DE102019135022A1 - Method for the computer-aided evaluation of measurements of electrical currents in a high-voltage electrical system of a given electrically powered motor vehicle - Google Patents

Abstract

The invention relates to a method for the computer-aided evaluation of measurements of electrical currents (i1, i2, ..., i5) in a high-voltage on-board network (BN) of a given electrically powered motor vehicle (VE), the high-voltage on-board network (BN) having a plurality of electrical components (HVB, EM, HE, CO, IN) comprising an electrical machine (EM) for driving the specified motor vehicle (VE) and an electrical energy store (HVB) for supplying electrical energy to the electrical machine (EM) and each electrical current (i1, i2, ..., i5) flows via the same node (N) in the high-voltage vehicle electrical system (BN). In the method, measured values of the electrical currents (i1 , i2, ..., i5) are predicted for a model trip and the mean current values of these predicted measured values are determined. Based on these variables, parameter values (pw1, pw2) of a parameter vector (PV) for correcting measurement errors when measuring electrical currents (i1, i2, ..., i5) are determined, with the parameter values (pw1, pw2) being optimized be determined with the optimization goal of a minimum number of parameter values (pw1, pw2) not equal to zero in the parameter vector (PV).

Description

The invention relates to a method and a system for the computer-aided evaluation of measurements of electrical currents in a high-voltage electrical system of a given electrically driven motor vehicle.

In electrically powered vehicles, such as purely electric vehicles or hybrid vehicles, the precise determination of electrical currents in the high-voltage on-board network is of great importance. The high-voltage electrical system is the electrical system that is used to drive the vehicle electrically. It comprises an electric drive machine and an energy store for this machine. If corresponding electrical currents are not determined precisely enough, this must be taken into account when designing the electrical energy store in the high-voltage vehicle electrical system, which means that the energy store is not used efficiently.

If you consider the currents measured at a node in a high-voltage on-board electrical system, a physical current of 0 A must result at this node in accordance with Kirchhoff's node rule. Usually, however, a value clearly deviating from zero is measured in high-voltage electrical systems of conventional motor vehicles, which can be attributed to measurement errors. This value is, for example, 1 A, which results in a deficiency in a 400 V on-board electrical system of 400 W. It is desirable to correct this deficiency so that the memory protection limits of the electrical energy store in the high-voltage vehicle electrical system can be set closer to the actual physical load limits. In addition to the actual physical load limit, the memory protection limits also take into account the maximum possible failure rate. In motorized operation of the motor vehicle, this means that if the memory protection limit is chosen too conservatively, the drive power is not released at a full acceleration desired by the driver, although it would be available. In the case of recuperation, taking into account the maximum defective power means that the electrical energy storage device cannot be charged to the physical charge limit.

The document [1] describes a fleet-based approach for the detection of errors when measuring electrical currents in an electrically driven motor vehicle. The electric current of the electric drive machine is predicted using a data-driven model and compared with corresponding predicted currents that were predicted with data-driven models in other motor vehicles. Although sources of error can be classified with this method, it is not possible to correct measurement errors in the corresponding motor vehicle.

The object of the invention is to create a method for evaluating measurements of electrical currents in a high-voltage electrical system of a given electrically driven motor vehicle, with which measurement errors can be reliably recognized and corrected.

The method according to the invention is used for the computer-aided evaluation of measurements of electrical currents in a high-voltage electrical system of a given electrically driven motor vehicle. The term electrically powered motor vehicle is to be understood broadly. An electrically driven motor vehicle can be a purely electric vehicle, but also a hybrid vehicle and in particular a plug-in hybrid vehicle. It is crucial that the motor vehicle has an electric drive, and an internal combustion engine drive can also be provided. The high-voltage on-board network, which has an on-board voltage of 400 V, for example, is used to drive the specified motor vehicle electrically. It contains a plurality of electrical components which contain an electrical machine for driving the specified motor vehicle and an electrical energy store for supplying electrical energy to this electrical machine. Every electrical current flows through the same node in the high-voltage vehicle electrical system. Preferably, each electrical current occurs on a different electrical component of the plurality of electrical components.

In the method according to the invention, a first data-driven model is learned in the specified motor vehicle in a step a) using a machine learning method and this learned first data-driven model is transmitted to a back-end server by the specified motor vehicle, with a back-end server not being used for Computer device belonging to the motor vehicle is to be understood as comprising one or possibly also several computers networked with one another.

The first data-driven model is learned on the basis of a large number of measurement data records that were each measured during operation of the specified motor vehicle at different times of operation. A respective measurement data set for a respective operating time includes measurement values from Input variables and measured values of the electrical currents, with the learned first data-driven model being able to predict measured values of the electrical currents based on measured values of the input variables.

In a variant of the invention, the first data-driven model is updated at regular time intervals in the specified motor vehicle based on newly added measurement data sets. The processing steps described below in the backend server are then repeated when an updated, learned first data-driven model is available.

In a step b), the learned first data-driven model and a plurality of further learned second data-driven models are processed in the backend server. Every second data-driven model was tested in a different motor vehicle of the same design as the specified motor vehicle with the same machine learning method as the first data-driven model based on a large number of measurement data sets measured in the respective other motor vehicle for the same input variables and the same electrical currents as the first learned data-driven model. As part of the processing in the backend server, current mean values for the electrical currents for each model drive operating time are determined for a model drive measurement series, which contains measured values for the input variables used when learning the first data-driven model and for a large number of model drive operating times within a given model drive . A respective current mean value for a respective electrical current and for a respective model drive operating time is the mean value of the measured values of the respective electrical current, which are generated by the first learned data-driven model and the plurality of second data-driven models based on the measured values of the input variables for the respective model drive -The time of operation can be predicted. In other words, the mean current value is an averaging of the predicted measured values over the first and the second data-driven models. A model trip is to be understood as a predetermined trip with certain operating parameters during the trip, which are to be understood as values of input variables at corresponding model trip operating times.

After step b) has been carried out, the following step c) is carried out under certain conditions or, if appropriate, also always. The following step c) is carried out in any case when an abnormal operating behavior of the given motor vehicle and the presence of measurement errors when measuring the electrical currents in the given motor vehicle are determined as a function of the measurement data records measured in the given motor vehicle. According to step c), a parameter vector is determined by means of the measured values of the respective electrical currents predicted by the first learned data-driven model and the current mean values for the respective electrical currents. This parameter vector consists of parameter values for parameters of a first function and of parameter values for parameters of a second function. These functions can be modeled differently; a linear function is preferably used as the basis for the first and / or second function. The first function maps current values of the electrical currents with normal operating behavior of the specified motor vehicle to current values of the electrical currents with abnormal operating behavior of the specified motor vehicle, whereas the second function maps current values of the electrical currents in the absence of measurement errors to current values of the electrical currents in the presence of measurement errors in the specified Depicts motor vehicle. The first function is preferably designed in such a way that in the event that the sum of the current values of all electrical currents that are to be mapped by the first function is zero, the sum of the current values of all electrical currents from the mapping with the first function is also zero result is zero. In doing so, use is made of the knowledge that Kirchhoff's knot rule is not influenced by abnormal operating behavior (i.e. by hardware errors in the specified motor vehicle).

The parameter values in step c) are determined by means of an optimization with the optimization goal of a minimum number of parameter values not equal to zero in the parameter vector, the parameter values of the parameters of the second function being transmitted to the specified motor vehicle to correct future measured values of at least some of the electrical currents.

According to the invention, use is made of the knowledge that suitable corrections for measurement errors can be determined in a fleet-based approach by means of corresponding predicted measured values and current mean values that were determined for a model trip. These corrections are reflected in the parameters of a corresponding second function. For correction, after the parameters of the second function have been transmitted to the specified motor vehicle, the inverse of the second function in the specified motor vehicle is determined, the value of this function providing the corrected measurement value for a corresponding measured value in the specified motor vehicle.

In addition, the invention is based on the knowledge that corresponding parameter vectors cannot be clearly determined without further boundary conditions. Accordingly, a suitable optimization with the criterion of the sparse population of the parameter vector is used. The optimization can, for example, be carried out using the method described in document [2].

• - eine erste Klasse, gemäß der weder ein anormales Betriebsverhalten des vorgegebenen Kraftfahrzeugs vorliegt noch Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug vorhanden sind;
• - eine zweite Klasse, gemäß der ein anormales Betriebsverhalten des vorgegebenen Kraftfahrzeugs vorliegt, jedoch keine Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug vorhanden sind;
• - eine dritte Klasse, gemäß der kein anormales Betriebsverhalten des vorgegebenen Kraftfahrzeugs vorliegt, jedoch Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug vorhanden sind;
• - eine vierte Klasse, gemäß der ein anormales Betriebsverhalten des vorgegebenen Kraftfahrzeugs vorliegt und Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug vorhanden sind.

In a particularly preferred embodiment of the method according to the invention, depending on the measurement data records measured in the specified motor vehicle, a classification into the following classes is carried out before step c) is carried out:

• a first class, according to which there is neither an abnormal operating behavior of the specified motor vehicle nor measurement errors for the measured values of the electrical currents in the specified motor vehicle;
• a second class, according to which there is an abnormal operating behavior of the specified motor vehicle, but no measurement errors are present for the measured values of the electrical currents in the specified motor vehicle;
• a third class, according to which there is no abnormal operating behavior of the specified motor vehicle, but measurement errors for the measured values of the electrical currents in the specified motor vehicle exist;
• a fourth class, according to which there is an abnormal operating behavior of the specified motor vehicle and measurement errors for the measured values of the electrical currents in the specified motor vehicle are present.

In the context of this embodiment, the method is terminated if the first class or the second class is present, because the method aims to correct measurement errors that are not present when they are classified in the first and second class.

In a particularly preferred variant of the above embodiment, the classification is based on first features and a second feature, each first feature being assigned to an electric current and a measure of the absolute deviation of the measured values of the assigned electric current predicted by the first learned data-driven model from the average current values for the model drive operating times. In contrast, the second feature is a measure of the size of the sums of the measured values of the electrical currents determined for the respective measurement data sets for the operating time of the corresponding measurement data set. With such features, the above classification can be carried out in a simple manner.

In a preferred embodiment of the variant just described, the classification is carried out using a threshold value classifier, an abnormal operating behavior of the specified motor vehicle being present when a first value, which becomes larger as the size of the first features increases and the smaller as the size of the second features increases exceeds a first threshold value. In contrast, there are measurement errors for the measured values of the electrical currents in the specified motor vehicle when a second value, which becomes greater as the size of the second feature increases, exceeds a second threshold value.

Instead of the above threshold value classifier, another classifier can also be used in a further variant of the invention. In particular, the classification can optionally also take place via a machine-learned decision tree classifier, which is learned from the first to fourth class by means of training data sets comprising first features and a second feature as well as an associated class. Such decision tree classifiers are well known and are therefore not described in further detail.

In a further, particularly preferred variant of the invention, if the above third class is present, instead of the parameter vector of step c) a reduced parameter vector compared to the parameter vector of step c) is determined from parameter values for the parameters of the second function. In other words, in this case it is not step c), but a modified step in which a parameter vector with fewer parameter values is determined. This parameter vector only contains parameter values for the parameters of the second function and not for the parameters of the first function. In this case, the parameter values can be determined via a regression of the measured values of the electrical currents predicted by the first data-driven model on the mean current values for the model travel operating times. The parameter values of the parameters of the second function are then matched to the specified Motor vehicle transmitted to correct future measured values at least a portion of the electrical currents. The correction can in turn take place in such a way that the inverse of the function is formed with the corresponding parameter values and a correction value for a corresponding measured current value is determined via this inverse.

As explained above, the high-voltage on-board network contains at least the electrical machine and the electrical energy store. However, the high-voltage on-board network preferably also contains further electrical components. These components are in particular a heater and / or an air conditioning system and / or a converter for voltage conversion for a low-voltage vehicle electrical system, e.g. a 12 V vehicle electrical system. Each of the components mentioned can also be present several times in the high-voltage vehicle electrical system.

Depending on the configuration of the method according to the invention, different types of first or second data-driven models can be learned. The first data-driven model and the plurality of second data-driven models preferably each represent a neural network structure known per se, preferably a so-called LSTM network structure (LSTM = Low Short-Term Memory). An LSTM network structure enables regressive learning, which makes it easy to update the data-driven models.

In an alternative, particularly preferred variant, the first data-driven model and the plurality of second data-driven models each represent a state space model. Such data-driven models are also known from the prior art. In the detailed description, a data-driven model based on a state space model is explained in more detail.

In addition to the method described above, the invention relates to a system for the computer-aided evaluation of a measurement of electrical currents in a high-voltage electrical system of a specified electrically driven motor vehicle, the high-voltage electrical system comprising a plurality of electrical components including an electrical machine for driving the specified motor vehicle and a Includes electrical energy storage for supplying electrical energy to the electrical machine, with each electrical current flowing through the same node in the high-voltage vehicle electrical system. This system is set up to carry out the method according to the invention or one or more preferred variants of the method according to the invention.

The invention also relates to a back-end server which is set up to work as a back-end server in the method according to the invention or one or more preferred variants of the method according to the invention. In other words, the back-end server contains means for carrying out those method steps that are carried out by it in the method according to the invention or in one or more preferred embodiments of the method according to the invention.

The invention also relates to an electrically driven motor vehicle, the motor vehicle being set up to operate as a specified motor vehicle in the method according to the invention or in one or more preferred embodiments of the method according to the invention. In other words, the motor vehicle contains means to carry out the steps carried out by the specified motor vehicle in the method according to the invention or in one or more preferred variants of the method according to the invention.

An exemplary embodiment of the invention is described in detail below with reference to the accompanying figures.

• 1 eine schematische Darstellung eines Hochvolt-Bordnetzes in einem elektrisch angetriebenen Kraftfahrzeug, wobei für dieses Bordnetz eine Variante des erfindungsgemäßen Verfahrens verwendet wird;
• 2 eine schematische Darstellung, mittels der der Ablauf einer Variante des erfindungsgemäßen Verfahrens verdeutlicht wird; und
• 3 ein Diagramm, das die Berücksichtigung der Kirchhoff'schen Knotenregel für einen Klassifikator erläutert, der in einer Ausführungsform des erfindungsgemäßen Verfahrens zum Einsatz kommt.

Show it:

• 1 a schematic representation of a high-voltage electrical system in an electrically driven motor vehicle, a variant of the method according to the invention being used for this electrical system;
• 2 a schematic representation by means of which the sequence of a variant of the method according to the invention is clarified; and
• 3 a diagram which explains the consideration of Kirchhoff's knot rule for a classifier which is used in an embodiment of the method according to the invention.

An embodiment of the method according to the invention for a high-voltage vehicle electrical system is described below, which is shown schematically in FIG 1 is shown and there with reference numerals BN is shown. The high-voltage on-board network BN includes a variety of electrical components, with the sake of simplicity Description in the following only five electrical components are considered. The on-board network BN is installed in an electrically powered motor vehicle and includes an electric machine to drive the motor vehicle EM that uses the electrical energy of a high-voltage battery HVB is fed. An electric current occurs at the high-voltage battery i1 and an electric current at the electric machine i2 on.

Furthermore, several electrical consumers are part of the high-voltage on-board network BN . In particular, there is an electric heater HE , air conditioning CO as well as a converter IN intended for voltage conversion for a low-voltage vehicle electrical system. On the electric heater HE the electric current occurs i3 , on the air conditioning CO the electric current i4 and the electrical current at the voltage converter i5 on. Corresponding measured values of these electrical currents are measured in node N. On the basis of Kirchhoff's knot rule, which is known per se, the sum of the currents should i1 to i5 at node N are zero, provided that no measurement errors occur when recording these currents. The aim of the variant of the method according to the invention described here is now to correct corresponding measurement errors in the currents i1 to i5 to be detected and corrected appropriately.

2 shows the general course of the procedure. The following is a specified motor vehicle VE considered in which the high-voltage electrical system BN the 1 is installed. In addition to this motor vehicle, other identical motor vehicles are used in the method VE ' a fleet considered, for the same way as the specified motor vehicle VE Corresponding corrections for the current values of the high-voltage on-board electrical systems installed in it can be determined. In the following, however, the correction is only made for the specified motor vehicle VE described.

When driving the specified motor vehicle VE are measured values of the currents corresponding to a large number of operating times i1 to i5 detected by means of current sensors at node N. Furthermore, measured values of further variables, which are referred to below as input variables, are also recorded at the respective operating times. These variables are related to the currents that occur. In the variant described here, the acceleration of the given motor vehicle, the speed of the given motor vehicle and the inclination of the given motor vehicle relative to the horizontal are considered as input variables. These sizes are in 1 With u1 , u2 and u3 designated. If necessary, further or other input variables can also be taken into account that are related to the measured currents. For example, the gear engaged in the specified motor vehicle or the power requirement and / or allocation to the electrical machine or a speed of the electrical machine can also be used as input variables.

The recorded measured values of the input variables and the correspondingly occurring currents result in a large number of measurement data records at the respective operating times. These measurement data sets are in 1 With MD designated. In the embodiment described here, they are used as training data for a data-driven model MO used that in the motor vehicle VE with a known learning process LE is learned. Depending on the design, different data-driven models can be used. In the embodiment described here, a data-driven model in the form of a state space model is considered. Such models with the associated learning processes are known per se to the person skilled in the art. The state space model used in the variant described here is explained in more detail below.

As a result of learning LE the learned data-driven model is obtained MO , which corresponds to a first data-driven model within the meaning of the claims. The model includes parameters in the form of corresponding matrices. The parameters and in this sense the data-driven model are transferred to a backend server after learning SE transfer. There are other data-driven models in the server MO ' of the other motor vehicles VE ' deposited in the fleet. These data-driven models, which correspond to second data-driven models within the meaning of the patent claims, were made in the same way as for the vehicle VE learned with corresponding measurement data of the respective motor vehicle and sent to the backend server SE transfer.

In one step S1 are now in the backend server SE Current mean values is using the data-driven models MO and MO ' for a model test series MUD a model ride UD (UD = Unity Drive). The model drive was suitably determined in advance and covers the operating conditions that usually occur in the motor vehicles under consideration very well VE and VE ' from. The model travel measurement series contains measured values of corresponding input variables for a large number of model travel operating times u1 , u2 and u3 . As part of the step S1 come with all data-driven models MO and MO ' the corresponding measured values of the currents i1 to i5 at the times of operation of the model run by means of the measured values of the input variables u1 to u3 predicted. This results in the respective Model drive operating time predicted measured values for all data-driven models. These predicted measured values are averaged over all data-driven models, so that a large number of current averages is can be obtained for the respective model drive operating times. In other words, current mean values exist at each time when the model is in operation is for the currents i1 to i5 .

A corresponding mean current value can be regarded as an approximation of the so-called "ground truth", which reflects the current values that actually occur for the model drive of a perfect motor vehicle, i.e. a motor vehicle in which neither measurement errors in the measurement of the currents nor hardware errors occur. It is assumed here that these errors are averaged out when averaging over a large number of vehicles. Hardware errors result from a faulty function of one or more hardware components in the corresponding motor vehicle. Such errors result in the operating behavior of the motor vehicle deviating from the normal operating behavior (i.e. there is an abnormal operating behavior).

• - die Klasse CL1, gemäß der weder Hardwarefehler noch Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug vorliegen;
• - eine Klasse CL2, gemäß der Hardwarefehler im vorgegebenen Kraftfahrzeug vorliegen, jedoch keine Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug aufgetreten sind;
• - eine Klasse CL3, gemäß der keine Hardwarefehler im vorgegebenen Kraftfahrzeug vorliegen, jedoch Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug aufgetreten sind;
• - eine Klasse CL4, gemäß der Hardwarefehler im vorgegebenen Kraftfahrzeug aufgetreten sind und ferner auch Messfehler für die Messwerte der elektrischen Ströme im vorgegebenen Kraftfahrzeug vorhanden sind.

The average current values obtained for the model run operating times is as well as those for learning the first data-driven model MO used current measured values that are sent to the backend server SE together with the learned data-driven model MO be transferred, and also those under step S1 predicted current measured values of the motor vehicle VE are then in one step S2 processed by a classifier, according to which a division into the four classes explained below takes place. The classification step S2 is explained again in detail below. According to the classification step, the following classes are considered:

• - the class CL1 , according to which there are neither hardware errors nor measurement errors for the measured values of the electrical currents in the specified motor vehicle;
• - a class CL2 , according to which there are hardware errors in the specified motor vehicle, but no measurement errors have occurred for the measured values of the electrical currents in the specified motor vehicle;
• - a class CL3 , according to which there are no hardware errors in the specified motor vehicle, but measurement errors have occurred for the measured values of the electrical currents in the specified motor vehicle;
• - a class CL4 , according to which hardware errors have occurred in the specified motor vehicle and, furthermore, measurement errors are also present for the measured values of the electrical currents in the specified motor vehicle.

Is there a classification or assignment to the class CL1 or CL2 , the procedure is aborted without further steps, since no measurement errors occur in these classes and a correction is therefore not required. If the class is available CL4 is based on a combination of two functions f1 and f2 a parameter vector PV determines the parameter values pw1 for parameters of the function f1 and parameter values pw2 for parameters of the function f2 specified. According to the function f1 hardware errors are taken into account, ie the function describes a mapping of the current values of the electrical currents i1 to i5 without hardware errors on current values when hardware errors occur. With the first function f1 the restriction based on Kirchhoff's knot rule is used, according to which, in the event that the sum of the current values associated with the function f1 are to be mapped, zero is also the sum of the current values that result from the mapping with the function f1 result is zero. This takes into account the fact that hardware errors change physical laws and thus also do not change Kirchhoff's knot rule.

In contrast to the function f1 are through the function f2 Measurement errors are taken into account, so that the above restriction given by Kirchhoff's knot rule does not apply to this function. The function f2 maps current values without the presence of measurement errors to corresponding current values in the presence of measurement errors in the specified motor vehicle. The parameter values pw1 and pw2 of the parameter vector PV are determined using a special mathematical method in which the sparse occupation of the parameter vector with parameter values other than 0 is assumed. This is explained in more detail below.

After determining the parameter vector PV become the parameter values pw2 the second function f2 , ie the parameter values which relate to measurement errors, to the specified motor vehicle VE transfer. The motor vehicle can then VE using these parameter values pw2 a correction of occurring measured values of the electrical currents i1 to i5 carry out. This is done in a simple manner in that in the motor vehicle VE the the parameter values pw2 underlying function f2 is inverted, whereupon current values adjusted with measurement errors can then be determined from current values, in which the measurement errors are corrected. In this way, greater accuracy can be achieved when measuring the currents in the motor vehicle VE can be achieved.

Used according to the classification in step S2 the class CL3 obtained, the determination of appropriate parameters can be easier than in the case of the class CL4 respectively. In particular, it is no longer necessary to use the function f1 to take into account what hardware failure modeled because in the class CL3 there are no hardware errors. Therefore becomes a parameter vector PV ' with fewer parameter values pw ' just for function f2 used. It is possible to use the function f2 by means of a regression that of the first data-driven model MO Predicted current values on the mean current values is to determine. In the case of a classification or assignment to the class CL4 become the parameter values pw ' to the motor vehicle VE transmitted and can there by inverting the function f2 can be used to correct measured values of electrical currents.

den Wert Null annimmt. Dabei ist zu beachten, dass im Falle von Messfehlern dieses Kriterium nicht gelten muss. Zur Vereinfachung der Notation werden nunmehr durch i bzw. im die K-dimensionalen Vektoren (i₁, i₂, ... , i_K)^T bzw. (i_1,m, i_2,m, ..., i_K,m)^T bezeichnet.This is explained in detail below using 2 explained procedures. In general, a high-voltage on-board network consisting of K currents i ₁ , i ₂ , i _{K is} considered, in the example of which 2 K = 5 applies. For each current i _k (k = 1, 2, ..., K), a measured value of i _k is generated according to the measurement data records MD provided. This measured value is referred to below as i _{k, m} . All K currents flow through node N of 1 , so that, according to Kirchhoff's knot rule, the sum ${\displaystyle \sum {}_{k=1}^K}{i}_k$

takes the value zero. It should be noted that this criterion does not have to apply in the event of measurement errors. To simplify the notation, the K-dimensional vectors (i ₁ , i ₂ , ..., i _K ) ^T and (i _{1, m} , i _{2, m} , ..., i _{K , m} ) ^T denotes.

While the specified motor vehicle is in motion, the measured currents {i _m (t)} _{i≤t≤T result} from the measurement data sets MD , where t corresponds to a respective operating time. Based on these data, the goal is now _{to determine a correction mapping i m} ↦ i. This mapping can then be used to correct measurement signals in real time. As described above, this correction mapping is provided by corresponding parameter values pw ' or pw2 described, which are attached to the specified motor vehicle VE be transmitted. When determining the correction mapping, there is the problem that only the measured values i _m (t) are available, but not the ground truth i (t) for the motor vehicle without measurement errors. Since this ground truth is not known, the corresponding mapping cannot be determined a priori via monitored learning.

To solve the above problem, the ideal ground truth i _s estimated as described below. This ground truth corresponds to the size is out 2 . A classifier then decides whether the specified motor vehicle has hardware errors or not, as will also be explained further below. Finally, depending on the classification result, a routine for learning the desired mapping for correcting measurement errors is carried out.

In the following, the estimation of the ideal ground truth is first presented i _s explained. For the ideal ground truth, it is assumed that there is a perfect motor vehicle, ie a motor vehicle with normal operating behavior without hardware errors, the current values of which are recorded without measurement errors. The k-th current in such a perfect vehicle is referred to below as i _{k, s} . There is thus the current vector i _s = (i _{1, s} , i _{2, s} , ..., i _{K, s} ) for a perfect vehicle. The vector i _s can be regarded as the nominal value of im, ie i _s = i _m should apply. For an imperfect vehicle, however, i _s ≠ i _m results, this deviation being caused by hardware errors or measurement errors or a combination of both types of errors.

Hardware faults (for example a flat tire) usually lead to the motor vehicle exhibiting abnormal dynamic operating behavior that deviates from the operating behavior of a perfect vehicle. This has the consequence that the electricity i _s changed to current i. Nevertheless, Kirchhoff's knot rule remains valid, ie the sum of the measured currents remains zero, so that 1 ^T i = 1 ^T i _s = 0 applies (1 = (1,1, ..., 1) ^T ).

On the other hand, measurement errors (e.g. sensor offsets or sensor drifts) cause a second deviation between im and i. In contrast to hardware errors, this deviation usually leads to a change in the sum of the measured currents, so that generally 1 ^T i _m ≠ 0 applies. Overall, the deviation of measured current values from the current values of a perfect motor vehicle can be described as follows: $$\underset{\text{Total deviation}}{{\displaystyle {\text{i}}_m-{\text{i}}_s}}=\underset{\text{Measurement error}}{{\displaystyle {\text{i}}_m-\text{i}}}+\underset{\text{Hardware failure}}{{\displaystyle \text{i}-{\text{i}}_s}}$$

bezeichnet. Für jedes Fahrzeug der Flotte wird ein diskretes Zustandsraummodell gelernt, das dem ersten datengetriebenen Modell für das vorgegebene Fahrzeug VE und den zweiten datengetriebenen Modellen für die anderen Fahrzeuge VE' entspricht. Zustandsraummodelle und deren Lernen sind dem Fachmann geläufig. Das hier verwendete Zustandsraummodell wird in dem Dokument [1] näher beschrieben.In the method described here, a first data-driven model is now used for the specified motor vehicle VE learned. In the same way, corresponding second data-driven models were also used for other vehicles VE ' learned of the fleet under consideration. It thus becomes the fleet of N (imperfect) motor vehicles including the vehicle VE and the vehicles VE ' considered. The measured current of an (arbitrary) j-th motor vehicle (j = 1, .., N) from the fleet is also included ${\text{i}}_m^j$

designated. A discrete state space model is learned for each vehicle in the fleet, which is the first data-driven model for the given vehicle VE and the second data-driven models for the other vehicles VE ' corresponds to. State space models and their learning are familiar to the person skilled in the art. The state space model used here is described in more detail in document [1].

basierend auf gemessenen Strömen ${\left\{{\text{i}}_m^j\left(t\right)\right\}}_{1\le t\le T^j}$

gelernt, wie dies auch in dem Dokument [1] beschrieben ist. In einem nächsten Schritt werden die Ströme ${\left\{{\text{i}}_m^j\right\}}_{1\le j\le N}$

abgeschätzt, die über die Flotte der Kraftfahrzeuge gemessen würden, falls alle Kraftfahrzeuge unter den gleichen Betriebsbedingungen fahren würden (siehe auch Dokument [1]). Mit anderen Worten wird die obige Modellfahrt betrachtet, die in 2 mit UD bezeichnet ist und entsprechende Werte von Eingangsvektoren u zu entsprechenden Modellfahrt-Betriebszeitpunkten enthält. Für diese Modellfahrt werden mit den trainierten Zustandsraummodellen entsprechende Stromwerte ${\left\{{\text{i}}_m^j\left(t\right)\right\}}_{1\le j\le N}$

prädiziert. Für einen entsprechenden Modellfahrt-Betriebszeitpunkt wird diese Prädiktion durch folgende Gleichung beschrieben: $$\begin{array}{l}\text{x}\left(t+1\right)={\text{A}}^j\text{x}\left(t\right)+{\text{B}}^j{\text{u}}_{ud}\left(t\right)\\ {\text{i}}_{sim}^j\left(t\right)={\text{C}}^j\text{x}\left(t\right)+{\text{D}}^j{\text{u}}_{ud}\left(t\right)\end{array}$$

In the context of learning the state space model, system matrices A ^j , B ^j , C ^j and D ^j with state vector x and input vector u are made up of the above input variables u1 , u2 and u3 and output vector ${\text{i}}_m^j$

based on measured currents ${\left\{{\text{i}}_m^j\left(t\right)\right\}}_{1\le t\le T^j}$

learned, as this is also described in document [1]. In a next step the currents ${\left\{{\text{i}}_m^j\right\}}_{1\le j\le N}$

estimated, which would be measured over the fleet of motor vehicles if all motor vehicles were to drive under the same operating conditions (see also document [1]). In other words, consider the above model ride, which is shown in 2 With UD and contains corresponding values of input vectors u at corresponding model travel operating times. For this model trip, corresponding current values are generated with the trained state space models ${\left\{{\text{i}}_m^j\left(t\right)\right\}}_{1\le j\le N}$

predicted. For a corresponding model travel operating time, this prediction is described by the following equation: $$\begin{array}{l}\text{x}\left(t+1\right)={\text{A.}}^j\text{x}\left(t\right)+{\text{B.}}^j{\text{u}}_{ud}\left(t\right)\\ {\text{i}}_{sim}^j\left(t\right)={\text{C.}}^j\text{x}\left(t\right)+{\text{D.}}^j{\text{u}}_{ud}\left(t\right)\end{array}$$

• x₀ = x(0) (beliebige Initialisierung) und u(t) = u_ud(t), t ∈ {1,2, ...,T_ud} (d.h. t bezeichnet einen Modellfahrt-Betriebszeitpunkt).

The following applies:

• x ₀ = x (0) (arbitrary initialization) and u (t) = u _ud (t), t ∈ {1,2, ..., T _ud } (ie t denotes a model travel operating time).

• Annahme 1: Für ein ausreichend großes N und in jedem Zeitschritt t kann i_s(t) genau durch den Mittelwert der simulierten bzw. prädizierten Ströme ${\text{i}}_{sim}^j\left(t\right)$
  
  über die N Kraftfahrzeuge der Flotte approximiert werden, d.h. es gilt: $${\text{i}}_s\left(t\right)\approx {\tilde{\text{i}}}_s\left(t\right):=\frac{1}{N}{\displaystyle \sum _{j=1}^N{\text{i}}_{sim}^j\left(t\right)}$$
  

Using the designation i _s (t) for the current value of a perfect vehicle at time t in the model drive UD the following assumption is used in the procedure described here:

• Assumption 1: For a sufficiently large N and in each time step t, i _s (t) can be calculated precisely through the mean value of the simulated or predicted currents ${\text{i}}_{sim}^j\left(t\right)$
  
  can be approximated via the N vehicles of the fleet, i.e. the following applies: $${\text{i}}_s\left(t\right)\approx {\tilde{\text{i}}}_s\left(t\right):=\frac{1}{N}{\displaystyle \sum _{j=1}^N{\text{i}}_{sim}^j\left(t\right)}$$
  

sowie der Annahme, dass sich für ein großes N die Messfehler, Hardwarefehler und Simulationsfehler zu einem bestimmten Zeitpunkt zu Null über die Flotte hinweg mitteln. Demzufolge wird gemäß der obigen Annahme der Vektor des Strommittelwerts bestimmt und in den weiteren Verfahrensschritten verarbeitet. Ebenso wird für das entsprechende Kraftfahrzeug der Vektor seiner simulierten Stromwerte ${\text{i}}_{sim}^j\left(t\right)$

weiterverarbeitet.This assumption is based on observation $$\begin{array}{c}\frac{1}{N}{\displaystyle \sum _{j=1}^N{\text{i}}_{sim}^j}=\frac{1}{N}{\displaystyle \sum _{j=1}^N\left({\text{i}}_s+{\text{i}}_m^j-{\text{i}}^j+{\text{i}}^j-{\text{i}}_s+{\text{i}}_{sim}^j-{\text{i}}_m^j\right)}\\ =i_s+\frac{1}{N}{\displaystyle \sum _{j=1}^N\left({\text{i}}_m^j-{\text{i}}^j\right)}+\frac{1}{N}{\displaystyle \sum _{j=1}^N\left({\text{i}}^j-{\text{i}}_s\right)}+\frac{1}{N}{\displaystyle \sum _{j=1}^N\left({\text{i}}_{sim}^j-{\text{i}}_m^j\right)}\end{array}$$

as well as the assumption that for a capital N the measurement errors, hardware errors and simulation errors average to zero at a certain point in time across the fleet. Accordingly, according to the above assumption the vector of the mean current value is determined and processed in the further process steps. The vector of its simulated current values is also used for the corresponding motor vehicle ${\text{i}}_{sim}^j\left(t\right)$

further processed.

The following now refers to the specified motor vehicle VE the 2 With reference, that is, the index j now relates to the specified vehicle VE . In the same way, the steps described below can also be used for the other motor vehicles VE ' of the fleet. As a rule, the method described here is also carried out for all vehicles in the fleet.

für jeden Zeitschritt t ∈ {1,2, ... ,T_ud} der Modellfahrt erhalten. Mit anderen Worten gilt ${\text{i}}_m^j\left(t\right)\approx {\text{i}}_{sim}^j\left(t\right)$

${\text{und i}}_s\left(t\right)\approx {\tilde{\text{i}}}_s\left(t\right).$

Das hier betrachtete Problem wäre somit für das vorgegebene Kraftfahrzeug fast gelöst, wenn das Auftreten von Hardwarefehlern ausgeschlossen werden könnte. In diesem Fall würde i^j(t) = i_s(t) ≈ ĩ_s(t) gelten, was dann zusammen mit ${\text{i}}_{sim}^j\left(t\right)$

genutzt werden kann, um die gewünschte Abbildung ${\text{i}}_m^j\mapsto {\text{i}}^j$

zu lernen (d.h. die Abbildung kann mit den Tupeln ${\left\{\left({\text{i}}_{sim}^j\left(t\right),{\tilde{\text{i}}}_s\left(t\right)\right)\right\}}_{1\le t\le T_{ud}}$

gelernt werden).According to the above calculations, an estimate was made for ${\text{i}}_m^j{\text{and i}}_s$

obtained for each time step t ∈ {1,2, ..., T _ud } of the model trip. In other words, ${\text{i}}_m^j\left(t\right)\approx {\text{i}}_{sim}^j\left(t\right)$

${\text{and i}}_s\left(t\right)\approx {\tilde{\text{i}}}_s\left(t\right).$

The problem considered here would thus be almost solved for the given motor vehicle if the occurrence of hardware errors could be excluded. In this case i ^j (t) = i _s (t) ≈ ĩ _s (t) would apply, which is then combined with ${\text{i}}_{sim}^j\left(t\right)$

can be used to make the illustration you want ${\text{i}}_m^j\mapsto {\text{i}}^j$

to learn (ie the mapping can with the tuples ${\left\{\left({\text{i}}_{sim}^j\left(t\right),{\tilde{\text{i}}}_s\left(t\right)\right)\right\}}_{1\le t\le T_{ud}}$

be learned).

gelernt werden, wie im obigen Absatz beschrieben wurde. Liegen hingegen Hardwarefehler vor, wird ein komplexerer Algorithmus zur Ermittlung von Messfehlern genutzt, wie weiter unten noch näher beschrieben wird.In order to reduce the complexity of the method, in the embodiment described here, a classification is used that corresponds to the step S2 the 2 corresponds to. This classification can be used to predict whether in the specified motor vehicle VE Hardware errors occur. If this is not the case, the above illustration can easily be compared with the sizes ${\left\{\left({\text{i}}_{sim}^j\left(t\right),{\tilde{\text{i}}}_s\left(t\right)\right)\right\}}_{1\le t\le T_{ud}}$

can be learned as described in the paragraph above. If, on the other hand, there are hardware errors, a more complex algorithm is used to determine measurement errors, as will be described in more detail below.

im Durchschnitt größer wird, wenn im vorgegebenen Kraftfahrzeug nicht nur Messfehler, sondern auch signifikante Hardwarefehler auftreten. Demzufolge wird mittels ${\text{i}}_{sim}^j\left(t\right)und{\tilde{\text{i}}}_s\left(t\right)$

für das vorgegebene Kraftfahrzeug und für jeden Strom k die durchschnittliche quadratische Abweichung zwischen $i_{k,m}^j\text{ und }{i}_{k,s}$

bestimmt. Mit anderen Worten wird das k-te Merkmal, das einem jeweiligen ersten Merkmal im Sinne der Patentansprüche entspricht, für das j-te Kraftfahrzeug (d.h. das vorgegebene Kraftfahrzeug) wie folgt bestimmt: $$f_k^j=\frac{1}{T_{ud}}{\displaystyle \sum _{t=1}^{T_{ud}}{\left(i_{k,sim}^j\left(t\right)-{\tilde{i}}_{k,s}\left(t\right)\right)}^2\approx \mathbb{E}}\left({\left(i_{k,m}^j-i_{k,s}\right)}^2\right)$$

The classification carried out is first explained below. The classification is based on K + 1 features f ₁ , f ₂ , ..., f _{K + 1} . It is assumed that the absolute deviation between ${\text{i}}_m^j{\text{and i}}_s$

becomes larger on average if not only measurement errors but also significant hardware errors occur in the given motor vehicle. As a result, means ${\text{i}}_{sim}^j\left(t\right)und{\tilde{\text{i}}}_s\left(t\right)$

for the given motor vehicle and for each current k, the mean square deviation between $i_{k,m}^j\text{and}{i}_{k,s}$

certainly. In other words, the kth feature, which corresponds to a respective first feature within the meaning of the patent claims, is determined for the jth motor vehicle (ie the specified motor vehicle) as follows: $$f_k^j=\frac{1}{T_{ud}}{\displaystyle \sum _{t=1}^{T_{ud}}{\left(i_{k,sim}^j\left(t\right)-{\tilde{i}}_{k,s}\left(t\right)\right)}^2\approx \mathbb{E.}}\left({\left(i_{k,m}^j-i_{k,s}\right)}^2\right)$$

entspricht dem Erwartungswert.The following applies: 1 ≤ k ≤ K. The size

corresponds to the expected value.

führen, dessen Einträge sich mit hoher Wahrscheinlichkeit nicht zu Null summieren, kann dieses Merkmal als quadratisches Mittel von $1^T{\text{i}}_m^j$

abgeschätzt werden, d.h. es gilt: $$f_{K+1}^j=\frac{1}{T^j}{\displaystyle \sum _{t=1}^{T^j}\left({\left(1^T{\text{i}}_m^j\left(t\right)\right)}^2\right)\approx \mathbb{E}}\left({\left(1^T{\text{i}}_m^j\right)}^2\right)$$

In addition, the (k + 1) th feature, which corresponds to a second feature within the meaning of the patent claims, is defined as a measure of the significance of measurement errors in the specified motor vehicle. Because measurement errors result in a current vector ${\text{i}}_m^j$

whose entries do not add up to zero with a high probability, this characteristic can be expressed as the root mean square of $1^T{\text{i}}_m^j$

can be estimated, i.e. the following applies: $$f_{K+1}^j=\frac{1}{T^j}{\displaystyle \sum _{t=1}^{T^j}\left({\left(1^T{\text{i}}_m^j\left(t\right)\right)}^2\right)\approx \mathbb{E.}}\left({\left(1^T{\text{i}}_m^j\right)}^2\right)$$

erforderlich, d.h. die gemessenen Ströme aus den Messdatensätze MD gemäß 2.The original measured currents are used for this ${\left\{{\text{i}}_m^j\left(t\right)\right\}}_{1\le t\le T^j}$

required, ie the measured currents from the measurement data records MD according to 2 .

beschrieben. Basierend auf diesem Vektor f^j wird eine entsprechende Klassifikation vorgenommen. In einer Variante wird dabei ein einfacher Schwellwertklassifikator genutzt. Bei diesem Klassifikator wird die Entscheidung, ob Hardwarefehler vorliegen, mittels folgender Formel getroffen: $${\eta }^j:={\displaystyle \sum _{i=1}^K{f}_i^j-\frac{1}{K}{f}_{K+1}^j>{\delta }_h}$$

The following are the first features and the second feature by the vector ${\text{f}}^j={\left(f_1^j,f_2^j,...,f_{K+1}^j\right)}^T$

described. A corresponding classification is carried out on the basis of this vector f ^j. In a variant, a simple threshold value classifier is used. With this classifier, the decision as to whether there are hardware errors is made using the following formula: $${\eta }^j:={\displaystyle \sum _{i=1}^K{f}_i^j-\frac{1}{K}{f}_{K+1}^j>{\delta }_H}$$

In other words, according to equation (7) above, there is a hardware fault if the variable η ^j exceeds the suitably selected threshold value δ _h. For an intuitive explanation, the above criterion can be rewritten as follows: $$\begin{array}{l}{\eta }^j\approx {\displaystyle \sum _{k=1}^K\mathbb{E.}}\left({\left(i_{k,m}^j-i_{k,s}\right)}^2\right)-\frac{1}{K}\mathbb{E.}\left({\left(1^T{\text{i}}_m^j\right)}^2\right)\\ =\mathbb{E.}\left({\displaystyle \sum _{k=1}^K{\left(i_{k,m}^j-i_{k,s}\right)}^2-{\left(\frac{1^T{\text{i}}_m^j}{\sqrt{K}}\right)}^2}\right)\\ =\mathbb{E.}\left({\Vert {\text{i}}_m^j-{\text{i}}_s\Vert }_2^2-{\Vert {\text{i}}_m^j-{\mathrm{\Pi }}_H\left({\text{i}}_m^j\right)\Vert }_2^2\right)>{\delta }_H\end{array}$$

Da der Strom i_s die Kirchhoff'sche Knotenregel erfüllt, gilt aufgrund des Satzes von Pythagoras folgende Beziehung: $${\Vert {\text{i}}_m^j-{\text{i}}_s\Vert }_2^2-{\Vert {\text{i}}_m^j-{\mathrm{\Pi }}_H\left({\text{i}}_m^j\right)\Vert }_2^2={\Vert {\mathrm{\Pi }}_H\left({\text{i}}_m^j\right)-{\text{i}}_s\Vert }_2^2$$

The above formula (5) was inserted into the criterion according to equation (7). The quantity Π _H is the projection operator onto the hypersurface H that is given by $H=\left\{\text{x}\in {ℝ}^K|1^T\text{x}=0.\right\}$

Because the stream i _s fulfills Kirchhoff's knot rule, the following relationship applies based on the Pythagorean theorem: $${\Vert {\text{i}}_m^j-{\text{i}}_s\Vert }_2^2-{\Vert {\text{i}}_m^j-{\mathrm{\Pi }}_H\left({\text{i}}_m^j\right)\Vert }_2^2={\Vert {\mathrm{\Pi }}_H\left({\text{i}}_m^j\right)-{\text{i}}_s\Vert }_2^2$$

wir durch den Pfeil ME in 3 angedeutet ist. Die Projektion des Punkts $i_m^j$

zurück zur Hyperebene H führt zu dem Punkt ${\mathrm{\Pi }}_H\left(i_m^j\right).$

In der obigen Gleichung (8) wird der Satz von Pythagoras auf ein Dreieck angewandt, das durch die Punkte $i_s,i_m^j\text{ und }{\mathrm{\Pi }}_H\left(i_m^j\right)$

aufgespannt ist. Dieses Dreieck hat einen rechten Winkel am Punkt ${\mathrm{\Pi }}_H\left(i_m^j\right).$

The above relationship is again in the diagram of 3 for K = 2, ie two electric currents i1 and i2 , shown. The hyperplane H is in 3 as a dashed line H reproduced. According to Kirchhoff's rule of knots, this line describes the relationship i ₁ + i ₂ = 0. The point i _s is the nominal current (ie the ground truth), which according to Kirchhoff's node rule in the hyperplane H lies. Hardware failures move the point i _s to the point i ^{j of} the hyperplane ℌ, as indicated by the arrow HE in 3 is indicated. In contrast, measurement errors move the point i ^j out of the hyperplane ℌ to the point $i_m^j,$

we by the arrow ME in 3 is indicated. The projection of the point $i_m^j$

back to the hyperplane H leads to the point ${\mathrm{\Pi }}_H\left(i_m^j\right).$

In equation (8) above, the Pythagorean theorem is applied to a triangle that passes through the points $i_s,i_m^j\text{and}{\mathrm{\Pi }}_H\left(i_m^j\right)$

is stretched. This triangle has a right angle at the point ${\mathrm{\Pi }}_H\left(i_m^j\right).$

als die bestmögliche Approximation für i^j bei Abwesenheit von weiteren Informationen betrachtet werden. Somit ist das obige Kriterium eine Approximation der gemittelten quadratischen Distanz ${\Vert {\text{i}}_m^j-{\text{i}}_s\Vert }_2^2,$

d.h. eine Approximation der mittleren quadratischen Abweichung, die durch Hardwarefehler verursacht wird (siehe auch Gleichung (1)).Knowing that i ^j satisfies Kirchhoff's knot rule (ie i ^j ∈ H), the vector can ${\mathrm{\Pi }}_H\left({\text{i}}_m^j\right)$

can be considered the best possible approximation for i ^j in the absence of further information. Thus, the above criterion is an approximation of the mean square distance ${\Vert {\text{i}}_m^j-{\text{i}}_s\Vert }_2^2,$

ie an approximation of the mean square deviation caused by hardware errors (see also equation (1)).

genutzt werden. Überschreitet dieses Merkmal den Schwellwert, liegen Messfehler vor. Ist dies nicht der Fall, liegen keine Messfehler vor.The classification for the presence of hardware errors was described above. In the same way, a classification can also be made with a corresponding threshold value can be made to determine whether there is a measurement error. A threshold value for the second feature can be used for this purpose $f_{K+1}^j$

be used. If this feature exceeds the threshold value, there are measurement errors. If this is not the case, there are no measurement errors.

Instead of the above threshold value classifier, a known decision tree classifier can also be used in a modified embodiment for the specified motor vehicle. This decision tree classifier is learned in advance on the basis of suitable training data in the form of feature vectors f ^j with the associated information as to which class the respective feature vector belongs to. Decision tree classifiers and corresponding learning methods for these classifiers are known per se and are therefore not described in further detail.

• - Klasse 1: Es werden weder Hardwarefehler noch Messwertfehler detektiert. In diesem Fall liegt ein nahezu perfektes Fahrzeug vor, so dass das Verfahren abgebrochen wird.
• - Klasse 2: Es werden nur Hardwarefehler detektiert. Auch in diesem Fall wird das Verfahren abgebrochen, da erfindungsgemäß nur Messfehler korrigiert werden sollen.
• - Klasse 3: Nur Messfehler werden detektiert. In diesem Fall kann auf einfache Weise eine Korrekturabbildung ${\text{i}}_m^j\mapsto {\text{i}}^j$
  
  unter Verwendung der Tupel ${\left\{\left({\text{i}}_{sim}^j\left(t\right),{\text{i}}_s\left(t\right)\right)\right\}}_t$
  
  gelernt werden.
• - Klasse 4: Es liegen sowohl Hardwarefehler als auch Messfehler vor, d.h. die Abweichung zwischen ${\text{i}}_m^j{\text{ und i}}_s$
  
  setzt sich aus zwei Komponenten zusammen, wie dies in der obigen Gleichung (1) dargestellt ist. Ohne weitere Annahmen ist dieses Problem jedoch nicht eindeutig lösbar. Um zu einer geeigneten Lösung zu kommen, werden Annahmen getroffen, mit denen es möglich ist, eine Lösung zu finden. Die entsprechenden Annahmen und die Lösung für die Klasse 4 werden später näher erläutert.

Overall, the following classes or cases result from the above classification:

• - Class 1: Neither hardware errors nor measured value errors are detected. In this case, the vehicle is almost perfect, so the method is terminated.
• - Class 2: Only hardware errors are detected. The method is also terminated in this case, since according to the invention only measurement errors are to be corrected.
• - Class 3: Only measurement errors are detected. In this case, a correction image can easily be made ${\text{i}}_m^j\mapsto {\text{i}}^j$
  
  using the tuple ${\left\{\left({\text{i}}_{sim}^j\left(t\right),{\text{i}}_s\left(t\right)\right)\right\}}_t$
  
  to be learned.
• - Class 4: There are both hardware errors and measurement errors, ie the deviation between ${\text{i}}_m^j{\text{and i}}_s$
  
  is composed of two components, as shown in equation (1) above. Without further assumptions, however, this problem cannot be clearly solved. In order to arrive at a suitable solution, assumptions are made with which it is possible to find a solution. The corresponding assumptions and the solution for class 4 are explained in more detail later.

modelliert werden, für die gilt: $${\text{i}}_s\in H\Rightarrow {\text{f}}_h\left({\text{i}}_s\right)\in H$$

Grades 1 to 4 above are in 2 as CL1 to CL4 designated. In the following, the error models that are considered when the above classes 3 and 4 are present are first explained. For reasons of clarity, the index j is now omitted. Since hardware errors do not change the validity of physical laws, ie in the present case Kirchhoff's knot rule, these errors can generally be caused by a continuous mapping or function ${\text{f}}_H:{\text{i}}_m^{}\mapsto \text{i}$

can be modeled for which the following applies: $${\text{i}}_s\in H\Rightarrow {\text{f}}_H\left({\text{i}}_s\right)\in H$$

This function corresponds to the first function within the meaning of the claims or the function f1 out 2 . In the embodiment described here, a linear model is used for f _h , namely $${\text{f}}_H\left({\text{i}}_s\right)=\left(1+d_H\right){\text{i}}_s+{\text{O}}_H.$$

Here d _h ∈ ℝ is a drift scalar and o _h ∈ H is an offset vector. The following can be derived: $${\text{i}}_s,{\text{O}}_H\in H\Rightarrow 1^T{\text{i}}_s=1^T{\text{O}}_H=0\Rightarrow \left(1+d_H\right)1^T{\text{i}}_s+1^T{\text{O}}_H=0\Rightarrow 1^T{\text{f}}_H\left({\text{i}}_s\right)=0\Rightarrow {\text{f}}_H\left({\text{i}}_s\right)\in H.$$

im modelliert werden. Diese Funktion unterliegt jedoch nicht den Einschränkungen gemäß obiger Gleichung (10), da der verrauschte Vektor im nicht die Kirchhoff'sche Knotenregel erfüllen muss. Die Funktion f_m entspricht einer zweiten Funktion im Sinne der Patentansprüche bzw. der Funktion f2 aus 2. Die Funktion f_m wird im hier beschriebenen Ausführungsbeispiel als folgende lineare Funktion modelliert: $${\text{f}}_m\left(\text{i}\right)=\left(\text{I}+{\text{D}}_m\right)\text{i}+{\text{o}}_m$$

In a similar way, measurement errors can be expressed as a linear mapping or function ${\text{f}}_m:\text{i}\mapsto {\text{i}}_m$

to be modeled. However, this function is not subject to the restrictions according to equation (10) above, since the noisy vector im does not have to fulfill Kirchhoff's knot rule. The function f _m corresponds to a second function within the meaning of the claims or the function f2 out 2 . In the _{exemplary embodiment described here, the function f m} is modeled as the following linear function: $${\text{f}}_m\left(\text{i}\right)=\left(\text{I.}+{\text{D.}}_m\right)\text{i}+{\text{O}}_m$$

bezeichnet eine diagonale Drift-Matrix und ${\text{o}}_m\in {ℝ}^K$

ist ein Offset-Vektor. Ziel der hier beschriebenen Ausführungsform ist es dabei, die Parameter ${\text{D}}_m\in {ℝ}^{K\times K}{\text{ und o}}_m\in {ℝ}^K$

zu lernen und die inverse Funktion ${\text{f}}_m^{-1}\left({\text{i}}_m\right)$

zu ermitteln. Wie bereits oben dargelegt, können mit dieser inversen Funktion dann entsprechende gemessene Stromwerte geeignet korrigiert werden. Die Parameter D_m und o_m werden dabei mit Hilfe der Tupel (i_m, i_s ) gelernt.Where I is the identity matrix, ${\text{D.}}_m\in {ℝ}^{K\times K}$

denotes a diagonal drift matrix and ${\text{O}}_m\in {ℝ}^K$

is an offset vector. The aim of the embodiment described here is the parameters ${\text{D.}}_m\in {ℝ}^{K\times K}{\text{and o}}_m\in {ℝ}^K$

to learn and the inverse function ${\text{f}}_m^{-1}\left({\text{i}}_m\right)$

to determine. As already explained above, corresponding measured current values can then be appropriately corrected with this inverse function. The parameters D _m and o _m are determined with the help of the tuples (i _m , i _s ) learned.

The following relationship results from the above equations (11) and (12): $$\begin{array}{c}{\text{i}}_m^{}={\text{f}}_m\circ {\text{f}}_H\left({\text{i}}_s\right)=\left(\text{I.}+{\text{D.}}_m\right)\left(\left(1+d_H\right){\text{i}}_s+{\text{O}}_H\right)+{\text{O}}_m\\ =\left(\text{I.}+{\text{D.}}_m\right)\left(1+d_H\right){\text{i}}_s+\left(\text{I.}+{\text{D.}}_m\right){\text{O}}_H+{\text{O}}_m\end{array}$$

In general, the parameters (especially D _m and o _m ) cannot be derived from the tuples in the form (i _m , i _s ) (and in particular also cannot be determined from the above tuples (i _sim , ĩ _s . In order to solve this problem, a distinction is made between the above cases of class 3 and class 4 in the embodiment described here.

If a classification or classification in class 3 has been carried out, there are no hardware errors, so that the above parameters d _h and o _h can be set to 0. This results in the relationship i _m = (I + D _m ) i _s + o _m from equation (13). In this case, the parameters D _m and o _m can be determined in a simple manner by a linear regression of the predicted current values i _sim (t) on the _{mean current values i s} (t). In other words, a total of K regressions are carried out, with a regression of the values i _{sim, k} (t) on the values i _{m, k} (t) being carried out for each k. The vector formed by the parameters D _m and o _m corresponds to in 2 the parameter vector PV ' with parameter values pw ' .

If the above class 4 is present, the problem cannot be clearly solved without further knowledge. For example, if the vector (d _h , o _h , D _m , o _m ) satisfies the above equation (13), then this also applies to the vector (d _h , o _h + e, D _m , o _m - (I + D _m ) e) as long as e E ℌ holds. The above vector (d _h , o _h , D _m , o _m ) corresponds to in 2 the parameter vector PV with corresponding parameter values pw1 (ie, values of d and _h o _h) and parameter values pw2 (ie, values of D _m and o _m ). In order to come to a clear solution in the case of a classification in class 4, the following assumption is used.

Assumption 2: The vector x, which contains all parameters (d _h , o _h , D _m , o _m ), is a sparsely populated vector, ie a vector that contains only a few parameter values that are not equal to 0.

This assumption is based on the knowledge that it is very unlikely that all or at least many sources of error are present in a motor vehicle at the same time. The presence of all sources of error at the same time would mean that all K sensors that are used for current measurement have offset or drift errors or both. Furthermore, K streams would be affected by hardware errors without exception. The assumption that there are only relatively few sources of error in the same vehicle at one point in time is therefore plausible.

Using the above assumption, the parameter vector x, which is the parameter vector PV the 2 can be determined as follows. First, the expressions (I + D _m ) (1 + d _h ) and (I + D _m ) o _h + o _m are estimated by the tuples (i _sim (t), ĩ _s (t)). In this estimation, a linear regression from i _sim (t) to i _s (t) is carried out. In the following, the k-th diagonal element of D _{m will be} referred to as d _{m, k} and the k-th elements of o _h and o _{m will be} referred to as o _{h, k} and o _{m, k} , respectively. For every k ∈ {1,2, ..., K} the following relationship arises: $$\begin{array}{l}\left(1+d_{m,k}\right)\left(1+d_H\right)\approx a_k\\ \left(1+d_{m,k}\right){\text{O}}_{H,k}+{\text{O}}_{m,k}\approx b_k\end{array}$$

Here, a _k and b _k denote estimates for the slope or the intercept.

These estimates are calculated by the regression of (i _{sim, k} (1), ..., i _{sim, k} (T _ud )) ^T on (ĩ _{s, k} (1), ..., h _{s, k} (T _ud )) ^T received. Based on the definitions y = (a ₁ , ..., a _K , f, b ₁ , ..., b _K ) ^T and x̃ = (d _h , o _h , ₁ ..., o _h , _K , d _{m, 1} , ..., d _{m, K} , o _{m, 1} , ..., 0 _{m, K} , 1) ^T ∈ ℝ ^{3K + 2} , each of the estimated terms can be put in a square form x̃ ^T Qx̃ of x̃, for example in the following form: $$\begin{array}{c}\left(1+d_{m,k}\right)\left(1+d_H\right)={\left({\tilde{\text{x}}}^T{\text{e}}_{3K+2}\right)}^2+{\tilde{\text{x}}}^T{\text{e}}_{3K+2}{\text{e}}_{K+1+k}^T\tilde{\text{x}}+{\tilde{\text{x}}}^T{\text{e}}_{3K+2}{\text{e}}_1^T\tilde{\text{x}}+{\tilde{\text{x}}}^T{\text{e}}_{K+1+k}{\text{e}}_1^T\tilde{\text{x}}\\ ={\tilde{\text{x}}}^T\left({\text{e}}_{3K+2}{\text{e}}_{3K+2}^T+{\text{e}}_{3K+2}{\text{e}}_{K+1+k}^T+{\text{e}}_{3K+2}{\text{e}}_1^T+{\text{e}}_{K+1+k}^{}{\text{e}}_1^T\right)\tilde{\text{x}}\\ :={\tilde{\text{x}}}^T{\text{Q}}_k\tilde{\text{x}},es\end{array}$$

The notation e _{3K + 2} denotes the canonical vector in relation to 3K + 2. In a similar way one obtains (1 + d _{m, k} ) o _{h, k} + o _{m, k} = x̃ ^T Q _{K + k} x̃ and thus equation (14) can be written as follows: $$\forall i\in \left\{\mathrm{1.2,}...\mathrm{,\; 2}K\right\}{\tilde{\text{x}}}^T{\text{Q}}_i\tilde{\text{x}}\approx {\text{y}}^T{\text{e}}_i.$$

Finally, the parameter vector x (or x̃) is obtained by solving the following optimization problem: $$\begin{array}{l}\underset{\tilde{\text{x}}}{{\displaystyle \text{min}}}{\Vert \tilde{\text{x}}\Vert }_1\text{under the condition}{\displaystyle \sum _{i=1}^{2K}{\left({\tilde{\text{x}}}^T{\text{Q}}_i\tilde{\text{x}}-{\text{y}}^T{\text{e}}_i\right)}^2}\le \epsilon ,{\tilde{\text{x}}}^T{\text{e}}_{3K+2}=\mathrm{1,}\\ {\tilde{\text{x}}}^T\left({\displaystyle \sum _{i=1}^{\mathrm{<figure-callout\; id="1"\; label="K\; e"\; filenames="DE102019135022A1\_0063.png"\; state="\{\{state\}\}">K</figure-callout>}}{\text{e}}_{}{}_{1+i}}\right)=0\end{array}$$

stellt sicher, dass sich die Hardware-Offsets o_h,1, ...,o_h,K zu Null summieren, um die Gleichung (10) zu erfüllen. Das Optimierungsproblem aus Gleichung (17) stellt eine Minimierung der Spärlichkeit des Vektors x dar. Mit anderen Worten hat diese Optimierung das Optimierungsziel, dass der Vektor x eine möglichst geringe Anzahl von Parameterwerten ungleich 0 aufweist. Dieses Optimierungsproblem kann beispielsweise mit dem Algorithmus „Quadratic Basic Pursuit“ der Druckschrift [2] gelöst werden. Dieser Algorithmus wandelt das obige Problem in ein konvexes Problem, so dass es rechnergestützt gelöst werden kann.Here, ∈ denotes an error threshold. The constraint x̃ ^T e _{3K + 2} = 1 ensures that the last element of x̃ takes the value 1. The condition ${\tilde{\text{x}}}^T\left({\displaystyle \sum {}_{i=1}^{\mathrm{<figure-callout\; id="1"\; label="K\; e"\; filenames="DE102019135022A1\_0063.png"\; state="\{\{state\}\}">K</figure-callout>}}{\text{e}}_{}{}_{1+i}}\right)=0$

ensures that the hardware offsets o _{h, 1} , ..., o _{h, K} add up to zero to satisfy equation (10). The optimization problem from equation (17) represents a minimization of the sparseness of the vector x. In other words, the optimization goal of this optimization is that the vector x has the smallest possible number of non-zero parameter values. This optimization problem can be solved, for example, with the “Quadratic Basic Pursuit” algorithm in the publication [2]. This algorithm converts the above problem into a convex problem so that it can be solved with the aid of a computer.

The overall result thus obtained is corresponding parameter vectors for the classifications in class 3 and in class 4, which are then sent by the backend server SE to the motor vehicle in question VE are transmitted and used there by inverting the corresponding function to correct current measured values.

In a modified embodiment, the above classification is omitted. Instead, the algorithm for class 4, which determines the parameter vector (d _h , o _h , D _m , o _m ), is always executed. In other words, the above optimization based on equation (17) can always be carried out, regardless of whether there are hardware errors or not. Without hardware errors, the optimization then delivers the result that the quantities d _h and o _h essentially assume the value zero. Nevertheless, the result is improved when the above classification is carried out.

The inventors tested the method described above using suitable measurement data. It has been shown that the method reliably determines the corresponding parameter vectors and is therefore very well suited for correcting measurement errors of corresponding current values in a motor vehicle.

The embodiments of the invention described above have a number of advantages. In particular, using a fleet-based approach, the learned data-driven models of several motor vehicles can be compared with one another in order to identify irregularities in the measurement of current values in a high-voltage electrical system. In addition, suitable parameter values are determined which can then be used in the respective motor vehicle to correct measurement errors for the corresponding measured current values.

List of reference symbols

BN

High-voltage electrical system

HVB

High-voltage battery

EM

electric machine

HE

heater

CO

air conditioning

IN

Converter

i1, i2, ..., i5

electrical currents

MO, MO '

data-driven models

VE, VE '

Motor vehicles

LE

Learning process

MD

Measurement datasets

MUD

Model run measurement series

UD

Model ride

u1, u2, u3

Input variables

is

Current mean values

CL1, CL2, CL3, CL4

Classes

f1, f2

Functions

PV, PV '

Parameter vectors

pw1, pw2, pw '

Parameter values

S1, S2

steps

SE

Backend server

H

Hyperplane ${\text{i}}_{\text{s}}{\text{, i}}^{\text{j}}{\text{, i}}_{\text{j}}^{\text{m}}\text{,}{\mathrm{\Pi }}_H\left(i_m^j\right)$

Current vectors

HE, ME

Displacements of current vectors

bibliography

• [1] Pfeiffer, J .; Wolf, P .; Pereira, R. A Fleet-Based Machine Learning Approach for Automatic Detection of Deviations between Measurements and Reality. 2019 IEEE Intelligent Vehicles Symposium (IV); IEEE Paris, France, 2019; Pages 2086-2092 . Loi: 10.1109 / IVS.2019.8813858.
• [2] Ohlsson, H .; Yang, AY; Dong, R .; Verhaegen, M .; Sastry, SS Quadratic Basis Pursuit. arXiv: 1301.7002 [cs, math, stat] 2013.arXiv: 1301.7002 .

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Non-patent literature cited

• Pfeiffer, J .; Wolf, P .; Pereira, R. A Fleet-Based Machine Learning Approach for Automatic Detection of Deviations between Measurements and Reality. 2019 IEEE Intelligent Vehicles Symposium (IV); IEEE Paris, France, 2019; Pages 2086-2092 [0088]
• Ohlsson, H .; Yang, A.Y .; Dong, R .; Verhaegen, M .; Sastry, S.S. Quadratic base pursuit. arXiv: 1301.7002 [cs, math, stat] 2013.arXiv: 1301.7002 [0088]

Claims (14)

Method for the computer-aided evaluation of measurements of electrical currents (i1, i2, ..., i5) in a high-voltage on-board network (BN) of a given electrically powered motor vehicle (VE), the high-voltage on-board network (BN) having a plurality of electrical components ( HVB, EM, HE, CO, IN) comprising an electrical machine (EM) for driving the specified motor vehicle (VE) and an electrical energy store (HVB) for supplying electrical energy to the electrical machine (EM) and each electrical current (i1, i2 , ..., i5) flows via the same node (N) in the high-voltage on-board network (BN), where: a) a first data-driven model (MO) is learned in the specified motor vehicle (VE) using a machine learning process (LE) and the learned first data-driven model (MO) is transmitted to a backend server (SE) by the specified motor vehicle (VE) is learned, the first data-driven model (MO) based on a plurality of measurement data sets (MD), each measured during operation of the specified motor vehicle (VE) at different operating times, a respective measurement data set (MD) for a respective operating time Includes measured values of input variables (u1, u2, u3) and measured values of electrical currents (i1, i2, ..., i5), with the learned first data-driven model (MO) based on measured values of the input variables (u1, u2, u3) Measured values of the electrical currents (i1, i2, ..., i5) can be predicted; b) in the backend server (SE) the learned first data-driven model (MO) and a plurality of further learned second data-driven models (MO '), each of which is in a different motor vehicle (VE') of the same type as the specified motor vehicle (VE) with the same machine learning method (LE) as the first data-driven model (MO) based on a large number of measurement data sets measured in the respective other motor vehicle (VE ') for the same input variables (u1, u2, u3) and the same electrical currents (i1, i2, ..., i5) how the first data-driven model (MO) were learned, are processed by, for a model driving measurement series (MUD), which measured values for the learning of the first data-driven model (u1, u2, u3) contains input variables used for a large number of model driving operating times within a given model driving, current mean values (is) for the electrical currents (i1, i2, ..., i5) for each model driving operation can be determined at the time of operation, with a respective current mean value (is) for a respective electrical current (i1, i2, ..., i5) and for a respective model drive operating time the mean value of the measured values of the respective electrical current (i1, i2, ... , i5) which are predicted by the first learned data-driven model (MO) and the plurality of second learned data-driven models (MO ') based on the measured values of the input variables (u1, u2, u3) for the respective model drive operating time; c) at least in the event that, as a function of the measurement data records (MD) measured in the specified motor vehicle (VE), an abnormal operating behavior of the specified motor vehicle (VE) and the presence of measurement errors when measuring the electrical currents (i1, i2, ... , i5) is determined in the specified motor vehicle, by means of the measured values of the respective electrical currents (i1, i2, ..., i5) and the mean current values (is) for the respective electrical currents (i1 , i2, ..., i5) a parameter vector (PV) is determined from parameter values (pw1, pw2) for parameters of a first function (f1) and from parameter values (pw2) for parameters of a second function (f2), the first function (f1) Current values of the electrical currents (i1, i2, ..., i5) with normal operating behavior of the specified motor vehicle (VE) to current values of the electrical currents (i1, i2, ..., i5) with abnormal operating behavior of the specified Motor vehicle (VE) and where the second function (f2) current values of the electrical currents (i1, i2, ..., i5) in the absence of measurement errors on current values of the electrical currents (i1, i2, ..., i5) in the presence of measurement errors in the specified motor vehicle (VE), the parameter values (pw1, pw2) being determined by means of an optimization with the optimization goal of a minimum number of parameter values (pw1, pw2) not equal to zero in the parameter vector (PV), the parameter values (pw2) the parameters of the second function (f2) are transmitted to the specified motor vehicle (VE) for correcting future measured values of at least part of the electrical currents (i1, i2, ..., i5). Procedure according to Claim 1 , characterized in that, depending on the measurement data records (MD) measured in the specified motor vehicle (VE), a classification is made into the following classes before step c) is carried out: a first class (CL1), according to which neither an abnormal operating behavior of the specified In the motor vehicle (VE) there are still measurement errors for the measured values of the electrical currents (i1, i2, ..., i5) in the specified motor vehicle (VE); - A second class (CL2), according to which there is an abnormal operating behavior of the specified motor vehicle (VE), but no measurement errors for the measured values of the electrical currents (i1, i2, ..., i5) in the specified motor vehicle (VE) are present; - a third class (CL3) according to which there is no abnormal operating behavior of the specified motor vehicle (VE), but measurement errors for the measured values of the electrical currents (i1, i2, ..., i5) in the specified motor vehicle (VE) are present; - A fourth class (CL4), according to which there is an abnormal operating behavior of the specified motor vehicle (VE) and measurement errors for the measured values of the electrical currents (i1, i2, ..., i5) in the specified motor vehicle (VE) are present; wherein the method is terminated if the first class is present and if the second class is present. Procedure according to Claim 2 , characterized in that the classification is based on first features and a second feature, each first feature being assigned to an electrical current (i1, i2, ..., i5) and a measure of the absolute deviation of the data-driven model learned by the first learned ( MO) is the predicted measured values of the assigned electric current (i1, i2, ..., i5) from the current mean values (is) for the model drive operating times and the second feature is a measure of the size of the measured data records (MD) determined Sums of the measured values of the electrical currents (i1, i2, ..., i5) for the time of operation of the corresponding measurement data set (MD). Procedure according to Claim 3 , characterized in that the classification is carried out using a threshold value classifier, an abnormal operating behavior of the specified motor vehicle (VE) being present if a first value, which becomes larger with increasing size of the first features and the smaller with increasing size of the second feature, the smaller exceeds a first threshold value, and measurement errors for the measured values of the electrical currents (i1, i2, ..., i5) in the specified motor vehicle (VE) are then present when a second value, which increases with the size of the second feature, the greater exceeds a second threshold value. Method according to one of the Claim 3 , characterized in that the classification is carried out using a machine-learned decision tree classifier, which is learned by means of training data sets comprising first features and a second feature as well as an associated class from the first to fourth class. Method according to one of the Claims 2 to 5 , characterized in that in the event that the third class (CL3) is present, instead of the parameter vector (PV) of step c) a reduced parameter vector (PV ') from parameter values (pw') compared to the parameter vector (PV) of step c) for the parameters of the second function (f2) is determined, the parameter values (pw ') via a regression of the measured values of the electrical currents (i1, i2, ..., i5) predicted by the first data-driven model (MO) on the mean current values (is) can be determined for the model drive operating times, the parameter values (pw ') of the parameters of the second function (f2) being applied to the specified motor vehicle (VE) for correcting future measured values of at least part of the electrical currents (i1, i2,. .., i5). Method according to one of the preceding claims, characterized in that the input variables (u1, u2, u3) include an acceleration of the specified motor vehicle (VE) and a speed of the specified motor vehicle (VE), the input variables (u1, u2, u3) preferably Furthermore, the inclination of the specified motor vehicle (VE) relative to the horizontal and / or the gear engaged in the specified motor vehicle (VE) and / or a power requirement and / or allocation to the electrical machine (EM) and / or a speed of the electrical machine ( EM). Method according to one of the preceding claims, characterized in that the high-voltage on-board network (BN), in addition to the electrical machine (EM) and the electrical energy store (HVB), a heater (HE) and / or an air conditioning system (CO) and / or a converter (IN) for voltage conversion for a low-voltage vehicle electrical system. Method according to one of the preceding claims, characterized in that the first data-driven model (MO) and the plurality of second data-driven models (MO ') each represent a neural network structure, in particular an LSTM network structure. Method according to one of the Claims 1 to 8th , characterized in that the first data-driven model (MO) and the plurality of second data-driven models (MO ') each represent a state space model. System for the computer-aided evaluation of measurements of electrical currents (i1, i2, ..., i5) in a high-voltage electrical system (BN) of a given electrically powered motor vehicle (VE), the High-voltage on-board network (BN) a plurality of electrical components (HVB, EM, HE, CO, IN) comprising an electrical machine (EM) for driving the specified motor vehicle (VE) and an electrical energy store (HVB) for supplying electrical energy to the electrical machine (EM) and each electrical current (i1, i2, ..., i5) flows via the same node (N) in the high-voltage on-board network (BN), the system being set up to carry out a method in which: a) a first data-driven model (MO) is learned in the specified motor vehicle (VE) using a machine learning process (LE) and the learned first data-driven model (MO) is transmitted to a backend server (SE) by the specified motor vehicle (VE), wherein the first data-driven model (MO) is learned based on a plurality of measurement data sets (MD) which were each measured during operation of the specified motor vehicle (VE) at different times of operation, wherein a respective measurement data set (MD) comprises measured values of input variables (u1, u2, u3) and measured values of the electrical currents (i1, i2, ..., i5) for a respective operating time, based on the learned first data-driven model (MO) Measured values of the input variables (u1, u2, u3) Measured values of the electrical currents (i1, i2, ..., i5) can be predicted; b) in the backend server (SE) the learned first data-driven model (MO) and a plurality of further learned second data-driven models (MO '), each of which is in a different motor vehicle (VE') of the same type as the specified motor vehicle (VE) with the same machine learning method (LE) as the first data-driven model (MO) based on a large number of measurement data sets measured in the respective other motor vehicle (VE ') for the same input variables (u1, u2, u3) and the same electrical currents (i1, i2, ..., i5) how the first data-driven model (MO) were learned, are processed by, for a model driving measurement series (MUD), which measured values for the learning of the first data-driven model (u1, u2, u3) contains input variables used for a large number of model driving operating times within a given model driving, current mean values (is) for the electrical currents (i1, i2, ..., i5) for each model driving operation can be determined at the time of operation, with a respective current mean value (is) for a respective electrical current (i1, i2, ..., i5) and for a respective model drive operating time the mean value of the measured values of the respective electrical current (i1, i2, ... , i5), which are predicted by the first learned data-driven model (MO) and the plurality of second learned data-driven models (MO ') based on the measured values of the input variables (u1, u2, u3) for the respective model drive operating time; c) at least in the event that, depending on the measurement data records (MD) measured in the specified motor vehicle (VE), an abnormal operating behavior of the specified motor vehicle (VE) and the presence of measurement errors when measuring the electrical currents (i1, i2, ... , i5) is determined in the specified motor vehicle, by means of the measured values of the respective electrical currents (i1, i2, ..., i5) and the mean current values (is) for the respective electrical currents (i1 , i2, ..., i5) a parameter vector (PV) is determined from parameter values (pw1, pw2) for parameters of a first function (f1) and from parameter values (pw2) for parameters of a second function (f2), the first function (f1) Current values of the electrical currents (i1, i2, ..., i5) with normal operating behavior of the specified motor vehicle (VE) to current values of the electrical currents (i1, i2, ..., i5) with abnormal operating behavior of the specified Motor vehicle (VE) and where the second function (f2) current values of the electrical currents (i1, i2, ..., i5) in the absence of measurement errors on current values of the electrical currents (i1, i2, ..., i5) in the presence of measurement errors in the specified motor vehicle (VE), the parameter values (pw1, pw2) being determined by means of an optimization with the optimization goal of a minimum number of parameter values (pw1, pw2) not equal to zero in the parameter vector (PV), the parameter values (pw2) the parameters of the second function (f2) are transmitted to the specified motor vehicle (VE) for correcting future measured values of at least part of the electrical currents (i1, i2, ..., i5). System according to Claim 11 , characterized in that the system for performing a method according to one of the Claims 2 to 10 is designed. Backend server acting as a backend server (SE) for a method according to one of the Claims 1 to 10 is set up, wherein the backend server (SE) is designed in such a way that - in the backend server (SE) a learned first data-driven model (MO) for a given motor vehicle (VE) as well as a plurality of further learned second data-driven models ( MO '), each in a different motor vehicle (VE') of the same design as the specified motor vehicle (VE) with the same machine learning method (LE) as the first data-driven model (MO) based on a large number of in the respective other motor vehicle (VE ') measured data sets for the same input variables (u1, u2, u3) and the same electrical currents (i1, i2, ..., i5) as the first data-driven model (MO) were learned, are processed by for a Model trip measurement series (MUD), which measured values for the input variables used when learning the first data-driven model (u1, u2, u3) for a large number of model trip operating times within a given contains a model drive, Current mean values (is) for the electrical currents (i1, i2, ..., i5) are determined for each model drive operating time, with a respective current mean value (is) for a respective electrical current (i1, i2, ..., i5) and the mean value of the measured values of the respective electrical current (i1, i2, ..., i5), which is determined by the first learned data-driven model (MO) and the plurality of second learned data-driven models (MO '), is for a respective model travel operating time. are predicted based on the measured values of the input variables (u1, u2, u3) for the respective model drive operating time; - at least in the event that, depending on the measurement data records (MD) measured in the specified motor vehicle (VE), an abnormal operating behavior of the specified motor vehicle (VE) and the presence of measurement errors when measuring the electrical currents (i1, i2, ..., i5) is determined in the specified motor vehicle, by means of the measured values of the respective electrical currents (i1, i2, ..., i5) and the mean current values (is) for the respective electrical currents (i1, i2, ..., i5) a parameter vector (PV) is determined from parameter values (pw1, pw2) for parameters of a first function (f1) and from parameter values (pw2) for parameters of a second function (f2), the first function ( f1) Current values of the electrical currents (i1, i2, ..., i5) with normal operating behavior of the specified motor vehicle (VE) to current values of the electrical currents (i1, i2, ..., i5) with abnormal operating behavior of the specified Motor vehicle (VE) and where the second function (f2) current values of the electrical currents (i1, i2, ..., i5) in the absence of measurement errors on current values of the electrical currents (i1, i2, ..., i5) in the presence of measurement errors in the specified motor vehicle (VE), the parameter values (pw1, pw2) being determined by means of an optimization with the optimization goal of a minimum number of parameter values (pw1, pw2) not equal to zero in the parameter vector (PV), the parameter values (pw2) the parameters of the second function (f2) are transmitted to the specified motor vehicle (VE) for correcting future measured values of at least part of the electrical currents (i1, i2, ..., i5). Electrically powered motor vehicle that is used as a predetermined motor vehicle (VE) for a method according to one of the Claims 1 to 10 is set up, wherein the motor vehicle (VE) is designed in such a way that - in the motor vehicle (VE) a first data-driven model (MO) is learned with a machine learning method (LE) and the learned first data-driven model (MO) is sent to a backend- Server (SE) is transmitted by the specified motor vehicle (VE), the first data-driven model (MO) being learned based on a large number of measurement data sets (MD), which were each measured during operation of the specified motor vehicle (VE) at different operating times , wherein a respective measurement data record (MD) for a respective operating time comprises measured values of input variables (u1, u2, u3) and measured values of the electrical currents (i1, i2, ..., i5), with the learned first data-driven model (MO) based on measured values of the input variables (u1, u2, u3) measured values of the electrical currents (i1, i2, ..., i5) can be predicted; - the motor vehicle receives parameter values (pw2) of parameters of a second function (f2) and corrects measured values of at least part of the electrical currents (i1, i2, ..., i5) with the received parameter values (pw2).

Images