JP3512845B2 - Process operation amount control method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for controlling a process operation amount by a control device, in particular a vehicle control device, using a characteristic map, in particular a process operation amount in a vehicle, The characteristic map is represented by a certain number of base (quasi) points defined by the operation amount of the process and stored in the memory of the controller, and for each base (quasi) point, an address of the characteristic map. At least one useful information regarding the value of the characteristic map at the position of the base (quasi) point in space is stored in the memory of the controller and is close to the operating point of the process. Starting from a method of the type in which at least one base (quasi) point is read out and at least one controlled variable for delivering a control signal is formed taking into account the base (quasi) point. .

[0002]

2. Description of the Related Art A method for operating a process using a characteristic map is disclosed in West German Patent Application Publication No. 34082115.
It is already known from the publication. The method described therein is used for controlling the operating characteristic quantity of an internal combustion engine. In particular, there are known methods for adjusting λ, adjusting knock, adjusting ignition timing and adjusting injection.

In this method, the characteristic map stored in the memory of the control device is used to perform preliminary control of the operation amount to be adjusted. In that case, the superimposed adjustment method implements a multiplying or even additive control action of the read characteristic map value. Furthermore, the characteristic map value can itself be changed by means of a superimposed adjustment. Moreover, this change adjustment can be carried out continuously during the operation, so that a self-alignment of the characteristic map is realized.

The characteristic map is represented by base (quasi) points. In this case, the base (quasi) points are fixedly arranged on the characteristic map according to a predetermined pattern. For each base (quasi) point, the value of the characteristic map at the position of the base (quasi) point is written in the memory of the control device. The characteristic map used is at most three-dimensional. When determining the characteristic map value for the operating point that is not located in the node of the predetermined pattern, interpolation is performed by using the base (quasi) points located around the operating point. To be done. The pattern feature map used therein presents a problem for time-varying systems where feature map matching is required.

The learning method for the pattern characteristic map is as follows.
Normally, measurements located at any point between the base (quasi) points must be extrapolated to surrounding base (quasi) points. As a result, a base (quasi) point match is often not possible in one step. Rather, there is a possibility that the base (quasi) point that has already been well matched by the correction value obtained in the periphery may be erroneous again. Under unfavorable preconditions (time course of operating points and selection of learning parameters), this coupling of the basal (quasi) points can lead to an oscillating tendency, which is a checkerboard-shaped, basal (quasi) point. The alternating changes of rising and falling of. Therefore, currently only coarse corrections are performed, eg only multiplicative or additive shifts of the entire characteristic map.

Furthermore, the basal (quasi) point densities are not locally matchable, but only linearization of the individual address components, which behaves uniformly in the entire address space, can be performed. Coarse quantization limits the maximum characteristic map dynamics, while finer quantization does not always match all base (quasi) points with sufficient frequency in some rarely used regions. ,
As a result, an uncaptured gap occurs. This results in the problem that in coarse quantization and characteristic map matching, information cannot be written at the location where it occurred.

[0007]

SUMMARY OF THE INVENTION The object of the present invention is to improve a method of the type mentioned at the outset such that the origin (quasi) point of the characteristic map can be placed at any position of the characteristic map irrespective of a fixed pattern. It is to be.

[0008]

[Means for Solving the Problem] This problem is that, in addition to each base (quasi) point, a base (quasi) in the address space of the characteristic map is added.
At least one piece of information about the position of the point is stored in the memory of the control device so that the base (quasi) point can be placed and operated anywhere in the address space of the characteristic map irrespective of the fixed pattern of actuation quantities. Solution by performing a distance calculation that calculates the distance from at least one base (quasi) point in the characteristic map to the operating point in order to select at least one base (quasi) point that is as close as possible to the point To be done.

[0009]

The method for operating a process using a characteristic map of the present invention is, compared with the prior art, that the origin (quasi) point of the characteristic map is set at an arbitrary position of the characteristic map regardless of a fixed pattern. It has the advantage that it can be placed in This allows the base (quasi) point density to be matched to local accuracy requirements. This makes it possible to simulate very high local characteristic map dynamics, ie gradient changes of the function to be stored, with a minimum training cost. In particular, the required base points (quasi-points) are determined not by the characteristic map dimensions (dimensions) but only by the required simulation accuracy, which makes it possible to effectively handle multidimensional problem settings.

Since the base (quasi) point can be written anywhere in the address space, measurement and optimization by the application side (application) is simplified. further,
Main problems in conforming characteristic maps, peripheral groups (quasi)
Extrapolation of the information to the points is avoided, which is an effective match of the characteristic map structure to manufacturing deviations, environmental influences and aging phenomena, i.e. simply to the known global shifts or subregions from the pattern characteristic map. Not only the shift, but also the matching of the individual base (quasi) points enables the accurate matching of the characteristic map structure.

The possibilities for effectively realizing a multidimensional characteristic map are obtained. The required storage capacity is
It does not grow exponentially as the dimension of the characteristic map increases, as in the case of the pattern characteristic map, but exclusively depends on the predetermined accuracy requirements. Furthermore, the group (quasi)
It is possible to accurately place the point at any predetermined position. This is, for example, based on a position which is of particular importance for exhaust gas inspection or which makes it possible to measure vehicle equipment measurements on the inspection table particularly easily.
It is advantageous for placing the points.

Further, the variable cost of influence of the base (quasi) point reduces training costs and avoids gaps in non-capturing states when using a learning method for matching the characteristic map to changed conditions. Is advantageous.

With the features of claims 2 to 32, advantageous embodiments and refinements of the method of claim 1 are possible. That is, it is particularly advantageous to store the address information of the base (quasi) point and the valid information with the same right. This allows reverse access of the characteristic map, ie the exchange of addresses and valid information. This is particularly advantageous in learning-capable systems. Because 2
This is because it is extremely difficult, even if possible, to make two separate pattern characteristic maps correspond to each other at the same time. further,
For individual (quasi) points or adjacent (quasi) points,
It is advantageous to store the peripheral information in the memory of the control device.
Since the operating point of a technical process usually moves continuously through the address space of the characteristic map with a limited rate of change, the control device can more quickly reach the base (quasi) point relevant to the operating point. Can be accessed. By introducing the surrounding structure, it is possible to eliminate the ambiguity that may exist in the reverse property map access. This is best possible with a very finely defined peripheral structure, the radix (quasi) point association for pointers.

It is advantageous to store a pointer to an adjacent base (quasi) point for each base (quasi) point. This allows for very fast access to adjacent radical (quasi) points. This allows
Searching for adjacent base (quasi) points is greatly simplified. Furthermore, it is advantageous to divide the address space of the characteristic map into a plurality of areas and to allocate a predetermined segment in the memory of the control device to each area and store the base (quasi) point of each area in a corresponding memory segment. . This also allows the peripheral information to be stored in the memory of the control device, so that the relevant (quasi) point for the relevant operating point can be detected more quickly. This method is particularly suitable for the use of parallel processing hardware structures, for example content-addressable memories, due to the division of the characteristic map into regions of adjacent base (quasi) points.

The required storage capacity for the characteristic map is vastly increased by performing an intermediate value calculation from the characteristic map values of the appropriate base (quasi) points to determine the characteristic map value for a given operating point of the process. It is possible to accurately reproduce the characteristic map without creating a cause. In that case, it is advantageous to select, as a suitable base (quasi) point, a base (quasi) point which is close to the motion point and is uniformly distributed around the motion point. In selecting this type of base (quasi) point, the probability for the most accurate characteristic map simulation is the greatest. This reduces the probability of performing extrapolation to determine the characteristic map value. In particular, claims 7 to 1
The criteria described in 0 provide an advantageous search strategy.

Furthermore, it is advantageous to form the L ₁ norm (city block distance) instead of the Euclidean norm in the distance calculation for detecting the base (quasi) point located as close as possible to the operating point. This is because the calculation of the L ₁ norm requires a small calculation cost. Furthermore, it is advantageous to use n + 1 base (quasi) points for the calculation of the intermediate values in the n-dimensional characteristic map. This further reduces the calculation cost. The multi-step configuration of property map access is likewise advantageous to enable rapid detection of suitable base (quasi) points. At that time, when accessing the base (quasi) point of the characteristic map,
In a step, a region containing a desired base (quasi-) point is determined, and in a second step, a segment in the memory of the controller corresponding to the determined region is unstructured in a suitable group (quasi-). ) It is advantageous to search for points.

Furthermore, it is advantageous to carry out a fixed address calculation in order to determine the region containing the desired base (quasi) point. This enables quick access to the rough neighborhood of the operating point at that time.

A preferred area division of the address space is an area division with orthogonally extending area boundaries, the division for each address component being carried out separately. This division is particularly advantageous for fixed address calculation for the first access stage. This is because the computational cost is small in this case. Implementation of area division according to the criterion that the number of base (quasi) points in each area is as large as possible is said to avoid a relatively large time difference when searching for an address space area. Advantageous for reasons. An advantageous division of the characteristic map is that, for each address component, the total (or quasi) point for each address component is (K-1) region boundaries of these address components, either manually or automatically. It is carried out by determining that it is divided into K pieces of adjacent base (quasi) points, which are as large as possible. With this division of the region boundaries, a relatively uniform division of the base (quasi) points for the individual regions is achieved.

Further advantageous possibilities of the characteristic map division are as follows. That is, each of the base (quasi) points of one area is associated with one representative point which is also written in the memory of the control device, and in this case, this representative point corresponds to the address component of that location in the characteristic map and this area. A pointer to the beginning of the corresponding memory segment in which the base point is stored. With this configuration, a more free form of the region boundaries is possible in order to evenly distribute the base (quasi) points to the region, for example in the case of locally (base) quasi-point density. A simple correspondence of a base (quasi) point to a representative point is to add all the base (quasi) points of the characteristic map that are located closer to each representative point than any other representative point. It is realized by associating. Conversely, the position of each representative point,
It is also advantageous if the average value of the address components of all the base (quasi) points associated with the representative point is determined by determining the position of the representative point. In order to carry out the first access step, the characteristic map area at that time can be determined very simply by determining the representative point located closest to the operation point at that time. At this time, an appropriate interpolation base (quasi) point can be selected by, for example, an unstructured search for the base (quasi) point associated with this point.

Furthermore, it is advantageous if the characteristic map is automatically matched to the changed process characteristics of the technological process. As a result, the learning ability of the control device regarding the characteristic map is realized. In matching the characteristic map with the changed process characteristic, it is advantageous to write a new origin (quasi) point in the memory of the control unit. In that case, the correction of the existing base (quasi) point is no longer necessary and the new base (quasi) point can be located anywhere in the address space. In order to further reduce the required storage capacity for the characteristic map, it is advantageous to erase the previous base (quasi) point.

Furthermore, it is advantageous to use an addressable memory of the content in order to store the (quasi) points of the characteristic map and to select the appropriate (quasi) points. The use of content addressable memory has the advantage that the appropriate origin (quasi) point can be quickly detected in a very short calculation time.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in detail with reference to the accompanying drawings with reference to the accompanying drawings.

In FIG. 1, a control device is shown by 10. The control device is provided for controlling the technical process. An example is an engine control device. This control device controls the combustion process of the internal combustion engine as a technical process. Reference numeral 14 designates the technical process in this case, i.e. the combustion process in an internal combustion engine. The configuration of controller 10 is shown only roughly. Reference numeral 11 here designates a micro computer. Reference numeral 12 designates the memory of the control device 10. Further, reference numeral 13 indicates an input / output unit. The specific configuration of the control device 10 for controlling the internal combustion engine is well known from the prior art. In a 4-cycle engine, ignition and gasoline injection can be electronically controlled. For controlling the ignition angle, the microcomputer 11 calculates the ignition angle between the two ignition processes from the information load and the rotation speed. The information load and the rotation speed are controlled by the control device 1 via the connected generator.
Supplied to zero. A complicated ignition characteristic map is stored in the memory 12 of the control device 10. Micro computer 11
For the calculation of the ignition angle between the two ignition processes, the values belonging to the load value and the rotational speed value at that time are extracted from the stored ignition characteristic map. The ignition angle value extracted from the characteristic map by the micro computer 11 is corrected by the micro computer 11 as necessary. This depends on one or more of the combustion chamber pressure course, knock signal, exhaust gas temperature, etc., depending on other variables that influence, such as engine temperature, intake air temperature or in the form of a closed loop regulation circuit. It can be carried out by feedback of the process quantity and by evaluation by means of suitable regulators.
The control device 10 then triggers the ignition process at the corresponding ignition timing. The same control device 10 similarly controls the mixture preparation. The injection valve is then operated at the right time, and then the injection valve injects a fixed quantity of fuel into the intake section of the internal combustion engine. The composition of the mixture then depends on the operating state of the internal combustion engine. Micro computer 11
Forms the injection time from the air amount signal, the rotation speed signal, and the correction coefficient as required. The air quantity signal and the speed signal provide a measure for the engine load (air quantity per stroke). The microcomputer 11 retrieves the optimum injection amount for the operating state from the injection characteristic map stored in the memory 12 of the control device 10. The injection characteristic map is defined by the operating parameters engine load and engine speed. Further details on engine control can be found in the printed material BoschTechnishe Unterrichtung “kombi
niertes Zuend und Benzineinspritzsystem ”(198
January, 3). The control device 10 and the process 14 to be controlled will now be described in detail only as far as the characteristic map in which it is stored and the access to these characteristic map data of the microcomputer 11 are concerned.

FIG. 2 shows an example of a simple characteristic map as stored in the memory 12 of the control device 10. The address space of the characteristic map is then fixed by the quantities X ₁ and X ₂ . These quantities correspond to the quantity engine load and the engine speed both in the ignition characteristic map and the injection characteristic map. The characteristic map is not completely contained in the memory 11 of the control device 10 but is represented by the selected base (quasi) point 20. Direct storage to the fixed (quasi) point 20 at that time via storage of fixed fixed (quasi) point patterns and fixed address calculation, for example using a fixed (quasi) point table, as is customary in conventional property map structures. Access is based on 20 points
Abandoned for free placement. Instead, at least one piece of valid information (output amount, function value), associated address information (input value, position in address space) and, in some cases, peripheral information (pointer to adjacent base (quasi) point, tree structure) A new base (quasi) point format is formed in which unit information consisting of array / order information in (1) is stored. The valid and address information in this case is to allow reverse access (exchange of valid information and address), if the problem setting requires this,
Treated with equal rights. This gives a base (quasi) point of 2
Zeros can be stored anywhere in the address space and the local radical (quasi) point densities can be matched arbitrarily. At the time of reading the effective information at that time, a base (quasi) point 20 that represents the operating point at that time as well as possible is searched. For interpolation, the base (quasi) points 20 that are located as close as possible to the operating point and are distributed as uniformly as possible around this are selected for interpolation (to avoid extrapolation, Operating point at the center of gravity of the base (quasi) point). A method for determining the most advantageous interpolation base (quasi) points for the operating points in the characteristic map address space and a few possible address space distributions of the characteristic map will now be described.

FIG. 3 shows a base (quasi) point 20 in which an address space for a three-dimensional characteristic map is written. Furthermore, the current operating point 21 of the technical process is illustrated. For each base (quasi) point 20, address information, that is, its coordinates (X ₁ ,
X ₂ ) and the associated function value are stored in the memory 12 of the controller 10.
Written in. An ordinary example in which the operating point 21 at that time does not coincide with the base (quasi) point 20 of the characteristic map is shown. In order to read the information of the characteristic map for this operating point 21, interpolation is performed from the base (quasi) point 22 closest to the operating point 21. At this time, the microcomputer 11 searches the entire address space, that is, all the base (quasi) points 20 in which the characteristic map is stored, by searching for suitable base (quasi) points 22. At that time, the microcomputer executes the distance calculation. Therefore, the micro computer reads the base (quasi) point 2
The distance between 0 and the operating point 21 at that time is calculated. Starting from this calculation, the micro-computer selects three suitable interpolation base (quasi) points 22. That is, an unstructured search for the peripheral base (quasi) point most relevant to the operating point 21 at that time is performed. The micro computer advantageously calculates this distance not according to the Euclidean norm, but according to the L ₁ norm, which is also known in the literature as the conceptual city block distance. The distance between the two points P1, P2 in the address space is then calculated according to the following formula:

[0026]

[Equation 1]

Where n = dimension of the characteristic map.

The distance calculation is indicated by the arrow in FIG. The micro-computer selects the three closest base (quasi) points for interpolation. When an n-dimensional characteristic map is handled, the micro computer selects n + 1 base (quasi) points for interpolation.

The microcomputer 11 determines the operating point 2 when it selects a base (quasi) point 22 suitable for interpolation.
It can also be programmed to check if 1 is within the envelope of the single convex surface formed by the origin (quasi) point 20 of interest. Small deviations up to a predetermined degree are then still tolerated.

FIG. 4 shows this criterion.
There is shown an operating point 21 surrounded by a reference (quasi) point 20 of interest. For a one-dimensional characteristic map, or characteristic curve, the operating point 21 lies between two selected base (quasi) points 20. This is shown in Figure 4a. For a two-dimensional characteristic map, the operating point 21 lies within the triangle formed by the selected base (quasi) point 20 (FIG. 4b). In the three-dimensional characteristic map, four base (quasi) points 20 are selected for interpolation. The operating point 21 must then be in the pyramid formed by the selected base (quasi) point 20. This is shown in Figure 4.
In c.

The microcomputer 11 of the control unit 10 operates at the operating point 2 for the selection of an appropriate base (quasi) point 22.
It can also be programmed to check the criteria for whether the (spatial) angles fixed by one and two base (quasi) points 20 are as large as possible. Here too, small deviations up to a predetermined degree are still acceptable. Suitable base (quasi) point 22
Another criterion for the selection of ## EQU1 ## arises from the fact that the base (quasi) points for the respective address components X ₁ , X ₂ are above and below the operating point 21. This criterion (Cartesian envelope) can be inspected very easily by the microcomputer 11. For this purpose, the microcomputer simply has to compare the address component (X ₁ , X ₂ ) of the selected base (quasi) point 20 with the address component of the operating point 21.

FIG. 5 shows the division of the three-dimensional address space into subregions which are subdivided to different degrees. The line L ₁ divides the address space into four partial areas B0 to B3. Partial area B
0 is also divided by line L ₂ into four subregions. The partial area B1 is divided into two parts by a line L ₃ . The partial area B2 is divided into two partial areas by the line L ₄ . The partial area B3 is not further divided. The base (quasi) point is shown in each subregion. This subdivision of the address space allows the basal (quasi) point density to meet accuracy requirements. In a characteristic map area having a very large dynamic characteristic, that is, in an area where the information content is high due to a significant change in the stored function gradient, the address space division is realized very finely, and in a characteristic map area having a small dynamic characteristic, the address space division is coarse. Only realized.

The partitioning of the address space as shown in FIG. 5 is shown in FIG. 6 in the form of a tree structure. Without partitioning, the entire address space would be represented by only one base (quasi) point. In that case, only the trunk S of the tree will be shown. After the first division of the address space by the line L ₁ , four partial areas B 0
-B3 occurs. The partial area B0 is further divided into four partial areas, so that four branches are derived from the partial area B0. The partial area B1 is divided into only two partial areas, so that from this only two branches are derived.
At this time, the division by L _{3 is performed} by the partial area U on one branch.
This is done so that B0 and UB2 are brought together and, in the other branch, subregions UB1 and UB3 are brought together. The partial area B2 of the first division stage is further divided into two partial areas by the line L ₄ . At that time, the line L ₄ is
In one branch the partial areas UB0 and UB1 are combined and in the other branch the partial areas UB2 and UB3
Are drawn to be put together. The partial area B3 of the first division stage is not further divided.

In the memory 12 of the control device 10, position information regarding the tree structure is stored for each base (quasi) point. Table 1 shows the individual base (quasi) points S0.
Position information using a binary number is shown for S8 to S8. Each binary number consists of two 2-digit binary numbers. 4 per division step, using 2 digit binary numbers
The decimal numbers 0 to 3 corresponding to the four possible areas are displayed. A 2-digit binary number is required for each division stage. The letter d indicates that writing in the corresponding binary digits may be arbitrary. In general, the following holds:
The order of the tree used depends on the characteristic map dimension. A binary tree can be used for the characteristic curves, a quat tree for the three-dimensional characteristic map, an octree for the three-dimensional, and a degree 2 ⁿ tree for the n-dimensional. Binary digit encoding can be used for the position indication inside the binary tree, and binary encoding in the two address directions for the quattree, respectively. The same applies to the following.

[0035]

[Table 1]

FIG. 7 shows another division of the address space of the three-dimensional characteristic map. Address space is line L
It is divided by 5 into 6 equal-sized rectangles 23. A segment in the memory 12 of the control device 10 is assigned to each area of the address space. The division of the address space into rectangles 23 (in the most general case a hypercube) corresponds to a coarse patterning of the address space.
The operating point 21 is also shown in FIG. When searching for a base (quasi) point 22 suitable for interpolation, the computer 11 first
It is determined whether the operating point 21 exists in the rectangle 23.
Then the base (quasi) points 20 contained therein are searched unstructured. At that time, distance calculation is also performed. Moreover, the same selection criteria as described in FIGS. 3-4 can be used. This form of address space partitioning, as shown in FIG. 7, is effective primarily for fixed coarse patterning, where the origin (quasi) point density is relatively constant in the overall address space. .

FIG. 8 shows another address space division of the three-dimensional characteristic map. The address space is again divided into rectangular shapes (rectangular solid shapes). Unlike FIG. 7, the rectangles are not all the same size here. The size of the rectangle should be such that as many base (quasi) points as possible are realized in each rectangle, ie SSG / SBG.
(SSG = total number of base (quasi) points, SBG = number of regions). This produces a rough peripheral structure. This division of the origin (quasi) point 20 in the address space is constantly modified by the continuous characteristic map matching during operation of the control unit 10. This is used to take into account changed operating conditions (eg aging phenomena) in the characteristic map. To achieve a uniform division of the base (quasi) points into rectangles, the division of the address space into rectangles is also
Must always be done using line L ₆ . In this case, the dynamic determination of the area boundaries is carried out separately for each address component. This means that the area boundaries of the respective address components are
Segment a coarse region with AK radix (quasi) points (AK
= The number of address classes of the address component under consideration). Boundaries G ₁ (X ₁ ) and G
₂ (X ₁ ) makes the entire address space 1
9/3 = 6 or in some cases it is divided into three regions containing seven base (quasi) points 20. The number of areas can be determined according to the number of base (quasi) points to be searched for read access. The number of address classes AKi of n address components can be of different sizes when the number of segments SEG ≠ AK ⁿ should be selected (eg 3 × 3 as shown in FIG. 8). Instead, SEG = 2 × 3 = 6).

Next, software capable of performing area division and access to the base (quasi) point 20 by the microcomputer 11 will be introduced and described with reference to FIG. Boundaries G _k (X _i ), (i = 1, ..., N), (k =
1, ..., M) are stored in separate tables and supply for each address component Xi a region number, in the illustrated embodiment a number between 1 and 3. N region numbers are required for each region (here 2). Two
One area number is used to read the starting address of the address space segment in which the operating point 21 exists from the memory 12. For this purpose, the memory 12 stores a two-dimensional matrix including the start address of the address space area. In FIG. 9, operating point 21 is at segment SEG _A. For this case, the following access is performed from the microcomputer: address component X ₁
Is larger than the boundary G ₁ (X ₁ ), but smaller than G ₂ (X ₁ ), that is, the area number 2 is obtained at X ₁ . In X ₂ , the area number 3 is obtained by the same method. Therefore, the matrix of the segment start address is accessed by this combination (2, 3). It is organized in three columns, each having three writing locations, which are sequentially stored in the memory of the controller 10, ie the element: start address + (2-1) × 3 + 3 is accessed. The start address of the memory segment to be searched is written here. If the individual areas are directly contiguous, i.e. should be stored without "blank" for new writes, the end address can be determined from "start address of next block-1". But if you want to leave room for new writes, it is sufficient to search to the last occupied cell (end mark), or the segment length is clearly indicated at the beginning of each segment. To be done. For time-critical applications, it is desirable to write the new base (quasi) point 20 first in the buffer region in directly contiguous segments. In this way, multiple new writes can be classified in the characteristic map at the same time and only one reclassification is required for this. The previous base (quasi) point 20 to be erased at the same time as the new write is also advantageously erased only during the classification of the new base (quasi) point 20. If a full update should be maintained before sorting, then the buffer region must be searched in addition to the current segment on each read access.

Upon writing a new base (quasi) point and simultaneously erasing the previous base (quasi) point, the number of base (quasi) points in the subregion defined by G _k (X _i ) is generally changed. . For this reason, G _k (X _i ) should be newly calculated after a certain number of characteristic map matches. For this purpose, the lengths of the segments (marked at the segment head) belonging to the partial area under consideration are added and the target value SS
It is compared with S / AK (X _i ). In the case of a deviation of + Δ, the Δ of the base (quasi) point 20 closest to the boundary under consideration is added to the corresponding region above this boundary and this boundary is newly established. This checking and possibly matching is carried out for all subregions, starting from the region of the lowest value of the address component under consideration in each case. The described method produces a classification in which one region contains a maximum of SSG / AK radix (quasi) points in the case of an inconvenient, eg strictly diagonal, radix (quasi) point distribution. In the case of a relatively uniform base (quasi) point distribution, the maximum number of base (quasi) points approaches the ideal value SSG / SEG.

Not enough information is covered between the operating point and the edge of the segment, ie in this direction the reference (quasi)
If no points are detected, adjacent segments are searched together to avoid extrapolation. In order to identify the relatively strong extrapolation, it is checked whether there is a base (quasi) point 20 for each address component (xi) above and below the operating point 21. In the main extrapolation direction, if extrapolation is detected and no suitable base (quasi) point 20 is detected in the extrapolation direction in the region at that time (this can occur especially at the region edge). Adjacent regions are searched together. In case of doubt, a region with a relatively small number of base (quasi) points is selected. This reduces the search time.

In the case of radical (quasi) point distribution, which is extremely inconvenient in address space division, a vacant area may occur. This leads to read access, which is very computationally expensive. The reason is that a plurality of adjacent regions must be searched for an appropriate interpolation base (quasi) point 22. In order to prevent this, it is advantageous to have a base (quasi) around the bordering area.
An artificial (auxiliary) basis (quasi) point, calculated from the point 20, can be written in the region center. Furthermore, by this means it is realized that a minimum base (quasi) point density is always guaranteed, which is dependent on local accuracy requirements. If a relatively high base (quasi) point density is desired, the minimum number of base (quasi) points 20 per region can be correspondingly increased. In this case, local accuracy requirements are automatically taken into account.
This is because the segmentation results in a linearization of the address quantity for (local) information content.

FIG. 10 shows three different areas 25.
Another division of the address space of the dimensional characteristic map is shown. The essential difference from the previous examples is that the region 25 is not divided into a rectangular parallelepiped shape. The entire set of all base (quasi) points 20 of the characteristic map is first divided into a group of SSG / SEG adjacent base (quasi) points, respectively. For this reason, it is desirable to use the standard base (quasi) point distribution for this application example. The base (quasi) points 20 of one group are each associated with one representative point. The representative point consists of n address components that define its position in the entrance space and a pointer to the beginning of the associated base (quasi) point list in memory 12 of controller 10. Its position in the address space is determined by the average value of the address components of all the base (quasi) points 20 belonging to it (center of gravity). When performing continuous characteristic map matching,
The new base (quasi) point 20 is processed by the weight 1 / K,
On the other hand, the previous position is weighted by (K-1) / K (K = basic (quasi) score of this class). Thereby, the position of the representative point 24 is matched with the changed base (quasi) point distribution.
During a read access, the distance of the operating point 21 to all the representative points 24 is formed. The nearest representative point is selected and the belonging base (quasi) point group is searched.

When the base (quasi) point 20 is newly written and the previous base (quasi) point is erased, the new base (quasi) point 20 is rewritten to the representative point 24 after a plurality of new writes. A new distribution and a new calculation of the position of the representative point 24 has to be performed.

Unlike the division of the address space into rectangular parallelepipeds, the coarse access to the region at that time is not performed here via address calculation, it is searched unstructured in two stages. To be done. As a result, the area 25
The flexibility of this method with respect to the placement and shape of the is increased. The above classification method reduces the search cost to 2 × (base (quasi) score) ^1/2 . For example 100
10 base (quasi) points, 10 per 10 base (quasi) points
Each representative point corresponds to 20 search steps. Figure 10
Shows the correspondence of the base (quasi) point to the representative point by a line segment.

The methods for address space partitioning that have been described thus far produce a coarse peripheral structure depending on the classification of the address space. By subdividing the peripheral structure down to the base (quasi) point level, the periphery of the operating point 21 can be directly accessed. For this purpose, the base (quasi) point is connected to the adjacent base (quasi) point selected via a pointer, for example. This is illustrated in FIG. The arrow in FIG. 11 indicates the connection of the base (quasi) point 20 to the adjacent base (quasi) point. In addition, the dashed line illustrates how the operating point 21 slowly moves through the address space of the characteristic map. This method makes use of the additional information that the rate of change of the operating point is limited in terms of access time intervals in the underlying technical problem setting. Therefore, it can be assumed that the operating point 21 is still in the vicinity of the base (quasi) point 20 used for interpolation in the previous characteristic map access in the current characteristic map access. . Thus, the concatenation of adjacent base (quasi) points 20 via pointers allows quick access to the appropriate interpolated base (quasi) points 12 without having to search for a relatively large address area. Initially, the minimum required number of pointers can be determined.
To ensure full coverage around the base (quasi) point, n + 1 independent linear pointers are sufficient in the n-dimensional characteristic map. The number of pointers at all base (quasi) points is chosen to be the same size in order to allow an efficient implementation and to guarantee a small storage requirement. Therefore,
All base (quasi) points 2 that are optimal for the interpolation method used
A concatenation of 0 is calculated. The selection of an appropriate adjacent base (quasi) point is performed according to the same criteria as the selection of the interpolation base (quasi) point surrounding an arbitrary operating point. On the first access to the characteristic map, the peripheral (quasi) point of the operating point 21 is not yet known. A suitable interpolation base (quasi) point can be detected via a fixed starting point, eg when the operating point always starts from the characteristic map origin, or once an unstructured search is performed. Thereafter, direct access to the adjacent base (quasi) point is always possible via the pointer. When a correction of the characteristic map is required, i.e. a new radical (quasi) point 20 has to be written and an earlier less relevant radical (quasi) point 20 has to be erased, the connections forming adjacencies in this region also. Must be changed. It saves the calculation time and the new writing of the base (quasi) point 20 causes the adjoining previous base (quasi) point 20 for connection.
A plurality of correction values can be assembled and commonly written in order to be able to be almost compensated by the elimination of For this purpose, new base points are initially stored in a buffer or they are written provisionally, i.e. in the characteristic map, without a request for optimal concatenation. After K new writes, the concatenation of the characteristic maps is checked globally. This calculation is done offline in the background program.

Automatic characteristic map generation: The base (quasi) point, that is, its position and, in some cases, its function value, and division into address space areas or setting of representative point positions, must be done manually or automatically. You can

If a predetermined number of basal (quasi) points should be automatically and optimally positioned in the characteristic map, the basal (quasi) point position and the function value are used as variables in the optimization program. It can be given in advance. Therefore, the optimization method uses the position and the function value, for example, of the interpolated characteristic map surface,
Change until the deviation from the predetermined measurement point amount (target characteristic map) is minimized. This deviation is determined, for example, for all measurement points by forming the distances of the measurement points from the values interpolated from the characteristic map at this point.

A predetermined function (measurement point amount) is used as small as possible with a predetermined maximum error (quasi).
If it is to be simulated through points (for information content), multiple optimization steps can be carried out as described above, with the base (quasi) score up to the first error point below. For a while.

In order to automatically divide the address space into areas by using the representative points, it is only necessary to optimally position the representative points. This can be done by giving the representative point position as a variable to the optimization program in advance. The optimization program changes its position, for example, until the average distance of all the base (quasi) points to their nearest neighbors is minimized.

Due to the area division of the address space, for a given application, in the area where the operating point occurs very frequently, a smaller number of base (quasi) points are written than in the area where the operating point becomes complicated. And is advantageous. Since this minimizes the average search time, ie the access time.

The method described here is applicable to a wide variety of fields. It is particularly suitable for use in mass-produced control devices. In addition to the above-mentioned applications, traveling device adjustment, diesel adjustment, and rear shaft steering can also be mentioned. The information read out from the characteristic map can then also be used to determine the actuation quantities that are important for controlling or adjusting the process but are not measured.

[Brief description of drawings]

FIG. 1 is a schematic diagram of a control device for an internal combustion engine.

FIG. 2 is a graph in which the base (quasi) point distribution according to the present invention is performed 3
It is a schematic diagram of a dimensional characteristic map.

FIG. 3 is a schematic diagram showing a first embodiment for division of an address space of a characteristic map showing selection of interpolation base (quasi) points in a three-dimensional characteristic map.

FIG. 4 is a schematic diagram showing operating points together with selected base (quasi) points of a characteristic map for one, two and three dimensions.

FIG. 5 is a schematic diagram showing a second embodiment for dividing the address space of a three-dimensional characteristic map.

6 is a diagram showing a tree structure of the address space division shown in FIG.

FIG. 7 is a schematic diagram showing a third embodiment for dividing the address space of a three-dimensional characteristic map.

FIG. 8 is a schematic diagram showing a fourth embodiment for dividing the address space of a three-dimensional characteristic map.

9 is a schematic diagram illustrating access to a base (quasi) point of the characteristic map in which the address space division of FIG. 8 is performed.

FIG. 10 is a schematic diagram showing a fifth embodiment for dividing the address space of a three-dimensional characteristic map.

FIG. 11 is a schematic diagram showing a sixth embodiment for dividing the address space of a three-dimensional characteristic map in which base (quasi) points are connected to each other.

[Explanation of symbols]

10 control devices, 12 memories, 20 units (quasi)
Points, 21 operating points

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Henning Torre Rosdorf Heinrich-Heine-Strasse 8 (56) References JP-A-4-303178 (JP, A) JP-A-2-306364 (JP, A) Japanese Unexamined Patent Publication No. 1-23297 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G05B 11/36 G05B 13/02

Claims (32)

(57) [Claims] 1. A method for controlling a process operation amount by a control device using a characteristic map, wherein the characteristic map is defined by the process operation amount and stored in a memory of the control device. Represented by a number of base (quasi) points, and for each said base (quasi) point at least one valid information relating to the value of said characteristic map at the position of the base (quasi) point in the address space of said characteristic map. At least one base (quasi) point, which is stored in the memory of the device and is close to the operating point of the process, is read from the characteristic map, and the base (quasi) point is read. In view of the above, in a method of the type in which at least one controlled variable is formed for sending the control signal, the address space of the characteristic map is additionally provided for each of the base (quasi) points (20). The stored group (quasi) points at least one of information relating to the position of (20) in the memory (12) of the control device (10) of said group (quasi) point (20),
At least one base (quasi) point that can be arranged at any position in the address space of the characteristic map and is as close as possible to the operating point (21) regardless of the fixed pattern of the operation amount is selected. In order to do so, a distance calculation is performed to calculate the distance from at least one base (quasi) point of the characteristic map to the operating point (21), and the process operation amount control method. 2. The operating amount of the process according to claim 1, wherein the address of the base (quasi) point (20) and the valid information are stored with an equivalent access right in order to enable reverse access to the characteristic map. Control method. 3. The control device (1) for an individual base (quasi) point (20) or a group of adjacent base (quasi) points (20).
3. The process operation amount control method according to claim 1, wherein peripheral information is stored in the memory (12) of 0). 4. A peripheral (quasi) point (2) is used as peripheral information.
4. The process operation amount control method according to claim 3, wherein a pointer to an adjacent base (quasi) point (20) for 0) is stored in the memory (21) of the controller. 5. The characteristic map address space is divided into a plurality of areas, and a predetermined segment in the memory (12) of the control device (10) is assigned to each area and a base (quasi-area) of each area is assigned. 5. A method according to claim 1, wherein the points (20) are stored in corresponding memory segments. 6. The characteristic map of the process is uniformly distributed around the operating point near the operating point (21) to determine a characteristic map value of the operating point (21). 6. A method according to any one of claims 1 to 5, characterized in that an intermediate value calculation, preferably an interpolation, is carried out from the characteristic map values of a certain number of base (quasi) points (20). 7. A base (quasi) point (20) uniformly distributed around the operation point (21) in the vicinity of the operation point (21), wherein the operation point (21) is the base (quasi) point. 7. The process operation amount control method according to claim 6, wherein the process quantity is selected so that it is within a simplex defined by. 8. A base (quasi) point (20), which is located near the operating point (21) and uniformly distributed around the operating point, is defined as the operating point (21) and each time. 7. The process actuation control method according to claim 6, wherein the angles defined by the two base (quasi) points (20) are selected to be as large as possible. 9. A base (quasi) point (20) uniformly distributed around the operation point (21) in the vicinity of the operation point (21) is used for each address component. Two
7. The method of claim 6, wherein 0) is above and below the operating point (21). 10. A base (quasi) point (20) uniformly distributed around the operating point in the vicinity of the operating point (21),
The operating point (21) is the selected base (quasi) point (2
7. The method for controlling the operation amount of a process according to claim 6, wherein the selection is performed so that the center of gravity is within the center of gravity of 0). 11. The process operation amount control method according to claim ₁ , wherein an L ₁ norm is formed in the distance calculation. 12. The method according to claim 6, wherein n + 1 number of base (quasi) points (20) are used for calculating the intermediate value in the n-dimensional characteristic map. . 13. A base (quasi) point (20) of the characteristic map.
13. The process operation amount control method according to claim 5, wherein the access to the process is performed in multiple stages, for example, in two stages. 14. A base (quasi) point (20) of the characteristic map.
In accessing the region, in the first stage, a region containing a desired base (quasi) point (20) is obtained, and the second region is obtained.
14. In the step of, the segment belonging to the obtained area is searched for an appropriate base (quasi) point (20) in an unstructured state in the memory (12) of the controller (10). Control method for operating amount of process. 15. The method according to claim 14, wherein a predetermined address calculation is carried out depending on the operating point (21) in order to determine the area containing the desired base (quasi) point (20). A method for controlling the operating amount of a process. 16. The address space area is divided into a rectangular parallelepiped shape, and the address space is divided for each address component (x1, x2) separately. A method for controlling the operating amount of a process. 17. The area division is performed by the base (quasi) point (2).
17. The process operation amount control method according to claim 5, wherein the number of 0) is selected so as to be as large as possible in each region. 18. Each address component (x1, x
18. The operating amount of a process according to claim 15, wherein the region boundaries between the individual address space regions for 2) are stored in a separate table in the memory (12) of the control device (10). Control method. 19. A region boundary is manually or automatically added to each address component (x1, x2) so that the entire set (20) of base (quasi) points is the address component (x).
, X2) is divided by a first number of region boundaries into a second number of adjoining basal (quasi) points (20) of as large a subset quantity as possible, where Two
17. The method according to claim 5, wherein the number corresponds to a number obtained by increasing the first number by one. 20. The start address of a segment is stored in the form of a matrix in the memory (12) of the control device (10).
A method for controlling the operation amount of the process according to the item. 21. The base (quasi) points (20) of one region are respectively associated with one representative point (24), and the representative points are also written in the memory (12) of the control device (10). However, the representative point (24) is relative to the address component (x1, x2) at that position in the characteristic map and the start of the corresponding memory segment in which the base (quasi) point (20) of the area is stored. 15. The process operation amount control method according to claim 5, further comprising a pointer. 22. The process operation amount control method according to claim 21, wherein a reference to an adjacent representative point (24) is assigned to the representative point (24) as peripheral information. 23. All the bases (quasi) of the characteristic map, which are located closer to each representative point (24) than any other representative point (24) with respect to the representative point (24).
23. A method as claimed in claim 21 or 22, wherein points (20) are assigned. 24. The position of each representative point (24)
All the base (quasi) points assigned to it (20)
24. The process operation amount control method according to any one of claims 21 to 23, wherein the method is determined by an average value of address components (x1, x2) of. 25. In order to obtain a region containing a desired base (quasi) point (20), an unstructured search is performed by using a representative point (2) located closest to the characteristic map region at that time.
25. The method for controlling an operation amount of a process according to any one of claims 21 to 24, which is carried out so as to be determined by obtaining 4). 26. In order to select a base (quasi) point (20) close to the operating point (21), the base (quasi) point (20) of the characteristic map is searched in an unstructured state. 5. A method for controlling an operation amount of a process according to claim 1, wherein 27. The method according to claim 1, wherein the characteristic map is manually or automatically matched to an unchanged process characteristic.
An operation amount control method for a process according to any one of items 1 to 7. 28. A method according to claim 27, wherein a new base point (20) is written in the memory (12) of the controller (10) to match the characteristic map. 29. The process operation amount control method according to claim 28, wherein in order to satisfy a predetermined required storage capacity, the preceding base (quasi) point (20) is similarly deleted. 30. A method according to claim 28 or 29, wherein the segmentation of the characteristic map is finally carried out after a predetermined number of new writes. 31. The region segmentation having a low priority,
The process operation amount control method according to claim 30, which is implemented in a background program of the control device (10). 32. The base (quasi) point (2) of the characteristic map.
0) and to select at least one base (quasi) point (20) close to said operating point (21),
Method according to claim 1 or 2, wherein a content addressable memory (12) is used.