Method for operating a drive unit for a motor vehicle and corresponding drive unit
By determining and adjusting reference input vectors based on individual errors, the method optimizes drive device operation in motor vehicles, enhancing model accuracy and adaptability with minimal computational effort.
Patent Information
- Application Number
- DE102020111206
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-04-24
- Publication Date
- 2024-12-05
- Estimated Expiration
- 2040-04-24
AI Technical Summary
Existing methods for operating drive devices in motor vehicles require significant computational effort for determining output vectors and lack the ability to adapt models during operation, leading to inefficiencies and suboptimal performance.
A method that involves determining individual errors of reference input vectors, removing vectors with the smallest errors, calculating output vectors using new input vectors, and adjusting the reference input vector set to improve model accuracy with minimal computational effort, allowing adaptation during operation.
This approach optimizes reference input vectors to enhance model accuracy with reduced computational requirements, enabling efficient and adaptive model improvements during vehicle operation.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
[0001] The invention relates to a method for operating a drive device for a motor vehicle, which has a control unit and a drive unit and is controlled by means of the control unit, wherein n-dimensional reference input vectors are each assigned a reference output vector, which are stored in the control unit and from which an output vector can be determined for an operating parameter for the drive unit of the motor vehicle, wherein when a new reference input vector with an associated new reference output vector is present, the following steps are carried out, in particular in the specified order: a. Determining a single error of the reference input vectors that describes the accuracy with which the reference output vector of one of the reference input vectors can be determined from the other reference input vectors without resorting to the one reference input vector and its reference output vector; b. buffering the reference input vector with the smallest single error and removing this reference input vector from the reference input vectors; c. Calculate the output vector using the new reference input vector as the input vector; d. Determining the individual error from a difference between the output vector and the reference output vector associated with the new reference input vector; e. Adding the removed reference input vector to the reference input vectors; f. Replacing the reference input vector with the smallest single error by the new reference input vector if the single error of the new reference input vector is larger than the smallest single error.
[0002] The invention further relates to a corresponding drive device.
[0003] For example, the prior art document DE 10 2010 028 266 A1 is known. This document describes a control unit in a vehicle that has means for calculating at least one output variable for controlling vehicle functions during vehicle operation based on at least one input variable determined during operation. The control unit has means for calculating the output variables using a Bayesian regression of training values determined for the output variable and the input variable prior to operation.
[0004] The document DE 43 04 441 A1 describes a method for operating a technical process, in particular a process in a motor vehicle, with the aid of a control unit. In the control unit, parameters for controlling the process are determined from a characteristic map. The characteristic map is spanned by operating variables of the process. The characteristic map is represented by a number of support points stored in the memory of the control unit. For each support point, at least one piece of address information about the position of the support point in the characteristic map and one piece of useful information about the value of the characteristic map at the position of the support point are stored in the memory of the control unit. The support points can thus be placed at any point in the characteristic map, independent of a fixed grid of operating variables. A distance calculation is performed to select a suitable support point for an operating point of the process.
[0005] Also known from the prior art are the publication DE 44 18 731 A1 and the non-patent literature SCHMITT, M.; ULLRICH, T.; TOLLE, H.: Associative datafields in automotive control. Proceedings of IEEE International Conference on Control and Applications, Vol. 2, 1994, pp. 1239-1244. IEEE Xplore [online]. DOI: 10.1109 / CCA.1994.381339, In: IEEE.
[0006] It is an object of the invention to propose a method for operating a drive device for a motor vehicle, which has advantages over known methods, in particular allows an improvement or at least a change of the model during operation of the drive device and / or enables a determination of an output vector for an input vector from a model with low computational effort.
[0007] This is achieved according to the invention with a method for operating a drive device for a motor vehicle having the features of claim 1. It is provided that the following steps are carried out in the specified order to optimize the reference input vectors: a. Determine the individual errors of the reference input vectors; b. Calculating an original total error from the individual errors; c. Perform the following steps for each of the reference input vectors: i. Removing the respective reference input vector from the reference input vectors, ii. Determine the individual error of each of the remaining reference input vectors, iii. Calculating a total reduction error of the removed reference input vectors from a sum of the individual errors of the remaining reference input vectors, iv. Assigning the reduction total error to the remote reference input vector; v. adding the removed reference input vector to the reference input vectors; d. Removing the reference input vector with the smallest reduction total error and / or with a reduction total error that is less than the original total error from the reference input vectors.
[0008] The control unit is used to control the motor vehicle, in particular to control a drive device of the motor vehicle. The drive device is used to drive the motor vehicle, i.e., to provide a drive torque directed toward driving the motor vehicle. In particular, the control unit is used to control a drive unit of the drive device, which ultimately generates the drive torque. The drive unit is, for example, in the form of an internal combustion engine, in particular a gasoline engine or a diesel engine.
[0009] The control unit is optionally used, at least temporarily, to determine the output vector matching the input vector using a model. In the case of an internal combustion engine, for example, at least one of the following values serves as the input vector: a rotational speed and an injection parameter, in particular a fuel quantity and / or an injection timing. Particularly preferably, the input vector is composed of several of the aforementioned values. An ambient condition, for example, an ambient temperature and / or an ambient air pressure, can also be used as an input vector or as a component of the input vector.
[0010] The input vector is available as an n-dimensional input vector. This means that it can be of any dimension. For example, the input vector is one-dimensional, i.e. scalar. However, the input vector is preferably multi-dimensional, for example two-dimensional or three-dimensional. The model is used to assign the output vector to the input vector. The output vector includes, for example, at least one of the following values: torque, fuel consumption, pollutant emissions. For example, the output vector is only directed at a single value, i.e., it is scalar. Of course, the output vector can also include multiple values. It can be provided that the input vector is filtered, in particular over time. This avoids jumps in the output vector and / or models a dynamic course of the output vector for the input vector.
[0011] To determine the output vector from the input vector, n-dimensional reference input vectors are stored in the control unit. Reference input vectors are generally understood to be a set of reference input vectors that can have any number of reference input vectors. The reference input vectors can therefore comprise no reference input vectors at all, exactly one reference input vector, or multiple reference input vectors. For the sake of readability, however, reference input vectors are always referred to in the plural, always referring to a reference input vector set with any number of reference input vectors. Incidentally, this also preferably applies to the other vectors, even without this being explicitly stated in each case.
[0012] Each reference input vector is assigned a reference output vector. The reference input vectors are each designed analogously to the input vector, and the reference output vector is designed analogously to the output vector. In particular, the reference input vectors each have the same dimension as the input vector and include the same variables or values. The reference output vector is also designed analogously to the output vector, i.e., has the same dimension and relates to the same variable or value. The reference output vector can therefore only have a scalar. In principle, any number of reference input vectors and associated reference output vectors can be stored in the control unit, for example, no reference input vector at all, just exactly one reference input vector, or multiple reference input vectors.For this purpose, the control unit has, for example, a corresponding memory.
[0013] It should be noted that if at least one vector is mentioned in this description, only exactly one vector can be present. However, multiple vectors are preferably present, and the statements regarding the vector or the at least one vector can be applied analogously to each of the multiple vectors. Conversely, if multiple vectors are mentioned, only a single vector or at least one of the mentioned vectors can of course be present. This applies in particular, but not exclusively, to the at least one reference input vector, the at least one neighboring vector, the at least one calculation vector, and / or the at least one classification vector.
[0014] The described method has the primary advantage that the database, i.e., the reference input vectors and the reference output vectors from which the output vector is determined, is adaptable. This adaptation can be performed not only before commissioning the control unit, but also during operation of the control unit, in particular during operation of the drive system or during operation of the motor vehicle.
[0015] For example, the new reference output vector corresponding to the new reference input vector is measured by a sensor on the drive unit after setting the reference input vector.
[0016] To improve the model using this measurement, it is then checked whether the new reference input vector and the new reference output vector contribute to the accuracy of the model. The new reference input vector and the corresponding new reference output vector are determined, for example, by setting the new reference input vector, particularly on the drive device or drive unit, and measuring the resulting reference output vector or determining it in another way.
[0017] First, the individual errors of the reference input vectors are calculated and assigned to the individual reference input vectors. Subsequently, for at least one of the reference input vectors, a check is carried out to determine whether the accuracy of the model can be improved by replacing the reference input vector with the new reference input vector. Particularly preferably, the procedure is carried out at least for the reference input vector with the smallest individual error. To this end, the at least one reference input vector, in particular the reference input vector with the smallest individual error, is first removed from the reference input vectors.
[0018] The new reference input vector is then used as the input vector, and the output vector is determined for it using the remaining reference input vectors. The single error of the new reference input vector is then determined from the output vector. For this purpose, the difference between the output vector and the reference output vector assigned to the new reference input vector is determined. This difference is temporarily stored as a single error.
[0019] The removed and temporarily stored reference input vector is then added back to the reference input vectors. Finally, after the described steps, the same reference input vectors are available as before. The single error of the new reference input vector is then compared with the smallest single error, i.e., the single error of the temporarily stored reference input vector. If the single error of the new reference input vector is greater than the smallest single error, the reference input vector to which the smallest single error is assigned is replaced by the new reference input vector. The reference input vectors are then preferably saved, i.e., stored as stored reference input vectors.
[0020] This represents a particularly simple and, in terms of computational effort, advantageous approach for optimizing the reference input vectors. Finally, based on the individual errors, it is assessed whether an improvement in accuracy can be achieved by replacing the least-needed reference input vector with the new reference input vector. Reference is also made to the further explanations in this description, in particular to the following explanations.
[0021] A further development of the invention provides that, in order to determine individual errors of the reference input vectors, the following steps are carried out for each of the reference input vectors, in particular in the specified order: a. Buffering the respective reference input vector as a test vector and removing the reference input vector from the reference input vectors, b. Calculate the output vector using the test vector as input vector, c. Determining the respective individual error from a difference between the output vector and the reference output vector assigned to the test vector, d. Assigning the respective individual error to the test vector, and e. Adding the test vector to the reference input vectors.
[0022] From time to time, and especially in the context of the optimization of the reference input vectors described above, it may be useful to determine an individual error for each of the reference input vectors. This individual error ultimately describes the accuracy with which the reference output vector of one of the reference input vectors can be determined from the other reference input vectors and their associated reference output vectors, without resorting to the one reference input vector and its reference output vector.
[0023] To do this, the reference input vector is temporarily stored as a test vector, for example, in a corresponding buffer. The reference input vector or test vector is then removed from the reference input vectors. The test vector is then used as the input vector, and the output vector is calculated from the remaining reference input vectors for the input vector. The individual error is then determined by calculating the difference between the output vector and the reference output vector assigned to the test vector.
[0024] The individual error is then assigned to the test vector and saved together with it. Finally, the test vector is added back to the difference vectors, so that the same reference input vectors are subsequently available as before the individual error was determined for the reference input vector. The described procedure is preferably performed for each of the reference input vectors. It is then possible to assess whether all reference input vectors are necessary to achieve sufficiently high accuracy in determining the output vector, or whether the model's accuracy can be improved using the new reference input vector.
[0025] A further development of the invention provides that the difference is sign-corrected to determine the individual error. When determining the individual error, the difference between the output vector and the reference output vector assigned to the test vector is calculated. The individual error now corresponds to the sign-corrected difference to achieve high significance. For example, it is provided to square the difference for sign-correcting. In this case, the individual error corresponds to the square root of the squared difference. This procedure represents an extremely precise and reliable method for determining the individual error.
[0026] The invention provides that, in order to optimize the reference input vectors, at least one of the following steps, in particular all of the following steps, are carried out, preferably in the specified order: a. Determine the individual errors of the reference input vectors; b. Calculating an original total error from the individual errors, in particular divided by the number of reference input vectors, preferably from squared individual errors, c. Perform the following steps, particularly in the order specified, for each of the reference input vectors: i. Removing the respective reference input vector from the reference input vectors, ii. Determine the individual error of each of the remaining reference input vectors, iii. Calculating a total reduction error of the removed reference input vector from the individual errors of the remaining reference input vectors, in particular divided by the number of reference input vectors, preferably from squared individual errors, iv. Assigning the reduction total error to the remote reference input vector; v. adding the removed reference input vector to the reference input vectors; d. Removing the reference input vector with the smallest reduction total error and / or with a reduction total error that is less than the original total error from the reference input vectors.
[0027] First, the individual errors of the reference input vectors are calculated in the manner described above and assigned to the individual reference input vectors. Subsequently, for example, to optimize the reference input vectors, the reference input vector with the smallest individual error can be removed from the reference input vectors and the optimization can then be terminated. However, the original total error is preferably calculated from the individual errors. For this purpose, the individual errors are summed, for example. Particularly preferably, the original total error is equal to the sum of the individual errors of all reference input vectors divided by the number of reference input vectors. It can also be provided that the individual errors are squared before summing.In this case, the original total error preferably corresponds to the square root of the summed squared individual errors divided by the number of reference input vectors.
[0028] The total reduction error is then calculated for each of the reference input vectors. This is done using a procedure similar to the procedure for determining the individual error. First, the respective reference input vector is removed from the reference input vectors, and the individual error of each of the remaining reference input vectors is calculated using the remaining reference input vectors. Each of the remaining reference input vectors is then assigned the individual error that would exist without the removed reference input vector.
[0029] The total reduction error is determined from the individual errors of the remaining reference input vectors. This is done analogously to the calculation of the total original error, so reference is made to the corresponding explanations. The reduction error is assigned to the removed reference input vector, which is then added back to the reference input vectors. Thus, the total reduction error, or a value for the total reduction error, is subsequently available for each of the reference input vectors.
[0030] The total reduction error describes the accuracy of the method for determining the output vector for the input vector when using only a portion of the reference input vectors, namely excluding the respective removed reference input vector. It allows a statement to be made as to whether the removed reference input vector contributes to the accuracy of the method or whether the reference input vector can be removed under certain circumstances.
[0031] It is intended to check the total reduction errors of the reference input vectors. It may be intended to determine the reference input vector with the smallest total reduction error and remove it from the reference input vectors. Because the reference input vector is associated with the smallest total reduction error, removing the reference input vector results in less accuracy loss than removing the other reference input vectors.
[0032] Additionally or alternatively, a check is performed to determine whether the reduction total error is smaller than the original total error. For example, all reference input vectors for which this condition is met are removed from the reference input vectors. However, it is particularly preferred to remove only the reference input vector with the smallest reduction total error from the reference input vectors, provided that the condition is also met that the reduction total error of this reference input vector is less than, i.e., smaller than, the original total error.
[0033] The described procedure enables a particularly effective reduction of the number of reference input vectors and, accordingly, a reduction of the memory requirements for storing the reference input vectors, in particular without an undue reduction of the accuracy of the model based on the reference input vectors.
[0034] A further development of the invention provides that, if a reference input vector is present in the reference input vectors that corresponds to the new reference input vector, the new reference output vector is determined from the reference output vector assigned to the corresponding reference input vector and the new reference output vector. For example, it is provided that the new reference output vector is set equal to the mean value of the reference output vector assigned to the corresponding reference input vector and the new reference output vector. If the new reference input vector is therefore equal to one of the already existing reference input vectors, the reference output vector assigned to this reference input vector is adjusted using the new reference output vector. The reference input vectors are then preferably stored, i.e., stored as stored reference input vectors.This results in an iterative improvement of the model.
[0035] A further development of the invention provides that the new reference output vector is determined by averaging the reference output vector and the new reference output vector. Thus, if the new reference input vector corresponds to one of the existing reference input vectors, the reference output vector of this reference input vector is adjusted using the new reference output vector. For this purpose, the mean value of both reference output vectors is calculated and assigned to the existing reference input vector. This eliminates or corrects any existing errors in the model over time.
[0036] A further development of the invention provides that the replacement of the reference input vector with the new reference input vector is only performed if the number of reference input vectors, in particular the number of stored reference input vectors, is equal to a specified maximum number. Typically, the memory space available in the control unit is limited. Therefore, it may be provided to limit the number of reference input vectors to the maximum number in order to save memory space.
[0037] As long as sufficient memory space is available, i.e., the number of reference input vectors is less than the maximum number, the new reference input vector can be added to the reference input vectors without removing any of the existing reference input vectors or replacing them with the new reference input vector. However, if the maximum number is reached by the number of reference input vectors, the new reference input vector can no longer be added without removing one of the existing reference input vectors. Accordingly, the replacement is performed, for example, in the manner described above. This achieves high memory efficiency for the control unit or the method.
[0038] A further development of the invention provides that when the new reference input vector with the associated new reference output vector is present, the following steps are carried out, in particular in the specified order: a. Determine the individual errors of the reference input vectors; b. Calculating an original total error from the individual errors, in particular divided by the number of reference input vectors, preferably from squared individual errors; c. Performing the following steps, in particular in the specified order, for one or more of the reference input vectors, in particular for the reference input vector with the smallest individual error: i. Replacing the respective reference input vector in the reference input vectors with the new reference input vector, ii. Determining the individual error of each of the reference input vectors, iii. Calculating a substitute total error of the reference input vectors from the individual errors of the reference input vectors, in particular divided by the number of reference input vectors, preferably from squared individual errors, iv. Assigning the equivalent total error to the replaced reference input vector, v. removing the new reference input vector from the reference input vectors and inserting the replaced reference input vector into the reference input vectors; d. Replacing the reference input vector with the smallest equivalent total error and / or with an equivalent total error that is less than the original total error with the new reference input vector.
[0039] The new reference input vector and the corresponding new reference output vector are determined, for example, by setting the new reference input vector, in particular on the drive device or drive unit, and measuring the resulting reference output vector or determining it in another way. First, the individual errors and the original total error are then calculated in the manner described above. Reference is made to the corresponding explanations.
[0040] Subsequently, for at least one of the reference input vectors, a check is performed to determine whether the accuracy of the model can be improved by replacing the reference input vector with the new reference input vector. Particularly preferably, the procedure is performed at least for the reference input vector with the smallest individual error. However, it can also be provided to perform the procedure for multiple reference input vectors, in particular the reference input vectors with the smallest individual errors, or for all reference input vectors.
[0041] Basically, the idea is to replace one reference input vector with the new reference input vector, i.e., to remove one reference input vector from the reference input vectors and insert the new reference input vector into the reference input vectors. The individual error is then determined for all reference input vectors, namely in the manner already described. The equivalent total error is determined from the individual errors of the reference input vectors. This is done analogously to determining the original total error, so reference is made to the corresponding explanations. The equivalent total error is assigned to the replaced reference input vector. Subsequently, the new reference input vector is removed from the reference input vectors, and the replaced reference input vector is inserted back into them. Finally, after the described steps, the same reference input vectors are available as before.
[0042] Subsequently, a check is performed to determine whether replacing the reference input vector or one of the reference input vectors with the new reference input vector offers an accuracy advantage. For example, the reference input vector with the smallest equivalent total error is replaced by the new reference input vector in the reference input vectors. Alternatively, the reference input vector whose equivalent total error is lower than the original total error is replaced by the new reference input vector. The reference input vectors are then preferably saved, i.e., stored as stored reference input vectors.
[0043] Particularly preferably, the two conditions are linked. In this respect, the reference input vector having the smallest equivalent total error is replaced by the new reference input vector, but only if the equivalent total error of the reference input vector is smaller than the original total error. In the manner described, the accuracy of the model for determining the output vector from the input vector can be significantly improved. Very particularly preferably, the described procedure for replacing the reference input vector with the new reference input vector is only carried out if a number of reference input vectors, in particular a number of stored reference input vectors, is equal to a specified maximum number. Only then is an exchange of existing reference input vectors usually necessary.
[0044] A further development of the invention provides that when the new reference input vector with the associated new reference output vector is present, the following steps are carried out, in particular in the specified order, preferably only if the number of reference input vectors is smaller than the specified maximum number: a. Determine the individual errors of the reference input vectors; b. Calculating the original total error as the sum of the individual errors, in particular divided by the number of reference input vectors, preferably from squared individual errors, c. Adding the new reference input vector to the reference input vectors; d. Re-determining the individual errors of the reference input vectors; e. Calculating a new original total error as the sum of the individual errors, in particular divided by the number of reference input vectors, preferably from squared individual errors, f. Removing the new reference input vector from the reference input vectors if the new original total error exceeds the original total error.
[0045] This procedure is used to determine whether adding the new reference input vector and the corresponding new reference output vector offers an accuracy advantage. To do this, the total original error and the new total original error are determined using the already known procedure. The total original error results from the sum of the individual errors before adding the new reference input vector and the new total original error after adding the new reference input vector.
[0046] Based on the original total error and the new original total error, a decision is made as to whether adding the new reference input vector to the reference input vectors is beneficial or not. The new reference input vector is removed from the reference input vectors if the new original total error is greater than the original total error. Otherwise, the new reference input vector is retained as part of the reference input vectors. This procedure is particularly preferred if none of the existing reference input vectors is replaced by the new reference input vector, particularly due to insufficient accuracy. The reference input vectors are then preferably saved, i.e., stored as stored reference input vectors.
[0047] For example, it is first checked whether one of the existing reference input vectors can be replaced by the new reference input vector, namely in the manner described above. If this is not the case, i.e., the corresponding conditions are not met, it is checked in the manner described above whether including the new reference input vector in the reference input vectors in addition to the existing reference input vectors offers any advantages. In this manner, a particularly high level of model accuracy can be achieved.
[0048] A further development of the invention provides that the following steps are carried out to determine the output vector for the input vector: a. Repeating the following steps, in particular in the specified order, until in an n-dimensional space, for example in one or more orthants, in particular in each orthant, of the n-dimensional space, there is a reference input vector around the input vector whose distance to the input vector is below a threshold value, and / or a maximum number of iterations is reached: i. Selecting at least one neighbor vector from the reference input vectors, and ii. Determining an additional reference input vector from the at least one neighboring vector and adding the additional reference input vector to the reference input vectors; b. Selecting at least one calculation vector closest to the input vector from the reference input vectors and calculating the output vector based on the at least one selected calculation vector.
[0049] To determine the output vector corresponding to the input vector from the reference input vectors, in a first step, at least one neighboring vector, but preferably several neighboring vectors, are selected from the reference input vectors. In a second step, the additional reference input vector is determined from this neighboring vector(s) and added to the reference input vectors. By adding the additional reference input vector to the reference input vectors, the resolution of the model is improved, so that the subsequent calculation of the output vector can be carried out extremely easily and with high accuracy.
[0050] The selection of the neighboring vector(s) from the reference input vectors is therefore preferably carried out - if possible - in such a way that the additional reference input vector determined from it(s) is closer to the input vector than the neighboring vector(s) or the reference input vectors different from the neighboring vector(s). For example, the reference input vector closest to the input vector is selected as the neighboring vector, or the reference input vectors closest to the input vector are selected as the neighboring vectors.
[0051] The additional reference input vector is preferably determined by averaging, for example by weighted averaging or by using the arithmetic mean of the neighboring vector(s). In the case of only one neighboring vector, the additional reference input vector corresponds to the neighboring vector after averaging. Naturally, the reference output vector associated with the additional reference input vector is determined analogously to this procedure from the reference output vector associated with the neighboring vector(s). The additional reference input vector is also assigned a reference output vector that has the same mathematical dimension as the reference output vectors of the reference input vectors.
[0052] The two steps, i.e. the first step and the second step, are repeated until at least one reference input vector is present whose distance from the input vector is less than the threshold value, or until a maximum number of iterations is reached. More generally, the steps are repeated, for example, until in the n-dimensional space, in particular in one or more of the orthants of the n-dimensional space, there is a reference input vector around the input vector whose distance from the input vector is less than the threshold value, or until the number of iterations, i.e. the number of times the two steps are carried out, has reached the maximum number. For each repetition of the first and second step, a number of iterations is incremented. If this number reaches the maximum number, the condition is met and the two steps are not repeated again.
[0053] The latter condition, in particular, prevents arbitrarily frequent execution if the additional reference input vector is not closer to the input vector than the original reference input vectors. It should be noted that, regardless of the number of reference input vectors and the number of neighboring vectors, the additional reference input vector is always determined, namely by averaging, and is also always added to the reference vectors. This also applies if there is no reference input vector at all or only a single reference input vector. In the second case, the additional reference input vector will usually correspond to the already existing reference input vector, so that two identical reference input vectors are subsequently available, from which the at least one calculation vector is selected.
[0054] The n-dimensional space spans around the input vector. The input vector thus represents a zero point of the n-dimensional space. The orthants are adjacent to the input vector or the zero point and do not overlap. The n-dimensional space is divided into 2 n The orthants are divided or subdividable. In the case of two-dimensional space, the orthants can be referred to as quadrants, and in the case of three-dimensional space, as octants. Particularly preferably, the above-mentioned steps are repeated until a reference input vector is arranged in several of the orthants, in particular in each of the orthants, which has a distance from the input vector that is smaller than the threshold value.
[0055] If the aforementioned condition is met or the maximum number of iterations is reached, the calculation vector(s) are selected from the reference input vectors, namely such that the calculation vector(s) are the reference input vector(s) closest to the input vector. For example, the calculation vector(s) are selected for several of the orthants. In other words, the reference input vector(s) that are closest to the input vector are selected from the reference input vectors, and the selected reference input vector(s) are used as the calculation vector(s). The selection of the calculation vector(s) from the reference input vectors is carried out, for example, such that at least some of the selected calculation vectors lie in different orthants.
[0056] Particularly preferably, the calculation vectors are selected from the reference input vectors in such a way that exactly one of the calculation vectors is located in several of the orthants. Very particularly preferably, exactly one calculation vector is selected from the reference input vectors for each of the orthants, so that exactly one calculation vector is located in each of the orthants. The output vector is then calculated from the selected calculation vector(s). This can generally be done in any desired manner. For example, it can be provided to set the output vector equal to the reference output vector of the calculation vector that is closest to the input vector. Preferably, however, an average is again calculated, for example using the arithmetic mean or inverse distance weighting.
[0057] Using the described procedure for operating the control unit, the output vector corresponding to the input vector can be determined in a special way. In particular, the computing power required for this is extremely low. It is also possible, with low computing power, to add at least one additional reference input vector and the corresponding reference output vector to the model, thus improving the model. Therefore, improving the model is preferably performed during operation of the control unit, in particular while the motor vehicle is in motion.
[0058] Particularly preferably, a value range is recorded for the reference input vectors. For at least one component of the reference input vectors, the value range contains the minimum value and the maximum value that the respective component has across all reference input vectors. Preferably, the value range is determined for all components of the reference input vectors. Preferably, the components of the reference input vectors are used at least partially in standardized form. This means that the value of each component of the reference input vectors is standardized with the difference between the maximum value and the minimum value of the respective component.
[0059] This normalized value is then used, for example, when selecting the at least one neighboring vector, in particular when calculating the distance of the reference input vectors from the input vector, and / or when calculating the output value, in particular during inverse distance weighting. For the selection of the at least one neighboring vector, the components of the input vector are preferably also normalized with the value ranges of the reference input vectors. Using the normalized components prevents undesired influences from different orders of magnitude of the components of the input vector or the reference input vectors. Without the normalization, components with a larger order of magnitude would be given a significantly greater weighting than components with a smaller order of magnitude.
[0060] A further development of the invention provides that the reference input vectors are reset to stored reference input vectors before determining the output vector for the input vector. Before the previously described procedure, in which at least one additional reference input vector is added to the reference input vectors, the reference input vectors are reset. For this purpose, stored reference input vectors are read out, in particular from the memory, and used as the reference input vectors. This means that any additional reference input vectors previously added to the reference input vectors are discarded. This reliably prevents any influence on the result or the output vector.
[0061] A further development of the invention provides that, in order to select the at least one neighboring vector, a classification vector is determined for each reference input vector, which classification vector results from the sign function of a difference between the input vector and the respective reference input vector. For each reference input vector, the corresponding classification vector is therefore first calculated, namely by forming the difference between the input vector and the respective reference input vector. The difference is subsequently subjected to the sign function, so that the classification vector ultimately only indicates the direction in which the reference input vector lies with respect to the input vector. The components of the classification vector can each assume the values -1, 0, and +1. This results in three possible classifications for each component. This enables a particularly simple and rapid selection of the neighboring vectors.
[0062] For each of the reference input vectors, there is a classification vector, so the number of classification vectors corresponds to the number of reference input vectors. The same applies to the classification vectors as stated for the reference input vectors. The classification vectors also form a set of classification vectors, which can contain any number of classification vectors—that is, no classification vectors at all, exactly one classification vector, or multiple classification vectors.
[0063] A further development of the invention provides that different classification vectors are determined from the classification vectors, and for different classification vectors, in particular for each different classification vector, exactly one reference input vector is selected as a neighboring vector. First, the classification vectors are examined to determine whether they differ from one another, and different classification vectors are compiled. For example, the reference input vectors are sorted according to their classification vectors.
[0064] Subsequently, exactly one reference input vector is selected as a neighbor vector for each different classification vector. This means that the selected neighbor vectors have different classification vectors. Particularly preferably, exactly one reference input vector is selected for each different classification vector. Thus, there are subsequently as many neighbor vectors as there are different classification vectors. Therefore, there can be exactly one neighbor vector or several neighbor vectors. This achieves a particularly high degree of accuracy in determining the output vector.
[0065] In simple terms, for several of the orthants, specifically for each of the orthants, exactly one neighboring vector is selected from the reference input vectors. The classification vectors ultimately indicate in which of the orthants each of the reference input vectors is located. For example, the values 0 and +1 of components of the classification vector are summarized. One of the orthants thus contains the classification vectors whose component has the value -1, whereas another of the orthants contains the classification vectors whose component has one of the values 0 and +1.
[0066] By selecting the neighboring vectors such that different classification vectors are assigned to the reference input vectors, the selected neighboring vectors lie in different orthants, so that in several of the orthants there is exactly one neighboring vector, whereas in the remaining orthants there is no neighboring vector. Particularly preferred here, too, is that in each of the orthants there is exactly one neighboring vector or one of the neighboring vectors. However, this is a simplified approach. Preferably, all possible values of the component are distinguished. The described procedure achieves the high accuracy already mentioned.
[0067] A further development of the invention provides that, from the reference input vectors whose classification vectors are identical, the one with the smallest distance to the input vector is selected as the neighboring vector. Therefore, if reference input vectors are present that have identical classification vectors, the distance to the input vector is determined for each of them. The smallest distance is then determined from the distances, and the reference input vector with the smallest distance to the input vector is used as the neighboring vector. This achieves particularly high accuracy. The described classification of the reference input vectors using the classification vectors is preferably performed each time an additional reference input vector is added to the reference input vectors.
[0068] Again, viewed in simplified terms, it is provided that for the orthants, in particular for each of the orthants, the reference input vector closest to the input vector in the respective orthant is selected as the neighboring vector. The reference input vectors are therefore distributed among the orthants. If there are several reference input vectors in the same orthant, the respective distance to the input vector is determined for each of them. The reference input vector with the smallest distance in the respective orthant is used as the neighboring vector. This procedure is particularly preferably carried out for all orthants, so that preferably one of the reference input vectors is present as a neighboring vector in each of the orthants. Of course, however, this only applies if there is actually a reference input vector in each of the orthants.In the orthant(s) in which no reference input vector is present, no neighboring vector can be selected from the reference input vectors.
[0069] A further development of the invention provides that the distance is calculated by normalizing components of the reference input vectors with components of the input vector and summing them. Each component of the reference input vector is thus normalized with the corresponding component of the input vector. The normalized components of the reference input vectors are then summed to determine the distances of the reference input vectors to the input vector. It can also be provided to squaring the components before summing in order to achieve sign correction. In this case, the distance preferably corresponds to the square root of the sum of the components of the respective reference input vector. In the manner described, the neighboring vectors can be selected particularly quickly and effectively.
[0070] A further development of the invention provides that the reference output vector of the additional reference input vector is calculated from the reference output vectors of the neighboring vectors, in particular by averaging or by means of inverse distance weighting. This procedure has already been mentioned above. The reference output vector is also assigned to the additional reference input vector. The reference output vector of the additional reference input vector is determined from the reference output vectors of the neighboring vectors.
[0071] For example, it may be provided to set the reference output vector of the additional reference input vector equal to the reference output vector of the neighboring vector that has the smallest distance from the reference input vector. However, a significant increase in accuracy can be achieved by using the reference output vectors of several of the neighboring vectors, in particular all neighboring vectors, to determine the reference output vector. For example, the reference output vector of the additional reference input vector is obtained by averaging, in particular by calculating the arithmetic mean, from the reference output vectors of the neighboring vectors.
[0072] A further development of the invention provides that the output vector is calculated from the reference output vector assigned to the at least one calculation vector, in particular by averaging or by means of inverse distance weighting. The same applies to the calculation of the output vector as to the calculation of the reference output vector of the additional reference input vector. Preferably, regardless of the number of selected calculation vectors, the output vector is calculated from all calculation vectors or the reference output vectors assigned to them. Reference is therefore made to the corresponding explanations.
[0073] Finally, the described method provides for first iteratively determining additional reference input vectors that are sufficiently close to the input vector. This is particularly preferably done by interpolation between the existing reference input vectors, in particular the stored reference input vectors. The iterative addition of the additional reference input vectors continues until at least one reference input vector has a sufficiently small distance from the input vector, or until the maximum number of iterations is reached.
[0074] Preferably, there are multiple reference input vectors whose distance from the input vector is less than the threshold value. Particularly preferably, such a reference input vector is determined for each of the orthants, if possible. In particular, at least one reference input vector must already be present in each of the orthants before the at least one additional reference input vector is added. This enables particularly high accuracy in determining the output vector.
[0075] The invention further relates to a drive device for a motor vehicle, which is operated according to the method described in this description. It is provided that the drive device has a control unit and a drive assembly and can be controlled by means of the control unit, wherein n-dimensional reference input vectors are each assigned a reference output vector, which are stored in the control unit and from which an output vector can be determined for an operating parameter for the drive assembly of the motor vehicle, and wherein the control unit is provided and designed to carry out the following steps, in particular in the specified order, when a new reference input vector with an associated new reference output vector is present: a. Determining a single error of the reference input vectors that describes the accuracy with which the reference output vector of one of the reference input vectors can be determined from the other reference input vectors without resorting to the one reference input vector and its reference output vector; b. buffering the reference input vector with the smallest single error and removing this reference input vector from the reference input vectors; c. Calculate the output vector using the new reference input vector as the input vector; d. Determining the individual error from a difference between the output vector and the reference output vector associated with the new reference input vector; e. Adding the removed reference input vector to the reference input vectors; f. Replacing the reference input vector with the smallest single error by the new reference input vector if the single error of the new reference input vector is larger than the smallest single error.
[0076] It is also intended that the following steps are carried out in the specified order to optimize the reference input vectors: a. Determine the individual errors of the reference input vectors; b. Calculating an original total error from the individual errors; c. Perform the following steps for each of the reference input vectors: i. Removing the respective reference input vector from the reference input vectors, ii. Determine the individual error of each of the remaining reference input vectors, iii. Calculating a total reduction error of the removed reference input vectors from a sum of the individual errors of the remaining reference input vectors, iv. Assigning the reduction total error to the remote reference input vector; v. adding the removed reference input vector to the reference input vectors; d. Removing the reference input vector with the smallest reduction total error and / or with a reduction total error that is less than the original total error from the reference input vectors.
[0077] The advantages of such a drive device design or such a procedure have already been pointed out. Both the drive device and the method for its operation can be further developed according to the explanations in this description, so reference is made to these in this regard.
[0078] In addition to the control unit, the drive device has a drive assembly, for example an internal combustion engine, which is controlled by the control unit. For example, the drive assembly and the control unit are supplied with the same input vector, which contains operating parameters for the drive assembly. The drive assembly is operated based on the operating parameters, whereas the control unit determines the output vector from the input vector, i.e. the operating parameters. A measured value, which represents an output variable of the drive assembly, is compared with the output vector and, in the event of a deviation, in particular a deviation exceeding a tolerance limit, a fault in the internal combustion engine is identified and / or the input vector is adjusted such that the measured value corresponds to the output vector.
[0079] The invention will be explained in more detail below with reference to the exemplary embodiments shown in the drawings, without limiting the invention. In the drawings: Fig. 1 a schematic representation of a drive device for a motor vehicle, with a drive unit and a control unit, and Fig. 2 a diagram explaining a method for operating the control unit.
[0080] The Fig. 1 shows a schematic representation of a drive device 1 for a motor vehicle, which comprises a drive unit 2 and a control unit 3 for controlling the drive unit 2. The drive unit 2 is preferably in the form of an internal combustion engine. An input vector 4 is set on the drive unit 2, which, in the exemplary embodiment shown here, is composed, for example, of a target speed 5 and at least one injection parameter 6. The injection parameter 6 is, for example, a fuel quantity to be injected, an injection time, or the like.
[0081] The operation of the drive unit 2 is based on the target speed 5 and the injection parameter 6. The operation of the drive unit 2 results in a measured value 7 for torque, a measured value 8 for fuel consumption, and a measured value 9 for emissions. The measured values 7, 8, and 9 for torque, fuel consumption, and emissions together form an actual value 10 or each an actual value 10. In the former case, the actual value 10 can also be referred to as an actual value vector.
[0082] Input vector 4 is provided as an input variable not only to drive unit 2 but also to control unit 3. A memory 11 in control unit 3 stores several n-dimensional reference input vectors, each of which is assigned a reference output vector. The reference input vector is of the same type as input vector 4, and the reference output vector is of the same type as actual value 10.
[0083] Based on the reference input vectors, control unit 3 determines an output vector 12 from input vector 4, which contains a model value 13 for torque, a model value 14 for fuel consumption, and a model value 15 for emissions. It can now be provided that the actual values 10 are fed back directly to control unit 3. However, it can also be provided that a difference 16 is first formed from the actual value 10 and the output vector 12, which is then fed back to control unit 3.
[0084] The Fig.2 shows a diagram using which a method for operating the control unit 3 is explained by way of example. Shown are reference input vectors 17 to 24, each of which is assigned a reference output vector. The input vector 4 is also indicated. An n-dimensional space around the input vector 4 is divided, for example, into orthants 25, 26, 27, and 28. In the two-dimensional space illustrated here as an example, the orthants 25, 26, 27, and 28 can also be referred to as quadrants. In any case, the orthants 25 to 28 emanate from the input vector 4. It can be seen that (purely by way of example) several of the reference input vectors 17 to 24 are present in each of the orthants 25 to 28.
[0085] Preferably, a classification vector is determined for each of the reference input vectors 17 to 24, and the reference input vectors 17 to 24 are classified based on the classification vectors. Ultimately, this allows for a more precise division than by simply dividing the reference input vectors 17 to 24 into the orthants 25, 26, 27, and 28, because the values -1, 0, and +1 are used for classification for each component of the classification vectors, resulting in three possible classes for each component. The reference input vectors 17 to 24 are classified based on the classification vectors in such a way that reference input vectors 17 to 24 with identical classification vectors are in the same class. All reference input vectors 17 to 24 that have the same classification vectors are therefore assigned to the same class.Overall, there are as many classes as there are different classification vectors.
[0086] Neighboring vectors are selected from the reference input vectors 17 to 24, wherein these are understood in particular to be the reference input vectors 17 to 24 that are closest to the input vector 4 in each of the classes or in each of the orthants 25 to 28. In the present embodiment, the reference input vectors 18, 20, 22, and 24 are used as neighboring vectors.
[0087] An additional reference input vector 29, along with the corresponding additional reference output vector, is now determined from the neighboring vectors 18, 20, 22, and 24. For example, the reference input vector 29 corresponds to the arithmetic midpoint between the neighboring vectors 18, 20, 22, and 24. In the example shown here, it lies at an intersection point of straight lines, with one of the straight lines passing through the reference input vectors 18 and 22, and a second straight line passing through the reference input vectors 20 and 24. It can be seen that the reference input vector 29 is closer to the input vector 4 than the previous reference input vectors 17 to 24.
[0088] The reference input vectors 17 to 24 and 29 closest to input vector 4 are now used again as neighboring vectors, for which purpose the classification described above is preferably carried out again. These are the reference input vectors 20, 22, 24, and 29. From these, a further additional reference input vector 30 is determined, which is added to the reference input vectors. This procedure is repeated until at least one of the reference input vectors 17 to 24, 29, and 30 has a distance from input vector 4 that is smaller than a threshold value. However, the procedure is particularly preferably repeated until for each class or in each of the orthants 25 to 28, there is a reference input vector 17 to 24, 29, and 30 that fulfills these conditions.
[0089] Subsequently, from the reference input vectors 17 to 24, 29, and 30, those closest to the input vector 4 are selected as calculation vectors. Particularly preferably, one of the reference input vectors 17 to 24, 29, and 30 is used as the calculation vector for each of the classes or from each of the orthants 25 to 28. The output vector is then calculated from these calculation vectors and the reference output vectors assigned to them, in particular by averaging or by means of inverse distance weighting.
[0090] The described procedure not only enables a rapid assignment of the output vector to the input vector, but also an improvement of the model, in particular, the addition or replacement of the reference input vectors 17 to 24 with further reference input vectors and the associated reference output vectors. Thus, in particular, it is not necessary to perform a computationally intensive Bayesian regression. Rather, the application and improvement of the model is possible and intended within the scope of normal operation of the drive device 1. LIST OF REFERENCE SYMBOLS: 1 drive device 2 drive unit 3 Control unit 4 Input vector 5 Target speed 6 Injection parameters 7 Torque 8 Fuel consumption 9 emissions 10 measured values 11 storage 12 Output vector 13 Model type 14 Model type 15 Model type 16 Difference 17 Reference input vector 18 Reference input vector 19 Reference input vector 20 Reference input vector 21 Reference input vector 22 Reference input vector 23 Reference input vector 24 Reference input vector 25 Orthant 26 Orthant 27 Orthant 28 Orthant 29 Reference input vector 30 Reference input vector
Claims
[1] Method for operating a drive device (1) for a motor vehicle, which has a control unit (3) and a drive unit (2) and is controlled by means of the control unit, wherein n-dimensional reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) are each assigned a reference output vector, which are stored in the control unit (3) and from which an output vector (12) can be determined for at least one operating parameter for the drive unit (2) of the motor vehicle, wherein when a new reference input vector with an associated new reference output vector is present, the following steps are carried out: a. Determining an individual error of the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), which describes the accuracy with which the reference output vector of one of the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) can be determined from the other reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) and the reference output vectors assigned to them, without resorting to the one reference input vector (17, 18, 19, 20, 21, 22, 23, 24) and its reference output vector; b. buffering the reference input vector (17,18,19,20,21,22,23,24) with the smallest single error and removing this reference input vector (17,18,19,20,21,22,23,24) from the reference input vectors (17,18,19,20,21,22,23,24); c. Calculating the output vector (12) with the new reference input vector as input vector (4); d. determining the individual error from a difference between the output vector (12) and the reference output vector associated with the new reference input vector; e. Adding the removed reference input vector (17,18,19,20,21,22,23,24) to the reference input vectors (17,18,19,20,21,22,23,24); f. Replacing the reference input vector (17,18,19,20,21,22,23,24) with the smallest single error by the new reference input vector if the single error of the new reference input vector is greater than the smallest single error, characterized by that to optimize the reference input vectors (17,18,19,20,21,22,23,24) the following steps are performed in the specified order: a. Determine the individual errors of the reference input vectors (17,18,19,20,21,22,23,24); b. Calculating an original total error from a sum of the individual errors; c. Perform the following steps for each of the reference input vectors (17,18,19,20,21,22,23,24): i. Removing the respective reference input vector (17,18,19,20,21,22,23,24) from the reference input vectors (17,18,19,20,21,22,23,24), ii. Determine the individual error of each of the remaining reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), iii. Calculating a total reduction error of the reduced reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) from a sum of the individual errors of the remaining reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), iv. Assigning the total reduction error to the remote reference input vector (17,18,19,20,21,22,23,24), v. adding the removed reference input vector (17,18,19,20,21,22,23,24) to the reference input vectors (17,18,19,20,21,22,23,24); d. Removing the reference input vector (17,18,19,20,21,22,23,24) with the smallest reduction total error and / or with a reduction total error that is less than the original total error from the reference input vectors (17,18,19,20,21,22,23,24). [2] Method according to claim 1, characterized by that to determine individual errors of the reference input vectors (17,18,19,20,21,22,23,24) for each of the reference input vectors (17,18,19,20,21,22,23,24) the following steps are carried out: a. buffering the respective reference input vector (17, 18, 19, 20, 21, 22, 23, 24) as a test vector and removing the reference input vector (17, 18, 19, 20, 21, 22, 23, 24) from the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24); b. Calculating the output vector (12) with the test vector as input vector (4); c. Determining the respective individual error from a difference between the output vector (12) and the reference output vector assigned to the test vector; d. Assigning the respective individual error to the test vector; and e. Adding the test vector to the reference input vectors (17,18,19,20,21,22,23,24). [3] Method according to one of the preceding claims, characterized by that if there is a reference input vector (17,18,19,20,21,22,23,24) in the reference input vectors (17,18,19,20,21,22,23,24) which corresponds to the new reference input vector, the new reference output vector is determined from the reference output vector assigned to the corresponding reference input vector (17,18,19,20,21,22,23,24) and the new reference output vector. [4] Method according to one of the preceding claims, characterized bythat the new reference output vector is determined by averaging the reference output vector and the new reference output vector. [5] Method according to one of the preceding claims, characterized by that the replacement of the reference input vector (17,18,19,20,21,22,23,24) with the new reference input vector is only performed if a number of the reference input vectors (17,18,19,20,21,22,23,24) is equal to a specified maximum number. [6] Method according to one of the preceding claims, characterized by that when the new reference input vector with the corresponding new reference output vector is available, the following steps are carried out: a. Determine the individual errors of the reference input vectors (17,18,19,20,21,22,23,24); b. Calculate the original total error as the sum of the individual errors; c. Adding the new reference input vector to the new reference input vectors (17,18,19,20,21,22,23,24); d. Recalculating the individual errors of the reference input vectors (17,18,19,20,21,22,23,24); e. Calculating a new original total error as the sum of the individual errors; f. Removing the new reference input vector from the reference input vectors (17,18,19,20,21,22,23,24) if the new original total error exceeds the original total error. [7] Method according to one of the preceding claims, characterized by that to determine the output vector to the input vector, the following steps are performed: a. Repeating the following steps until a reference input vector (17, 18, 19, 20, 21, 22, 23, 24) is present in an n-dimensional space around the input vector (4) whose distance from the input vector (4) is below a threshold value and / or a maximum number of iterations is reached: i. Selecting at least one neighboring vector from the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), ii. determining an additional reference input vector from the at least one neighboring vector and adding the additional reference input vector (29, 30) to the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24); b. Selecting at least one calculation vector closest to the input vector (4) from the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) and calculating the output vector based on the at least one selected calculation vector. [8] Method according to claim 7, characterized bythat for selecting the neighboring vectors for each of the reference input vectors (17,18,19,20,21,22,23,24) a classification vector is determined which results from the sign function of a difference between the input vector (4) and the respective reference input vector (17,18,19,20,21,22,23,24), and that different classification vectors are determined from the classification vectors and for different classification vectors exactly one reference input vector (17,18,19,20,21,22,23,24) is selected as a neighboring vector. [9] Drive device (1) for a motor vehicle, which is operated according to the method according to claim 1, wherein the drive device (1) has a control unit (3) and a drive unit (2) and can be controlled by means of the control unit, wherein n-dimensional reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) are each assigned a reference output vector, which are stored in the control unit (3) and from which an output vector (12) can be determined for at least one operating parameter for the drive unit (2) of the motor vehicle, wherein the control unit (3) is provided and designed to carry out the following steps when a new reference input vector with an associated new reference output vector is present: a. Determining an individual error of the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), which describes the accuracy with which the reference output vector of one of the reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) can be determined from the other reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) and the reference output vectors assigned to them, without resorting to the one reference input vector (17, 18, 19, 20, 21, 22, 23, 24) and its reference output vector; b. buffering the reference input vector (17,18,19,20,21,22,23,24) with the smallest single error and removing this reference input vector (17,18,19,20,21,22,23,24) from the reference input vectors (17,18,19,20,21,22,23,24); c. Calculating the output vector (12) with the new reference input vector as input vector (4); d. determining the individual error from a difference between the output vector (12) and the reference output vector associated with the new reference input vector; e. Adding the removed reference input vector (17,18,19,20,21,22,23,24) to the reference input vectors (17,18,19,20,21,22,23,24); f. Replacing the reference input vector (17,18,19,20,21,22,23,24) with the smallest single error by the new reference input vector if the single error of the new reference input vector is greater than the smallest single error characterized by that to optimize the reference input vectors (17,18,19,20,21,22,23,24) the following steps are performed in the given order: a. Determine the individual errors of the reference input vectors (17,18,19,20,21,22,23,24); b. Calculating an original total error from a sum of the individual errors; c. Perform the following steps for each of the reference input vectors (17,18,19,20,21,22,23,24): i. Removing the respective reference input vector (17,18,19,20,21,22,23,24) from the reference input vectors (17,18,19,20,21,22,23,24), ii. Determine the individual error of each of the remaining reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), iii. Calculating a total reduction error of the reduced reference input vectors (17, 18, 19, 20, 21, 22, 23, 24) from a sum of the individual errors of the remaining reference input vectors (17, 18, 19, 20, 21, 22, 23, 24), iv. Assigning the total reduction error to the remote reference input vector (17,18,19,20,21,22,23,24), v. adding the removed reference input vector (17,18,19,20,21,22,23,24) to the reference input vectors (17,18,19,20,21,22,23,24); d. Removing the reference input vector (17,18,19,20,21,22,23,24) with the smallest reduction total error and / or with a reduction total error that is less than the original total error from the reference input vectors (17,18,19,20,21,22,23,24).
Citation Information
Patent Citations
Control unit and method for calculating an output variable for a control system
DE102010028266A1
Method for operating a process using a family of characteristic curves
DE4304441A1
Control and regulation of processes in motor vehicles
DE4418731A1
