阅读说明：本技术 用于降低具有内燃机的车辆的驱动系统的废气排放的方法 (Method for reducing exhaust gas emissions of a drive system of a vehicle having an internal combustion engine ) 是由 H.马克特 S.安格迈尔 于 2020-04-15 设计创作，主要内容包括：本发明涉及一种用于降低具有内燃机的车辆的驱动系统的废气排放的方法,该方法具有以下步骤：通过计算机实现的机器学习系统产生多个第一行驶走势,其中,第一行驶走势的统计分布取决于在行驶运行中测量的第二行驶走势的统计分布；借助于车辆的或车辆的驱动系统的计算机实现的建模计算用于第一行驶走势的相应的废气排放；根据所计算的废气排放中的至少一种来匹配车辆的驱动系统,其中,根据所计算的废气排放中的至少一种的程度或走势以及根据相应的第一行驶走势的统计频度来进行所述匹配,其中,借助于第一行驶走势的统计分布来测定相应的第一行驶走势的统计频度。(The invention relates to a method for reducing exhaust gas emissions from a drive train of a vehicle having an internal combustion engine, comprising the following steps: generating a plurality of first driving gestures by a computer-implemented machine learning system, wherein a statistical distribution of the first driving gestures depends on a statistical distribution of second driving gestures measured during driving operation; calculating a corresponding exhaust emission for the first driving profile by means of computer-implemented modeling of the vehicle or of a drive system of the vehicle; the drive system of the vehicle is adapted according to at least one of the calculated exhaust emissions, wherein the adaptation is performed according to the degree or the behavior of the at least one of the calculated exhaust emissions and according to the statistical frequency of the respective first driving behavior, wherein the statistical frequency of the respective first driving behavior is determined by means of a statistical distribution of the first driving behavior.)

1. A method for reducing exhaust emissions of a drive system of a vehicle having an internal combustion engine, characterized by the steps of:

-generating a plurality of first driving gestures (51) by means of a computer-implemented machine learning system, wherein a statistical distribution of the first driving gestures depends on a statistical distribution of second driving gestures measured during driving operation,

-calculating a respective exhaust emission (52) for a first driving profile by means of computer-implemented modeling of the vehicle or of a drive system of the vehicle,

-matching the drive system of the vehicle (53) according to at least one of the calculated exhaust emissions, wherein the matching is performed according to the degree or the behavior of the at least one of the calculated exhaust emissions and according to the statistical frequency of the respective first driving tendency, wherein the statistical frequency of the respective first driving tendency is determined by means of a statistical distribution of the first driving tendency.

2. A method according to claim 1, characterised in that the first and second driving profiles represent a physical or technical characteristic of the vehicle's power train that can be measured with sensors, which physical or technical characteristic characterizes the continued movement of the vehicle.

3. Method according to any one of the preceding claims, characterized in that the first and second running tendency represent a speed tendency, a tendency of a position of an accelerator pedal, a tendency of a position of a clutch pedal, a tendency of a position of a brake pedal or a tendency of a transmission ratio.

4. Method according to one of the preceding claims, characterized in that a first travel profile is generated for the respectively associated first travel route.

5. Method according to claim 4, characterized in that the respective exhaust gas emissions for the first driving tendency are calculated as a function of the first driving tendency and/or the associated first driving route.

6. The method according to one of claims 4 or 5, characterized in that the statistical distribution of the first driving tendency comprises a statistical distribution of the first driving tendency with the respective associated first driving route, and the statistical distribution of the second driving tendency measured during the driving operation comprises a statistical distribution of the second driving tendency with the respective associated second driving route measured during the driving operation.

7. The method of claim 5, wherein the first travel route is generated by the computer-implemented machine learning system or by another computer-implemented machine learning system.

8. The method according to any one of claims 4 to 7, characterized in that the first travel route comprises route characteristics, in particular geographical characteristics, traffic flow characteristics, lane characteristics, traffic guidance characteristics and/or weather characteristics of the route.

9. The method of any preceding claim, wherein the machine learning system learns using computer-implemented training, the training comprising the steps of:

-selecting a first driving route from a first database with driving routes,

a generator of the machine learning system obtains a first travel route as an input variable and generates a corresponding associated first travel profile for the first travel route,

the second database stores the travel routes and the associated travel patterns detected during the travel operation,

selecting a second travel route and a corresponding associated second travel pattern detected during the travel operation from a second database,

the discriminator of the machine learning system obtains as input variables a pairing comprising one of the first travel routes having the respective associated generated first travel gesture and a pairing comprising a second travel route having the respective associated second travel gesture detected during the travel operation,

the discriminator calculates, as a function of the input variables, an output which characterizes each pairing obtained as an input variable as a pairing with the generated first driving behavior and also as a pairing with the second driving behavior detected during driving operation,

optimizing an objective function, which represents the distance between the paired distribution with the generated first driving behavior and the paired distribution with the second driving behavior detected during the driving operation, as a function of the output of the discriminator.

10. The method of any one of claims 1 to 9, wherein the machine learning system learns using computer-implemented training, the training comprising the steps of:

a generator of the machine learning system obtains the first random vector as an input variable and generates a first travel route and an associated first travel profile for the first random vector,

the database stores the travel routes and the associated travel patterns detected during the travel operation,

selecting a second travel route and a corresponding associated second travel pattern detected during the travel operation from the database,

the evaluator of the machine learning system receives as input variables a first pairing of the generated first travel route and the associated generated first travel gesture and a second pairing of the second travel route and the associated second travel gesture detected during the travel operation,

the discriminator calculates, as a function of the input variables, an output which characterizes each pairing obtained as an input variable by a first pairing of a first generated travel route and a corresponding associated first generated travel gesture and by a second pairing of a second travel route and a corresponding associated second detected travel gesture during travel operation,

-optimizing an objective function representing a distance between the distribution of the first pair and the distribution of the second pair based on the output of the discriminator.

11. Method according to one of claims 9 or 10, characterized in that the parameters of the machine learning system are adapted according to an optimization of the objective function such that

-the discriminator is optimized for discriminating between the generated first driving tendency and the second driving tendency detected during the driving operation,

the generator is optimized for generating a first generated driving profile with a first distribution, which is as difficult as possible to distinguish from a second driving profile, which is present with a second distribution and is detected during driving operation, by means of the discriminator.

12. The method of claim 11, wherein parameters of the machine learning system are matched according to a gradient of the objective function.

13. Method according to any one of claims 9 to 12, characterized in that the statistical distance between the first distribution of the first running tendency and the second distribution of the second running tendency is achieved as an objective function.

14. Method according to one of claims 9 to 13, characterized in that the generator and/or the discriminator are each implemented as a neural network, in particular as a recurrent neural network.

15. Method according to any of the preceding claims, characterized in that the modelling of the vehicle or of the drive system of the vehicle comprises a model of the combustion engine and/or a model of the exhaust gas aftertreatment system of the combustion engine and/or a model of the engine controller and/or a model of the combustion process.

16. Method according to any of the preceding claims, characterized in that the matching is performed by matching the topology or components of the drive system, by matching the control functions of the software provided in the drive system, by matching parameters in the application of the drive system or by matching control variables in the operation of the drive system in the vehicle.

17. Method according to any one of the preceding claims, characterized in that the second driving tendency is selected from a number of third driving tendencies.

18. Method according to claim 17, characterized in that the second driving tendency is selected such that the distribution of the second driving tendency corresponds to the distribution of the actual journey.

19. A computer program arranged to perform the method according to any one of the preceding claims.

20. A storage medium readable by machine, on which a computer program according to claim 19 is stored.

Technical Field

The present invention relates to a method for reducing exhaust gas emissions of a drive system of a vehicle having an internal combustion engine and to a computer program provided for this purpose.

Background

DE 102017107271 a1 discloses a method for determining the prevailing driving cycle of a driving test for determining the exhaust emissions of a motor vehicle. Here, the speed profiles for the different driving cycles are derived on the basis of the parameter set. The aim is to determine a dominant cycle which reflects as much as possible the "maximum" emission within given boundary conditions.

Disclosure of Invention

In fact, currently, the determination of the emissions of the drive system is mainly carried out in two stages:

1) one or more representative driving cycles are determined which are intended to describe the driving behavior of all vehicles. Such a driving cycle should also map driving situations that are demanding for the drive system (e.g. including parts with high dynamics and/or multiple starting processes). This also includes, for example, the Test cycles specified by legislators, such as WLTP (world Light automobile Test protocol).

2) The optimization and verification of the drive system takes place with the aid of these driving cycles. The vehicle is provided with corresponding measuring equipment and the measurements are carried out according to the test protocol determined in 1). The optimization of the system topology, regulatory functions and parameters is essentially performed at steady state of the drive system and according to these cycles.

However, it is not possible to provide a drive system which, in terms of its exhaust emissions, meets the high requirements with minimal environmental impact in real driving operation. The verification results are only of limited interest, since the entire operating state space of the drive system is only covered very randomly and firstly the statistical distribution of the operating states is incorrectly represented by these driving cycles.

This results in the risk of not complying with legislation on exhaust gas emissions, on the one hand, and the drive system is optimized in terms of emissions not as a whole for all strokes and taking into account the corresponding frequency.

In some countries, legislation stipulates that new motor vehicles driven by means of an internal combustion engine are permitted on the basis of the emissions generated during actual driving operation. For this reason, the english name "real driving emissions" is also common. Such motor vehicles include, for example, those driven solely by an internal combustion engine, but also those having a hybrid drive train.

For this purpose, the test personnel test (bestiente) one or more driving cycles with a motor vehicle and measure the emissions occurring there. The license of the motor vehicle is then correlated with these measured emissions. In this case, the driving cycle can be freely selected by the test person within wide limits. A typical duration of the driving cycle may be, for example, 90 to 120 minutes.

The challenge for the manufacturers of motor vehicles in the development of motor vehicles is therefore to foresee in advance during the development of new motor vehicles whether the emissions of the motor vehicle remain within the legally prescribed limits for each permitted driving cycle.

It is therefore important to provide a method and a device which make it possible to reliably predict the expected emissions of a motor vehicle already in the development phase of the motor vehicle, in order to be able to carry out a change of the motor vehicle in the event of an expected exceeding of a threshold value. Such an estimation, which is based solely on measurements on test stands or in a moving vehicle, is extremely costly due to the large diversity of the driving cycles that can be taken into account.

In the prior art, it is therefore proposed, for example, to determine a so-called pilot cycle for which meeting the emission regulations is particularly challenging. It is assumed here that if this is the case for the most challenging cycle, it is possible to meet the emission regulations for all conceivable driving cycles.

However, in addition to the requirement that exhaust gas regulations have to be met in each conceivable or permissible driving cycle, an important goal of vehicle or drive development is to minimize the total emissions of the vehicle drive system in real operation. When adapting or optimizing the vehicle drive system to the most or particularly critical driving cycles, although it may be ensured that the criteria are met in all cycles, there is thus a risk of significant deterioration of the emissions in less critical cycles. If the less critical cycles in real driving operation are still more frequent cycles, which is usually the case, then the overall system is degraded in terms of emissions in real operation by this optimization. For example, optimizing emissions for critical, but in reality very unusual, driving cycles with extreme speed profiles (e.g. extreme bumpy driving with hard acceleration) can lead to emissions being worse for less critical but more frequent driving cycles with common speed profiles (e.g. short city trips with traffic lights), which overall can lead to higher emissions in real operation.

It is therefore very advantageous for the development of vehicles with emission optimization of internal combustion engines and the adaptation of their emission optimization to be able to automatically generate a large number of real speed profiles, the distribution of which corresponds to or approximates the real expected distribution. Thus, the starting point for the matched drive system is the generated speed profile with a profile representing the real operation.

The computer-aided generation of the speed profile is therefore an important technical starting basis in a profile representing a real operation, which in various scenarios can decisively improve the development or optimization of the vehicle drive system and thus contribute to a less and more efficient vehicle, in particular to a less and more efficient vehicle drive system.

A method for reducing exhaust gas emissions of a drive system of a vehicle having an internal combustion engine is therefore proposed, characterized by the following steps:

-generating a plurality of first driving gestures (51) by means of a computer-implemented machine learning system, wherein a statistical distribution of the first driving gestures depends on, in particular follows, a statistical distribution of the second driving gestures measured during the driving operation,

-calculating a respective exhaust emission (52) for the first driving profile by means of computer-implemented modeling of the vehicle or of a drive system of the vehicle,

-matching the drive system of the vehicle (53) according to at least one of the calculated exhaust emissions, wherein the matching is performed according to the degree or the behavior of the at least one of the calculated exhaust emissions and according to the statistical frequency of the respective first driving tendency, wherein the statistical frequency of the respective first driving tendency is determined by means of a statistical distribution of the first driving tendency.

The driving behavior is a behavior of a driving behavior of the vehicle, wherein the driving behavior is a physical or technical characteristic of a drive train of the vehicle, in particular, which is measurable with a sensor, and which characterizes a further movement of the vehicle. As the most important variant, the speed profile of the vehicle falls within the range of the driving profile. The speed profile of the vehicle is for a specific journey one or more of the dominant parameters for determining emissions, consumption, wear and the like. The speed trend can be determined by the speed value, but also by a variable derivable therefrom, such as an acceleration value. Other main driving characteristics, the trend of which is important for applications such as determining emissions, consumption or wear, include in particular the position of the accelerator pedal, the position of the clutch pedal, the position of the brake pedal or the transmission gear ratio.

The proposed method enables an efficient verification and optimization of the exhaust emissions of the drive system globally not only for individual cycles or scenarios but also for field operation.

In this case, it may also be preferred to select the second driving profile from a plurality of third driving profiles, wherein the second driving profile is selected such that the distribution of the second driving profile corresponds to the distribution of the actual journey. If a statistically representative trip cannot be resorted to for the proposed method, the method can be significantly improved when such a representation is established or approximated by appropriate selection from available data.

In contrast to prior art methods, the virtual test environment allows for a reduction of exhaust emissions of the drive system based on a plurality of generated driving cycles compared to a selected sampled dominant cycle. The driving cycle is generated by means of a model of the real driving, which ensures on the one hand that the entire state space is covered sufficiently closely and on the other hand that the generated driving cycle corresponds well to the statistics of the real driving with respect to specific problems (for example emissions of the drive system). This can be achieved in particular by the correlation of the generated driving tendency with the driving tendency measured during driving operation. The distribution of the driving tendency considered for the simulation may depend on other statistics (e.g., driving statistics in a particular area, weather statistics, etc.).

In the proposed method, therefore, the test scenario is no longer predetermined on the basis of the knowledge collected on the preceding project and the system is tested and optimized according to this test scenario, but the operational capacity of the system with respect to emissions is determined over the entire state space by means of simulation. Here, the result maps the true probability of occurrence.

Thereby, unnecessary attention to test cases that are not very helpful for the current drive system, e.g. test cases that show up as having problems in previous projects, but that have no problems in the development project or are not suitable for showing problem fields of the actual development project, can be avoided, since these are not yet known in previous projects.

Too much attention to particularly demanding test cases, which often lead to over-specification or over-matching of the system, even if it is for example rare, can also be avoided. This strong interest often results from the lack of knowledge in which situation and frequency a single problem area occurs during real driving operation.

As already explained, a matching can be carried out here by means of a representative distribution of the generated driving profiles, which makes it possible for the drive system not to be optimized with regard to emissions for individual or particularly critical driving profiles. In contrast, the drive system is optimized in such a way that the emissions expected in real operation are minimized overall, that is to say that the sum of the end pipe emissions is minimized over all the strokes.

The optimization can be carried out by automatic matching of the topology of the components of the drive system, automatic matching of the components of the drive system or automatic matching of software functions for the drive system in the development of the drive system. The correspondingly adapted drive system is then correspondingly produced and used in the vehicle. The optimization can also be carried out by adapting the data in the application of the drive system, wherein the drive system of the respective application is used in the vehicle. In addition, the optimization can be carried out by adapting the control functions or control variables of the control software during the operation of the drive train in the vehicle.

In a preferred embodiment, the exhaust gas emissions are calculated not only from the generated driving profile, for example from a speed profile for the specific driving route, but also from route information from the associated driving route, for example from the gradient profile of the route. More accurate or more realistic exhaust emission values can thereby be calculated, which thereby improves the subsequent matching.

The statistical distribution of the first travel tendency preferably comprises the statistical distribution of the first travel tendency together with the respectively associated first travel route. Accordingly, the statistical distribution of the second driving tendency measured during the driving maneuver preferably includes the second driving tendency together with the statistical distribution of the respectively associated second driving route measured during the driving maneuver. In both cases, the distribution for the respective pairing of the driving profile and the associated driving route is therefore taken into account. Here, the driving route comprises route characteristics, such as, in particular, geographical characteristics, traffic flow characteristics, lane characteristics, traffic guidance characteristics and/or weather characteristics of the route.

A preferred computer implemented training of a machine learning system for generating driving tendency comprises the steps of:

-selecting a first driving route from a first database with driving routes,

a generator of the machine learning system obtains the first travel route as an input variable and generates a corresponding associated first travel profile for the first travel route,

the second database stores the travel routes and the associated travel patterns detected during the travel operation,

selecting a second travel route and a corresponding associated second travel pattern detected during the travel operation from a second database,

the discriminator of the machine learning system obtains as input variables a pairing comprising one of the first travel routes having the respective associated generated first travel gesture and a pairing comprising a second travel route having the respective associated second travel gesture detected during the travel operation,

the discriminator calculates, as a function of the input variables, an output which characterizes each pairing obtained as an input variable as a pairing with the generated first driving behavior and also as a pairing with the second driving behavior detected during driving operation,

optimizing an objective function, which represents the distance between the paired distribution with the generated first driving behavior and the paired distribution with the second driving behavior detected during the driving operation, as a function of the output of the discriminator.

An alternative preferred computer implemented training for generating a machine learning system of driving tendency comprises the steps of:

a generator of the machine learning system obtains the first random vector as an input variable and generates a first travel route and an associated first travel profile for the first random vector,

the database stores the travel routes and the associated travel patterns detected during the travel operation,

selecting a second travel route and a corresponding associated second travel pattern detected during the travel operation from the database,

the evaluator of the machine learning system receives as input variables a first pairing of the generated first travel route and the associated generated first travel gesture and a second pairing of the second travel route and the associated second travel gesture detected during the travel operation,

the discriminator calculates, as a function of the input variables, an output which characterizes each pairing obtained as an input variable by a first pairing of a first generated travel path and a corresponding associated first generated travel path and by a second pairing of a second travel path and a corresponding associated second detected travel path during travel operation,

-optimizing an objective function representing a distance between the distribution of the first pair and the distribution of the second pair based on the output of the discriminator.

Advantageously, the parameters of the machine learning system are adapted according to an optimization of the objective function in such a way that

The discriminator is optimized for distinguishing between the generated first driving tendency and the second driving tendency detected during the driving operation,

the generator is optimized for generating a first generated driving behavior with a first distribution, which is as difficult as possible to distinguish from a second driving behavior present with a second distribution and detected during driving operation by means of the discriminator.

The described training method provides a computer-implemented machine learning system with which a representative driving situation can be generated, whereby measures such as emissions optimization or system verification with regard to emissions can in turn be carried out taking into account the actual representative influence.

In an advantageous embodiment, the modeling of the vehicle or of the drive train of the vehicle comprises a model of the internal combustion engine and/or of the exhaust gas aftertreatment system of the internal combustion engine and/or of the engine controller and/or of the combustion process, as a result of which particularly precise simulation results are achieved.

To perform the described computer-implemented method, a computer program may be provided and stored on a machine-readable memory. A computer-implemented learning system comprising such a machine-readable memory may be arranged to perform a method wherein the calculations to be performed are implemented by one or more processors of the computer-implemented learning system.

Detailed Description

Fig. 1 shows a conventional method for reducing exhaust gas emissions of a vehicle having an internal combustion engine. In step 101, the results or empirical values of the previous project are observed and in step 102 test cases are found which in particular should cover the critical driving cycle. The specified test protocol from legislation is listed in step 103. In step 104, the definition of the vehicle or drive system to be optimized is carried out. In step 105, for the vehicle or drive system to be optimized, measurements of exhaust emissions are performed for the test cycles selected in steps 102 and 103 or for the specified test protocol. Here, the measurement may be, for example, in the order of about 100h and the corresponding result is stored or analyzed in step 106. The vehicle or the drive system can be optimized depending on the measurements performed.

Fig. 2 shows a proposed method for reducing exhaust gas emissions of a vehicle having an internal combustion engine. In step 201, a representative pairing of the driving situation and the associated driving route is obtained by a generative model of the machine learning system of the real driving. In step 202, a vehicle model or a model of a drive system is established for the vehicle or drive system to be optimized. The modeling of the vehicle or the drive system in the figure comprises in particular a submodel of the exhaust gas aftertreatment system, a combustion model and/or a model of a controller, in particular of an engine controller.

In step 203, the exhaust emissions are calculated in a simulated manner for the representative pairing of step 201, which is composed of the driving profile and the associated driving route, using a vehicle model or a model of the drive system of step 202. The results of the simulation are then stored or analyzed in step 204. The number of simulations can be in the order of approximately 10000h, for example. Furthermore, the simulated distribution may correspond to or approximate the actual distribution of the driving tendency and the driving route in real operation. The correlation or statistical frequency of a particular driving situation and driving route can be taken into account when adapting the drive system according to the calculated exhaust emissions.

Fig. 3 shows an example of a computer-implemented training method for a machine learning system, with which a pairing of a driving situation and an associated driving route can be generated, which pairing represents the method described for fig. 2.

The travel route or route of the vehicle is stored in the database 301.

In fig. 3, an exemplary route in database 301 is indicated at 311. The driving route or routes of the vehicle are stored in the database 302 together with the associated driving situation. In fig. 3, an exemplary pairing of a route and an associated driving situation in the database 302 is denoted by 321. The driving tendency in the database 302 corresponds to the driving tendency determined or measured during the driving operation of the vehicle. That is, the driving profile is preferably detected and stored by sensors of the vehicle when the vehicle actually travels along the route. Databases 301 and 302 are implemented throughout the system, particularly on machine-readable storage. Here, the database represents only data that is systematically stored in a machine-readable memory.

Now, in the machine learning system 304, the generator 341 is to be trained for generating a driving profile for the routes of the database 301. These driving situations are preferably determined on the basis of random input variables, for which purpose random variables, such as random vectors, may be provided in block 303. In particular, a random generator may be implemented in block 303, wherein the random generator may also be referred to herein as a pseudo-random generator.

The driving pattern generated by the generator 341 should preferably be as indistinguishable or hardly distinguishable from the driving pattern determined during driving operation from the database 302. For this purpose, the discriminator 342 is trained to discriminate as well as possible between the driving tendency generated by the generator 341 and the driving tendency extracted from the database 302, or between corresponding pairs of driving tendency and route characteristics. The learning system should not only generate individual driving patterns, which are as far as possible indistinguishable from the driving patterns measured during individual driving operations. Conversely, the distribution of the generated driving tendency in the parameter space of the input variable should also be as close as possible to the distribution of the driving tendency measured during driving operation in the parameter space of the input variable, i.e. a representative distribution of the driving tendency is achieved.

The training of the machine learning system 304 for this purpose comprises an optimization of an objective function 305, according to which the parameters of the generator 341 and of the discriminator 342 are matched.

The proposed training of the machine learning system 304 will be described in more detail below with reference to fig. 3.

The routes in the database 301 are stored in particular as a sequence of discrete data points, with the route characteristics being stored for each data point or for each discretized step.

For example, the route r in the database 1 has a length N: r ═ r (r)₁，...，r_N). Each data point r_tCorresponding to one discretization step. Particularly preferred are implementations in which the discretization step corresponds to a temporal or spatial discretization. In the case of time discretization, the data points each correspond to the time elapsed since the start of the route and the sequence of data points therefore corresponds to the time trend. In the spatial discretization, the data points each correspond to a segment of the road traversed along the route.

The sampling rate is typically constant. In the time discretization, the sampling rate can be defined, for example, as x seconds, and in the spatial discretization, as x meters.

Each data point r of the route_tThe characteristics of the route at the corresponding discretized step, that is to say r, are described_t∈R^D. D is the number of route characteristics, where each dimension of the multidimensional route characteristic is counted here as a dimension of the one-dimensional route characteristic.

Such route characteristics can, for example, relate to discretized steps, in particular to time points or time intervals or locations or road sections or distances:

-a geographical feature, such as absolute altitude or grade,

-traffic flow characteristics, such as a time-dependent average speed of traffic,

-lane characteristics, such as number of lanes, lane type or lane curvature,

traffic guidance properties, such as speed limits, number of traffic lights or number of specific traffic signs, in particular parking or giving a way, or pedestrian crossings,

weather characteristics, such as rain, wind speed, presence of fog at a predefined point in time.

A route is selected from the database 301 and transmitted to the generator 341 in step 313.

Also preferably, a random vector is determined in block 303 and transmitted to generator 341 in step 331. The random vector z is extracted, i.e. randomly determined. In this case, z ∈ R applies in particular^LWhere L may optionally depend on the length N of the route. The distribution is preferably fixed to a simple family of distributions, such as gaussian distributions or equal distributions, from which z is extracted.

The input variables of generator 341 now preferably consist of variables, random vector z and path r. Instead of generating the input into the generator 341 purely randomly, the generated driving profile can therefore be adjusted for a specific route characteristic (kondationiren). For example, different driving situations can be generated for the same predefined route r by sampling different z. Here, the route characteristics of the route r in the database 301 may be actually measured route characteristics, route characteristics defined by an expert, or route characteristics learned by a machine learning system, such as a neural network. Routes having route characteristics created by two or three of these variations may also be provided in the database 301.

In an exemplary application, the driving profile generated in this case is used to determine emission characteristics of the drive system of the vehicle, for example, by specifically modifying some of the route characteristics, generating a matching driving profile and simulating emissions for these curves, in order to specifically study the extent to which the particular route characteristics influence the emissions during the combustion process. This allows the parameters of the drive system to be specifically optimized, for example, for specific, for example, for a particularly required course curve in, for example, a control unit, in particular parameters of the control means of the drive system.

The generator 341 now generates the driving profile from the input quantities, the random vector (step 331) and from the selected route (step 313). For this purpose, the generator 342 has a computer-implemented algorithm with which to implement the generation of the model and the output of the driving profile (step 343).

Such a driving profile generated by the generator 341 may be output, for example, as pi ═ pi (pi ═ pi-₁，...，π