Therefore, a control algorithm for operating the brake is read (S214), and the engine is further stopped (S215).

  Then, when the margin of the width is intermediate between “present” and “absent” (in the case of “intermediate” in S211), for example, in the case of an unexpected black circle D in FIG. 20, for example, genetic A new control algorithm is created using dynamic network programming (S216).

  Here, FIG. 22 shows an example of the directed graph (network) structure of the individuals in the initial group when a new control algorithm is created. The individuals of the initial group may be those of the driving field of the white circle E and those of the driving field selected from the correspondence table. Here, the control algorithm is evaluated based on the degree of damage for the purpose of minimizing the degree of damage that occurs when the host vehicle collides with an obstacle. Then, genetic operations such as crossover and mutation are repeated for the directed graph structure of the driving field of the white circle E and the directed graph structure of the driving field selected from the correspondence table to generate a new control algorithm.

  Then, for example, a control algorithm is generated that stops the host vehicle while rubbing the side of the accident vehicle that is an obstacle.

  Then, it learns the quality of the generated control algorithm as a response of the driver (S266), and the generated control algorithm is stored in the third storage unit 80. In this way, the place where the vehicle encounters is automatically estimated, and when the vehicle encounters a place that was not assumed at the design stage of the control algorithm of the vehicle control device, it corresponds to that place A new control algorithm is automatically generated. In addition to the existing control algorithms, by accumulating new control algorithms corresponding to unexpected driving fields, it becomes possible to perform vehicle control corresponding to more diverse driving fields.

(2. Second Embodiment)
Hereinafter, a second embodiment of the present invention will be described.
The configuration and basic operation of the vehicle control device according to the second embodiment are the same as those of the first embodiment described above.
However, in 2nd Embodiment, the estimation part 20 estimates the driving field which the vehicle has encountered using an identity mapping model.

  FIG. 23 shows a schematic diagram of the identity mapping model. The identity mapping model is a hierarchical neural network that presents the same data to an input layer and an output layer and learns an intermediate layer output as an identity map. In the example shown in FIG. 23, the inter-vehicle distance between the preceding vehicle and the host vehicle is input to the input layer and the output layer as external environment data, and the vehicle speed, the brake depression amount, the accelerator depression amount, and the steering are input as vehicle state data. The corner is entered.

  In the example shown in FIG. 23, as an intermediate layer, car behavior (stop, start, running), vehicle relations (following, free, shrink, spread, approach, collision), driver reaction (usually, precursor, near-miss), and road gradient (Flat, Uphill, Downhill) is output as a driving field.

  FIG. 24 shows a schematic diagram of the intermediate layer output of the identity mapping model. The horizontal axis x1 and the vertical axis x2 in FIG. 24 are output parameters of the intermediate layer, respectively. Then, as shown in FIG. 24, a traveling place (for example, overtaking, passing, parallel parking, parking, right turn, crossing an intersection, left turn) is specified by the intermediate layer output.

  Further, in the second embodiment, the traveling place stored in the first storage unit 30 is specified by the intermediate layer output. 25A and 25B are schematic diagrams of the assumed range. Lines II in FIGS. 25A and 25B schematically show the assumed range of the assumed field stored in the first storage unit 30. The horizontal and vertical axes in FIGS. 25A and 25B are parameters of the intermediate layer output. And in FIG. 25 (A), the estimated traveling place e1 shown with a broken line is included in the assumption range enclosed with the line II. Therefore, it is determined that the estimated traveling place e1 is an expected traveling place.

  On the other hand, in FIG. 25B, the estimated traveling place e2 indicated by a broken line is not included in the assumed range surrounded by the line II. Therefore, it is determined that the estimated traveling place e2 is an unexpected traveling place. In addition, FIG. 25B schematically shows a range T of the driving field that is not based on the past experience, in which the assumptions for the driving field within the assumption are listed in the correspondence table.

  In the example shown in FIG. 25 (B), the driving field t1 is selected as the driving field in the correspondence table that most closely approximates the estimated driving field e2 that is unexpected and specified by the intermediate layer output of the identity mapping model, The driving field b is selected as the driving field within the assumption that is closest to the estimated driving field e2. In the same manner as in the first embodiment described above, a new control algorithm corresponding to an unexpected traveling place is created.

  In the above-mentioned embodiment, although the example which comprised this invention on the specific conditions was demonstrated, this invention can perform a various change and combination, and is not limited to this.

DESCRIPTION OF SYMBOLS 10 Detection part 11 Driver state monitor 12 Outside vehicle environment monitor 13 Vehicle state monitor 20 Estimation part 30 1st memory | storage part 40 Judgment part 50 Emergence part 60 Conversion part 100 ECU

Claims (3)

Detecting means for detecting at least one of a state of a driver driving the vehicle, an external vehicle environment state, and a vehicle state;
Based on the detection state detected by the detection means, an estimation means for estimating a running place that the vehicle encounters using an identity mapping model ;
First storage means for storing a known driving field and a vehicle control algorithm in association with each other;
Determining whether the estimated traveling place estimated by the estimating means is a known traveling place stored in the first storage means or an unknown traveling place not stored in the first storage means Means,
Emerging means for generating a new control algorithm corresponding to the estimated travel location using genetic network programming when the estimated travel location is determined to be an unknown travel location by the determination means;
When the determining means determines that the estimated traveling place is a known traveling place, the control algorithm corresponding to the estimated traveling place obtained from the first storage means is converted into a vehicle control parameter, and the estimated traveling place is A conversion means for converting a new control algorithm created by the emergence means into a vehicle control parameter when it is determined that the vehicle is an unknown traveling place;
A vehicle control device characterized by comprising: The stored in the first storage means, the control algorithm for the known travel field, the vehicle control device according to claim 1, wherein a is obtained by emergent using genetic network programming. Further, the vehicle control device according to claim 1, wherein further comprising a third storage means for storing a new control algorithms emergent by the emergent means.

Images