JP2014238035A - Vehicle control device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress the deterioration of the learning accuracy of a learning value at a lattice point (end point) at an end of a learning map, related to a vehicle control device.SOLUTION: A vehicle control device comprises a learning map which is formed into a lattice shape, and in which a learning value of a control parameter used in the control of an internal combustion engine 10 is associated with each lattice point. Weighting corresponding to a distance between a learning objective lattice point and learning data is performed to each of a plurality of pieces of learning data of the control parameter included in a prescribed learning range from a position of a prescribed learning objective lattice point being a learning object, and a learning value of the learning objective lattice point is calculated by using the learning data which are applied with the weighting. The learning value of the learning objective lattice point is updated by using the calculated learning value. The learning value which is associated with a lattice point at an end of the learning map is a fixed value.

Description

  The present invention relates to a vehicle control device, and more particularly, to a vehicle control device suitable for controlling a vehicle having a control parameter learning map used for controlling an onboard device such as an internal combustion engine.

  Conventionally, for example, Patent Document 1 discloses a control device for an internal combustion engine including a control parameter learning map. A learning value for correcting the control parameter is stored in association with each lattice point of the learning map. In this conventional control device, when learning data of a control parameter is acquired, four lattice points located around the acquired value are selected on the learning map, and the learning values of these four lattice points are updated. It is configured to do. In this learning control, the acquired value of the control parameter is weighted and then reflected in the learned value of surrounding grid points. The weight at this time increases as the distance between the position of the acquired value and the grid point decreases. Is set to be

JP 2009-046988 A JP-A-62-045955 Japanese Patent Laid-Open No. 11-125138

  As in the control apparatus for an internal combustion engine described in Patent Document 1, in the configuration including a learning map in which learning values of control parameters used for vehicle control are associated with each of a plurality of grid points, When the learning value that reflects the acquired value of the learning data of the control parameter is calculated at the predetermined learning target lattice point, the learning target lattice point is the lattice point (end point) at the end on the learning map. Therefore, a part of the effective range (learning range) of the learning data used for calculation of the learning value is outside the learning map setting area (the area where learning data cannot be obtained). In other words, at the lattice points (end points) at the end of the learning map, the area where the learning data to be reflected in the learning value is limited. For this reason, there will be a bias in the distribution of learning data that is the target of calculation of learning values at the end points. As a result, there is a concern that the calculation accuracy of the learning value at the end point is deteriorated.

  The present invention has been made in order to solve the above-described problems, and provides a vehicle control apparatus that can suppress deterioration in learning value calculation accuracy at lattice points (end points) at the ends of a learning map. The purpose is to provide.

A first invention is a control device for a vehicle,
A learning map that is formed in a lattice shape, and learning values of control parameters used for vehicle control are associated with each lattice point;
Weighting according to the distance between the learning target grid point and the learning data is applied to each of the plurality of learning data of the control parameter included in the predetermined learning range from the position of the predetermined learning target grid point to be learned. Learning value calculating means for calculating a learning value of the learning target grid point using the weighted learning data;
Learning value updating means for updating the learning value of the learning target lattice point using the learning value calculated by the learning value calculating means;
A vehicle control device comprising:
The learning value associated with the grid point at the end of the learning map is a fixed value.

The second invention is the first invention, wherein
The learning map includes a real map area having a plurality of grid points, and a virtual map area formed by extending the real map area,
The learning value associated with the first grid point at the end of the learning map including the virtual map region is a fixed value,
The learning value calculation means uses the learning value of the second grid point at the end of the real map area as the learning value of the first grid point and the grid inside the second grid point in the real map area. It is calculated by interpolation based on the learning value of the point.

The third invention is the first or second invention, wherein
For a lattice point having a bias of a predetermined degree or more in the distribution of learning data included in the learning range, a learning value associated with the lattice point is a fixed value.

  According to the first aspect of the present invention, the learning value associated with the lattice point (end point) at the end of the learning map is set to a fixed value, so that the distribution of the learning data for which the learning value is calculated at the end point is obtained. It can suppress that the calculation precision of the learning value in the said end point deteriorates by the influence of bias | inclination.

  According to the second invention, the map shape around the second grid point (end point of the actual map area) even when the learning value of the grid point inside one of the end points of the actual map area changes greatly. Can be suppressed. As a result, even in the above case, it is possible to satisfactorily secure the calculation accuracy of the learning value at the second grid point (end point of the actual map area).

  According to the third aspect of the present invention, it is possible to prevent the learning value calculation accuracy from deteriorating due to the influence of bias on the distribution of learning data to be subjected to learning value calculation.

It is a whole block diagram for demonstrating the system configuration | structure of Embodiment 1 of this invention. In Embodiment 1 of this invention, it is explanatory drawing which shows typically an example of the learning map used for weighted learning control. It is the figure which illustrated the relative relationship between the map area | region and learning range of a learning map. It is a figure for demonstrating the effect by the learning method of Embodiment 1 of this invention. It is the figure which represented the learning result by the learning method of Embodiment 1 of this invention compared with the ideal learning result. It is a figure for demonstrating the subject of the learning method of Embodiment 1 of this invention. It is a flowchart which shows the procedure of the learning method of Embodiment 2 of this invention. It is a figure showing an example of the learning result at the time of using the learning method of Embodiment 2 of this invention. It is a figure for demonstrating the effect by the learning method of Embodiment 2 of this invention. It is a figure showing an example of the acquisition situation of the learning data on a learning map. It is a figure for demonstrating the learning method of Embodiment 3 of this invention.

Embodiment 1 FIG.
Hereinafter, Embodiment 1 of the present invention will be described with reference to FIGS.
[Hardware Configuration of Embodiment 1]
FIG. 1 is an overall configuration diagram for explaining a system configuration according to the first embodiment of the present invention. The system shown in FIG. 1 includes an internal combustion engine (as an example, a spark ignition internal combustion engine) 10 as a vehicle-mounted device (not shown). In each cylinder of the internal combustion engine 10, a combustion chamber 14 is formed on the top side of the piston 12, and the piston 12 is connected to a crankshaft 16. Further, the internal combustion engine 10 includes an intake passage 18 that sucks intake air into each cylinder. The intake passage 18 is provided with an electronically controlled throttle valve 20 that adjusts the amount of intake air. On the other hand, the internal combustion engine 10 includes an exhaust passage 22 that exhausts the exhaust gas of each cylinder, and a catalyst 24 such as a three-way catalyst that purifies the exhaust gas is provided in the exhaust passage 22. Each cylinder of the internal combustion engine 10 includes a fuel injection valve 26 that injects fuel into the intake port, an ignition plug 28 that ignites the air-fuel mixture, an intake valve 30 that opens and closes the intake port, and an exhaust valve that opens and closes the exhaust port. 32.

  Next, a control system installed in the system of this embodiment will be described. The system of the present embodiment includes a sensor system including various sensors necessary for driving a vehicle including the internal combustion engine 10 and an ECU (Engine Control Unit) 50 that controls the operation state of the internal combustion engine 10. First, the sensor system will be described. The crank angle sensor 34 detects the crank angle and the engine speed, and the air flow sensor 36 detects the intake air amount. The water temperature sensor 38 detects the water temperature of the engine cooling water, and the in-cylinder pressure sensor 40 detects the in-cylinder pressure.

  The ECU 50 is configured by an arithmetic processing unit that includes a storage circuit including a ROM, a RAM, a nonvolatile memory, and the like, and an input / output port. A learning map to be described later is stored in the nonvolatile memory of the ECU 50. Each sensor of the sensor system is connected to the input side of the ECU 50. An actuator such as a throttle valve 20, a fuel injection valve 26, and a spark plug 28 is connected to the output side of the ECU 50. And ECU50 drives each actuator based on the driving | operation information of the internal combustion engine 10 detected by the sensor system, and performs driving | operation control. Specifically, the engine speed and the crank angle are detected based on the output of the crank angle sensor 34, and the intake air amount is detected by the air flow sensor 36. Further, the engine load is calculated based on the engine speed and the intake air amount, the fuel injection amount is calculated based on the intake air amount, the engine load, the water temperature, etc., and the fuel injection timing and ignition timing are calculated based on the crank angle. To decide. Then, the fuel injection valve 26 is driven when the fuel injection timing comes, and the spark plug 28 is driven when the ignition timing comes. Thus, the air-fuel mixture is combusted in each cylinder, and the internal combustion engine 10 is operated. The ECU 50 executes predetermined engine control such as air-fuel ratio feedback control in addition to the above-described ignition timing control and fuel injection control.

[Characteristics of Embodiment 1]
(Overview of weighted learning control)
In general, in engine control, learning control for learning control parameters based on acquired values (learning data) of various control parameters is performed. In this specification, “acquisition” includes the meanings of detection, measurement, measurement, calculation, estimation, and the like. In this embodiment, weighted learning control using weighted average processing described below is executed as learning control. The ECU 50 constitutes a learning device that performs weighted learning control, and includes a learning map having a plurality of lattice points.

  FIG. 2 is an explanatory diagram schematically showing an example of a learning map used for weighted learning control in Embodiment 1 of the present invention. In FIG. 2, one learning value (eg, valve timing or fuel injection amount increase value) is calculated based on two reference parameters (eg, engine load and engine speed) corresponding to the X axis and Y axis. A two-dimensional (one example) learning map is illustrated. A learning value (map value) of a control parameter used for engine control is stored in association with each lattice point of the learning map.

  According to the weighted learning control of the present embodiment, in a situation where a plurality of control parameter learning data is acquired, a plurality of learning parameters included in a predetermined learning range from the position of a predetermined learning target lattice point to be learned. Each learning data is weighted according to the distance between the learning target lattice point and each learning data, and the learning value of the learning target lattice point is calculated using the weighted learning data. Then, the learning value of the learning target lattice point is updated using the calculated learning value.

  More specifically, the weighted learning control of the present embodiment is performed using a plurality of learning data obtained during a predetermined time. FIG. 2 illustrates a case where four learning data (square marks) are obtained around the lattice point A. In this case, with the lattice point A as the learning target lattice point, the weighted average is used using the four learning data included in the learning range (the range indicated by the dotted circle) centered on the learning target lattice point A. A value is calculated, and a process of reflecting the calculated weighted average value on the learning value of the learning target grid point A is performed. Such learning is not limited to one lattice point (lattice point A in FIG. 2), but when there is a lattice point from which a plurality of learning data is obtained within its own learning range, This is executed as a learning target grid point.

  The weighted average process when reflecting the learning data in the learning value can be performed by the following method, for example. That is, taking the case of the lattice point A shown in FIG. 2 as an example, for the acquired learning data xi (i is 1 to 4 in the example of FIG. 2), the learning target lattice point A and the learning data xi The weight wi according to the distance is set. More specifically, the weight wi is set to be larger as the distance between the learning target grid point A and the learning data xi is shorter.

  Then, the learning value of the learning target lattice point A is calculated based on the learning data xi and the weight wi. More specifically, the product of the learning data xi and the weight wi is calculated for each learning data xi, and the sum of the integrated values wixi for each learning data xi is divided by the sum of the weights wi of each learning data xi. Thus, the weighted average value is calculated. Then, the learning value of the learning target grid point A is calculated and updated based on the calculated weighted average value. Thereby, the larger the weight wi assigned to the learning data xi, the larger the learning data xi is reflected in the learning value. Therefore, the degree to which the learning data xi is reflected in the learning value of the learning target lattice point A (learning effect) becomes larger as the learning data xi is closer to the learning target lattice point A, and the learning data xi is changed from the learning target lattice point A. It gets smaller as it gets farther away.

  The process of updating the learning value may be, for example, using the calculated weighted average value as a new learning value, or correcting the current learning value using the calculated weighted average value. May be. In addition, the setting of the weight wi in the present embodiment is not particularly limited as long as it is a method that can be set to a value corresponding to the distance between the learning target grid point and the learning data xi, but for example, performed using a Gaussian function. Can do.

(Problems when learning grid points (end points) at the end of the learning map)
FIG. 3 is a diagram illustrating the relative relationship between the map area of the learning map and the learning range.
Two circles represented by broken lines in FIG. 3 indicate learning ranges when the lattice points A and B are learning target lattice points, respectively. That is, at the lattice points A and B, learning using learning data acquired within the respective learning ranges is executed.

  As in the ECU 50 of the present embodiment, in a configuration including a learning map in which learning values of control parameters used for controlling the internal combustion engine 10 are associated with a plurality of lattice points, a predetermined learning object to be learned When calculating a learning value reflecting the acquired value of the learning data of the control parameter at the lattice point, the lattice point at which the learning target lattice point is at the end on the learning map is the same as the lattice point B in FIG. (End point), a part of the effective range (learning range) of the learning data used to calculate the learning value is outside the learning map setting area (map area) (area where learning data cannot be obtained) turn into. As a result, if the lattice point is A, the entire learning range is the map region, but no learning data is obtained outside the map region, so the substantial learning range of the lattice point B is shown in FIG. As shown, it will be only the range of the left semicircle.

  As described above, at the lattice points (end points) at the end of the learning map, a region where learning data to be reflected in the learning value is obtained is limited. For this reason, there will be a bias in the distribution of learning data that is the target of calculation of learning values at the end points. As a result, there is a concern that the calculation accuracy of the learning value at the end point is deteriorated.

(Characteristic learning method in the first embodiment)
Therefore, in this embodiment, the learning value (map value) associated with the lattice point at the end among the plurality of lattice points included in the learning map is set to a fixed value set in advance. In other words, the end point is not learned.

  FIG. 4 is a diagram for explaining the effect of the learning method according to the first embodiment of the present invention. More specifically, FIG. 4B shows an example of a learning result when the learning method of the present embodiment is used, and FIG. 4A is referred for comparison with the learning method of the present embodiment. It is a figure which shows an example of a learning result. 4A and 4B, the horizontal axis in the upper diagram corresponds to the row (horizontal column) to which the grid points A and B belong in the grid-like learning map in the lower diagram. The shape obtained by connecting the circles with dotted lines indicates the initial map shape (the shape obtained by connecting each initial value (map value when creating the learning map) and each grid point).

  First, a case where the learning method of the present embodiment is not used (that is, a case where learning is performed at the end point B) will be described with reference to FIG. The learning result shown in FIG. 4 (A) is obtained by using each of the learning points shown in FIG. 4 (A) as learning target lattice points for each lattice point of the row to which lattice points A and B belong. It is a thing when it goes. As a result, from FIG. 4A, in the map values after learning of the lattice points excluding the end points (diamond marks), the learning data located in the vicinity of each lattice point (in the left and right directions of each lattice point) has been described above. Since the weight is reflected on the map value, it can be seen that the calculation (learning result) of the map value along the distribution of the learning data is obtained. However, the end point B (same as the end point opposite to the end point B) limits the range in which the learning data can be obtained as described above with reference to FIG. 3, so that the learning value (map value) is calculated. The distribution of the learning data used is biased. As shown in FIG. 4A, map data (learning values) that are apart from the neighboring learning data and the initial value are calculated at the end point B and the opposite end point as affected by the bias of the learning data. End up. As a result, the calculation accuracy of the learning value at the end point deteriorates (a large learning error occurs).

  On the other hand, according to the learning method of the present embodiment that does not perform learning at the end points, the map value of the end points is a fixed value (it remains the initial value), so as shown in FIG. Compared with the case of FIG. 4A, the map value (learning result) after learning can be made closer to the acquired value of learning data also at the end point B and the opposite end point. Thereby, compared with the case of FIG. 4 (A), a learning error can be suppressed.

FIG. 5 is a diagram illustrating a learning result obtained by the learning method according to the first embodiment of the present invention in comparison with an ideal learning result.
As described above with reference to FIG. 4, according to the learning method of the first embodiment in which the map value of the end point is fixed and learning is not performed at the end point, compared with the conventional method in which learning is performed at the end point. The map shape after learning between the end point B and the adjacent lattice point A (the same applies to the end point on the opposite side) can be made closer to the ideal learning result (map shape obtained by connecting each learning data). it can. As a result, the calculation accuracy of the learning value at the end point can be secured satisfactorily (learning error can be suppressed).

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described with reference to FIGS.
The hardware configuration of the second embodiment is the same as that of the first embodiment described above.

[Characteristics of Embodiment 2]
(Problem of learning method of Embodiment 1)
FIG. 6 is a diagram for explaining the problem of the learning method according to the first embodiment of the present invention.
FIG. 6 is a diagram showing a map value learning result by the learning method of the first embodiment described above (a method that does not perform learning at the end points).

  In the learning result shown in FIG. 6, the learned map value (diamond mark) of the lattice point A inside one of the end points B is greatly changed (separated) from the initial value (circle mark). In such a case, if the learning method according to the first embodiment in which the map value of the end point is fixed at the initial value is used, the end point B is accompanied by a change in the map value of the lattice point A as shown in FIG. The map shape around the area will change. As a result, the learning error of the map value of the end point B becomes large.

(Characteristic learning method in Embodiment 2)
FIG. 7 is a flowchart showing the procedure of the learning method according to the second embodiment of the present invention. FIG. 8 is a diagram illustrating an example of a learning result when the learning method according to the second embodiment of the present invention is used. Note that the processing shown in FIG. 7 may be performed every time the end point map value is learned, or the map value of the grid point inside one of the end points has greatly changed from the initial value. It may be performed only in a case (case as shown in FIG. 6 above).

  According to the processing procedure shown in FIG. 7, first, as shown in FIG. 8B, a learning map area (hereinafter referred to as “real map area”) in which a grid point associated with a map value is present. A virtual map area is created by expanding and providing a grid point positioned one outward from the end point of the real map area (the grid point on the thick frame to which the end point B belongs in FIG. 8B) ( Step 100).

  Next, a predetermined map value is set for a grid point on the virtual map area (the end point C corresponds to this), and the map value is not learned at the grid point on the virtual map area ( Step 102). That is, in the present embodiment, the learning concept of the first embodiment described above in which learning is not performed at end points using a fixed value is applied to lattice points on the virtual map area.

  Next, as shown in FIG. 8A, the map value after learning at the lattice point A inside the end point B of the real map region and the map value (fixed value) of the lattice point C on the virtual map region. A value (triangle mark) taken at the end point B by a line connecting the lines (a straight line as an example in FIG. 8A) is calculated as a map value of the end point B (step 104). That is, according to the processing of step 104, based on the map value after learning at the lattice point A and the map value (fixed value) of the lattice point C on the virtual map area, the lattice point A and the lattice point C The map value of the end point B located between the points is calculated by interpolation. Thus, also in the present embodiment, learning using the weighted average process is not performed for the end point B of the actual map area.

FIG. 9 is a diagram for explaining the effect of the learning method according to the second embodiment of the present invention.
According to the learning method of the present embodiment described above, even when the map value of the grid point A inside the end point B of the actual map region is greatly changed, as shown in FIG. Compared with the learning method of Form 1, the change in the map shape around the end point B can be suppressed. As a result, even in the above case, the calculation accuracy of the map value (learning value) of the end point B (as well as the end point on the opposite side of the end point B as shown in FIG. 9) can be ensured satisfactorily (learning). Error can be suppressed).

  When the learning method of the present embodiment is used, the map value of the end point B according to the present learning method is changed by appropriately changing the distance D from the end point B of the real map region to the lattice point C on the virtual map region. It is possible to adjust the degree of correction.

Embodiment 3 FIG.
Next, Embodiment 3 of the present invention will be described with reference to FIG. 10 and FIG.
The hardware configuration of the third embodiment is the same as that of the first embodiment described above.

[Characteristics of Embodiment 3]
(Problem assumed in the third embodiment)
FIG. 10 is a diagram illustrating an example of an acquisition state of learning data on the learning map.
As shown in FIG. 10, there are areas where learning data is easily obtained and areas where it is difficult to obtain learning data in the area on the learning map. As a result, the distribution of learning data to be learned at each lattice point is biased. Specifically, in the situation shown in FIG. 10, the distribution of learning data in the learning range is uniform (no bias) for lattice point A, but the learning data in the learning range for lattice point B. The distribution of is non-uniform (biased). For this reason, the learning result at the lattice point B is affected by the bias of the learning data distribution, and there is a concern that the learning value calculation accuracy deteriorates.

(Characteristic learning method in Embodiment 3)
FIG. 11 is a diagram for explaining the learning method according to the third embodiment of the present invention.
In order to solve the above problem, in the present embodiment, as shown in FIG. 11A, the map value (learned value) of the grid point (white circle) where it is difficult to obtain the learning data is a fixed value set in advance. It was made to be a value. That is, learning is not performed at the target lattice point (the learning result is not validated). More specifically, at the time of creating the learning map, a grid point that is expected to cause a deviation of a predetermined degree or more in the distribution of the learning data included in the self learning range is selected, and the map value of the selected grid point is It is a fixed value. Note that the learning method of the first or second embodiment described above is applied to the end points.

  On the other hand, in the case of a grid point where learning data is difficult to obtain, even if it is a grid point set in advance, learning data can be obtained appropriately afterwards due to a change in the learning data acquisition status. Sometimes it becomes. Therefore, in the present embodiment, the learning result is validated (that is, the map value is unfixed) for the lattice points for which it is determined that the learning data is appropriately obtained.

  Judgment as to whether or not learning data has been appropriately obtained can be made using the following method. That is, for example, it can be performed based on whether learning is performed at a lattice point around the lattice point to be determined (for example, the nearest eight lattice points) (whether the map value is not fixed). . Or, for example, whether the learning data exists evenly within the learning range of the lattice point to be determined (more specifically, the determination target is like the lattice point indicated by a black circle in FIG. 11B). If there is learning data in all four quadrants within the learning range when the lattice point becomes the center, it can be performed. Contrary to the above processing, if there is a lattice point where learning data is difficult to obtain due to a change in the learning data acquisition status, the learning result is invalidated (that is, the map value is fixed). You may do it.

  According to the learning method of the present embodiment described above, for the grid points where it is difficult to obtain learning data, the map value is fixed and learning is not performed. It is possible to prevent the calculation accuracy of the value (learning value) from deteriorating (learning errors can be suppressed). Further, according to the learning method described above, such a countermeasure can be appropriately performed based on a change in the acquisition state of learning data.

  By the way, in the first to third embodiments described above, a configuration including a learning map for learning values of control parameters used for controlling the internal combustion engine 10 that is one of the on-board devices is exemplified. The learning method was explained. However, the learning map that is the subject of the present invention is not limited to the target for the internal combustion engine as long as it is for the learning value of the control parameter used for controlling the vehicle, and is used for the control of other on-vehicle equipment. It may be for the learning value of the control parameter.

DESCRIPTION OF SYMBOLS 10 Internal combustion engine 12 Piston 14 Combustion chamber 16 Crankshaft 18 Intake passage 20 Throttle valve 22 Exhaust passage 24 Catalyst 26 Fuel injection valve 28 Spark plug 30 Intake valve 32 Exhaust valve 34 Crank angle sensor 36 Air flow sensor 38 Water temperature sensor 40 In-cylinder pressure sensor 50 ECU (Electronic Control Unit)

Claims (3)

A learning map that is formed in a lattice shape, and learning values of control parameters used for vehicle control are associated with each lattice point;
Weighting according to the distance between the learning target grid point and the learning data is applied to each of the plurality of learning data of the control parameter included in the predetermined learning range from the position of the predetermined learning target grid point to be learned. Learning value calculating means for calculating a learning value of the learning target grid point using the weighted learning data;
Learning value updating means for updating the learning value of the learning target lattice point using the learning value calculated by the learning value calculating means;
A vehicle control device comprising:
The vehicle control apparatus according to claim 1, wherein a learning value associated with a lattice point at an end of the learning map is a fixed value. The learning map includes a real map area having a plurality of grid points, and a virtual map area formed by extending the real map area,
The learning value associated with the first grid point at the end of the learning map including the virtual map region is a fixed value,
The learning value calculation means uses the learning value of the second grid point at the end of the real map area as the learning value of the first grid point and the grid inside the second grid point in the real map area. The vehicle control device according to claim 1, wherein the vehicle control device is calculated by interpolation based on a learned value of a point.   3. A lattice value having a bias of a predetermined degree or more in a distribution of learning data included in the learning range, the learning value associated with the lattice point is a fixed value. The vehicle control device described in 1.

Images