DE102019216312A1 - Prediction and / or assessment of the emission quantity of substances in the exhaust gas of internal combustion engines - Google Patents

Abstract

Method (100) for predicting and / or assessing (1a) the amount of emissions of at least one substance in the exhaust gas of an internal combustion engine (1) with the following steps: a plurality of values, the state variables (2a-2f) of a drive train (2) with the internal combustion engine (1) at least two different previous times (t-τ1, t-τ2, ..., t-τD) in a time horizon (τD), a classifier (3a) and / or a regressor (3b), as input variables (31) supplied (110); the input variables (31) are mapped (120) by the classifier (3a) and / or the regressor (3b) onto at least one output variable (32); Output variable (32), the prediction or assessment (1a) sought is evaluated (130). Method for training a classifier (3a) and / or regressor (3b) for use in the method (100) using values of state variables ( 2a-2j) of the drive train (2) as well as values of actual emission quantities (1b) of the mi At least one substance that relates to a plurality of previous points in time (t-τ1, t-τ2, ..., t-τD) within a time horizon (τD).

Description

The present invention relates to the prediction and / or assessment of the emission behavior during the operation of internal combustion engines.

State of the art

The maximum limits for pollutant emissions from internal combustion engines for vehicles have been progressively tightened in recent years. The limit values that apply today can only be achieved with exhaust aftertreatment. Passive systems such as particle filters are no longer sufficient, especially for diesel engines. Rather, additional systems are required to reduce nitrogen oxide emissions, such as a storage catalytic converter (NOx Storage Catalyst, NSC) or a selective catalytic reduction (SCR).

Function monitoring is prescribed for such systems so that, in the event of any emissions-related errors occurring, the user of the vehicle can be prompted to rectify the error in a workshop by activating the engine control lamp. A method for monitoring the function of an SCR system is in the DE 10 2017 204 300 A1 disclosed.

Disclosure of the invention

In the context of the invention, a method for predicting and / or assessing the amount of emissions of at least one substance in the exhaust gas of an internal combustion engine was developed. The substance can in particular be a pollutant whose emission quantity in the exhaust gas is subject to legal restrictions, such as nitrogen oxide, NO _x , for example. The emission quantity can in particular be specified, for example, as an emission rate per unit of time, but also, for example, as a cumulative quantity over a predetermined period.

In this method, a plurality of values that assume state variables of a drive train with the internal combustion engine at at least two different previous points in time in a time horizon are fed to a classifier and / or a regressor as input variables. The input variables are mapped to at least one output variable by the classifier and / or by the regressor. The prediction or assessment sought is evaluated from the at least one output variable.

The classifier and / or the regressor is not restricted to only assessing a steady state of the internal combustion engine. Rather, it can also take into account the dynamics of the state variables and their effects on the emission behavior. For example, the emission of the substance can be increased at the moment of a sudden acceleration of the internal combustion engine compared to steady-state operation at a new speed. As a result, a more accurate prediction or assessment is obtained on average over real operating cycles with changing loads.

The classification can be, for example, a binary classification in terms of whether the emission quantity of the substance corresponds to the norm or not. A more detailed classification can provide for gradations in this regard, so that it is possible to react in a correspondingly graded manner. The regression can, for example, provide a value for the emission amount and / or it can supply a value of another variable from which the emission amount can be determined in connection with other known variables. The value of the emission quantity ultimately obtained can in turn be reacted to in a graduated manner according to predetermined criteria.

For example, in response to the fact that a limit value for the emission quantity is slightly exceeded, only the engine control lamp can be activated. In the event of larger excesses, for example, a warning tone can sound at progressively shorter time intervals in order to emphasize the request for a repair. Further measures are also possible and useful in order to enforce a compulsory repair. For example, after the internal combustion engine has been switched off, a restart can be prevented until the error has been eliminated and the limit value is adhered to again. Before this, for example, a waiting period can be granted, which is measured in operating hours or, if used in a vehicle, in kilometers of driving distance. Furthermore, for example, at least one driving dynamics system of the vehicle can be activated in order to make the driving dynamics of the vehicle less comfortable. For example, devices that make staying in the vehicle more comfortable, such as the radio, heating and / or air conditioning, can also be deactivated. Such comfort restrictions do not force the user of the vehicle to have the vehicle towed to a workshop at great expense, but they make it easier for him to put the error on the back burner and continue to use the vehicle unchanged.

In this context, it is of particular advantage that the method can not only predict certain individual errors that are potentially relevant to exhaust emissions, but also directly provides a prediction and / or assessment of the amount of emissions. This records whether the error actually has an impact on the amount of emissions in the respective situation and how strong the environmental impact, if any. For example, one and the same error can have a greater or lesser effect on the amount of emissions, depending on the operating situation.

In a particularly advantageous embodiment, the time horizon over which the previous points in time extend is based on the length of the path from a fuel supply and / or an air supply into the internal combustion engine to the discharge of the exhaust gas from the internal combustion engine, and / or on the basis of the time duration the fuel and / or the air required to cover this distance is determined. In this respect, cars in particular differ from trucks and other commercial vehicles. In the case of a car, the distances are significantly shorter, so that it is sufficient to take a shorter time horizon into account.

• • eine Zufuhrmenge des Brennstoffs, und/oder
• • ein Zufuhrdruck des Brennstoffs, und/oder
• • eine Zufuhrmenge von Luft, und/oder
• • eine Katalysatortemperatur in einem Abgasnachbehandlungssystem, und/oder
• • eine vor dem Eintritt in das Abgasnachbehandlungssystem gemessene Menge der mindestens einen Substanz im Abgas, und/oder
• • eine gemessene oder vorhergesagte Emissionsmenge der mindestens einen Substanz zu mindestens einem zurückliegenden Zeitpunkt.

The state variables on the basis of which the output variable is formed and thus ultimately the prediction or assessment sought is evaluated are particularly suitable

• • a supply amount of the fuel, and / or
• • a supply pressure of the fuel, and / or
• • a supply amount of air, and / or
• • a catalyst temperature in an exhaust gas aftertreatment system, and / or
• A quantity of the at least one substance in the exhaust gas measured before entering the exhaust gas aftertreatment system, and / or
• • a measured or predicted amount of emissions of the at least one substance at at least one point in time in the past.

Taking into account an earlier emission quantity can in particular prevent physically implausibly rapid changes in the emission quantity from being predicted. The other variables have proven to be particularly meaningful with regard to the stoichiometry of the processes that lead to the formation of NO _x in the exhaust gas of diesel engines.

• • das Drehmoment, und/oder
• • die Motordrehzahl, und/oder
• • der Ladedruck eines Turboladers, und/oder
• • eine durch die Abgasrückführung zurückgeführte Menge an Abgas.

In addition, other state variables can also be taken into account, such as, for example

• • the torque, and / or
• • the engine speed, and / or
• • the boost pressure of a turbocharger, and / or
• • an amount of exhaust gas recirculated by exhaust gas recirculation.

The total number of state variables to be taken into account can depend, for example, on how many state variables are ultimately independent of one another, or which of the state variables can be expressed by combinations of other state variables.

In a particularly advantageous embodiment, the state variables are determined from measured values that are supplied by sensors required for normal operation of the drive train. The method can then be used to monitor the functioning of an internal combustion engine without the need for additional sensors or cables.

In a further particularly advantageous embodiment, the classifier and / or the regressor contains a modeling of the combustion process in the internal combustion engine as a Gaussian process. Such a modeling is particularly well tailored to the situation that state variables of the drive train can generally only be observed at discrete points in time, while in truth they are continuous variables.

The method described above does not necessarily require that prior knowledge is already available about which state variables of the drive train are particularly meaningful for the prediction or assessment of the emission quantity. Rather, this knowledge can be acquired entirely with machine learning.

The invention therefore also relates to a method for training a classifier and / or regressor for use in the method described above. The method is based on the assumption that values of several state variables of the drive train as well as values of actual emission quantities of at least one substance are available for a plurality of points in time in the past within a time horizon.

As part of the method, for each state variable and each of the previous points in time, it is pre-classified in binary form as to whether the value of the state variable at the respective point in time is relevant for the prediction and / or assessment of the emission quantity. Parameters that characterize the behavior of the classifier and / or the regressor are optimized in such a way that the values of the state variables pre-classified as relevant are mapped to at least one output variable at the respective previous points in time by the classifier and / or by the regressor which, according to a model cost function, is as closely as possible in line with the values of the actual emission quantities.

It was recognized that in this way the effort for training a classifier and / or regressor, which, as an alternative or in addition to current values of the state variables, also takes into account values of the state variables that were different in the past, can be drastically reduced. The step from a stationary model, which only takes into account current values of state variables, to a dynamic model, which also takes past values into account, multiplies the total number of variables to be processed. If, for example, 11 state variables are measured, taking into account 10 previous points in time leads to a total of 121 variables to be processed. Particularly in the case of a classifier that includes modeling as a Gaussian process, the effort increases disproportionately with the number of variables on which the model depends.

As explained above, however, not all of these quantities are independent of one another; some quantities can be expressed by others. In particular, through physical dependencies and time constants of the processes in an internal combustion engine, values of a state variable at a first point in time can depend on values of other state variables at points in time further back. Therefore, only a small part of the prima facie values to be taken into account (in the example mentioned 121) will actually be relevant. Omitting the other variables only minimally, if at all, degrades the accuracy of the classifier and / or the regressor. On the one hand, this saves computing time and, on the other hand, it prevents the information contained in the actually relevant quantities from being diluted by a large excess of less relevant quantities.

The computational effort saved in this way more than compensates for the additional effort that is required for the binary pre-classification.

In a particularly advantageous embodiment, at least one candidate for mapping the state variables and past times to the classes “relevant” and “not relevant” is determined as part of the pre-classification. Such a mapping can be expressed as a two-dimensional binary matrix, for example, which assigns a 1 for “relevant” or a 0 for “not relevant” to every state variable and every point in time (about 1 to 10 discrete time steps in the past).

A pre-cost function is used to assess the extent to which the pre-classification is applicable to the candidate. The candidate is optimized with the aim of improving the assessment using the pre-cost function.

The pre-cost function can be any function that assigns each candidate a measure of the extent to which a pre-classification with this candidate is applicable. This pre-cost function will not be continuous in the space of the candidates, since a single classification of a previous value of a state variable can only ever be switched discretely between “relevant” and “not relevant”. The optimization method for the optimization of the candidate should therefore not require any continuity of the pre-cost function, but is otherwise not restricted.

In a particularly advantageous embodiment, the values of the state variables classified as relevant according to the candidate are mapped to at least one output variable by a training classifier and / or training regressor in connection with the respective previous times to which they relate. An assessment of the extent to which this output variable is in line with the values of the actual emission quantities is included in the assessment using the pre-cost function. For this assessment, an output variable that has the character of a qualitative estimate and can therefore be determined very quickly is completely sufficient. The more precise determination of the output variable, which is ultimately used for the prediction and / or assessment of the emission quantity, is carried out later by the trained classifier or regressor. A training classifier or training regressor is therefore particularly advantageously selected, the evaluation of which is at least a factor of 10 faster than the evaluation of the classifier or regressor to be trained.

The pre-cost function can also contain, for example, a measure for the mutual information of the state variables. This is based on the previously explained knowledge that not all state variables are independent of one another, but, for example, some state variables can be expressed by combinations of other state variables.

In a further particularly advantageous embodiment, the available pairs of state variables and previous points in time (that is, for example “amount of fuel a second ago” or “fuel pressure three seconds ago”) are binary preclassified one after the other, with the sequence in particular being able to be selected at random, for example. The next pair is then always drawn from the pool of pairs in order to then decide whether the respective state variable is relevant at the respective previous point in time or not. A pre-classification in this way is particularly fast and can also be carried out, for example, with a pre-classification that uses a Training classifier or training regressor works, combine. For example, by means of an individual assessment, pairs of state variables and previous points in time can initially be excluded that are obviously not or only very weakly correlated with the quantity of emissions. With regard to the amount of emissions, for example, it is negligible whether the low beam is switched on.

The result of the training is condensed in the binary pre-classification and in the parameters ultimately obtained that characterize the behavior of the classifier and / or regressor. Anyone who has this information can directly specify a classifier and / or a regressor for the processing of state variables for the application to which they relate and thus determine the prediction and / or assessment of the emission quantity sought without performing any training to have to. In particular, in many applications the training is not performed by the same entity that determines the ultimate prediction and / or assessment. Particularly in the case of applications on vehicles, the prediction and / or assessment of the emission quantity integrated in a control unit of the vehicle is aimed at the end user of the vehicle, who should be requested to visit a workshop in the event of a fault. However, the end user cannot then be expected to carry out training first, but this has already been carried out beforehand by the manufacturer of the control device.

• • eine binäre Vorklassifikation von Werten von Zustandsgrößen eines Antriebsstrangs mit einer Brennkraftmaschine, die sich auf mehrere zurückliegende Zeitpunkte beziehen, als relevant oder nicht relevant für die Vorhersage und/oder Beurteilung der Emissionsmenge mindestens einer Substanz im Abgas der Brennkraftmaschine, und/oder
• • Parameter, die das Verhalten eines Klassifikators und/oder Regressors charakterisieren, der Werte von Zustandsgrößen zu den jeweiligen zurückliegenden Zeitpunkten auf mindestens eine im Hinblick auf die

The invention therefore also relates to a parameter set obtained with the training method, the

• A binary pre-classification of values of state variables of a drive train with an internal combustion engine, which relate to several previous points in time, as relevant or not relevant for the prediction and / or assessment of the emission quantity of at least one substance in the exhaust gas of the internal combustion engine, and / or
• • Parameters that characterize the behavior of a classifier and / or regressor, the values of state variables at the respective previous points in time to at least one with regard to the

Prediction and / or assessment depicts evaluable output variable, includes. For the reasons mentioned, this parameter set is a product that can be sold independently.

In particular, the methods can be implemented entirely or partially by computer. The invention therefore also relates to a computer program with machine-readable instructions which, when they are executed on one or more computers, cause the computer or computers to carry out one of the described methods. In this sense, control devices for vehicles and embedded systems for technical devices, which are also able to execute machine-readable instructions, are to be regarded as computers.

The invention also relates to a machine-readable data carrier and / or to a download product with the computer program. A download product is a digital product that can be transmitted via a data network, i.e. that can be downloaded by a user of the data network and that can be offered for immediate download in an online shop, for example.

Furthermore, a computer can be equipped with the parameter set, with the computer program, with the machine-readable data carrier or with the download product.

Finally, in a further particularly advantageous embodiment, a field-programmable logic gate arrangement (Field Programmable Gate Array, FPGA) can also be configured for executing one of the methods described. FPGAs are particularly suitable for use in vehicles because they are particularly energy-efficient in relation to the computing power. When implementing the factors described in one or more control units for a vehicle, there are usually specifications with regard to the maximum electrical power consumption and / or with regard to the maximum permissible heat development.

Further measures improving the invention are illustrated in more detail below together with the description of the preferred exemplary embodiments of the invention with reference to figures.

Figure list

• 1 Ausführungsbeispiel des Verfahrens 100 zur Vorhersage und/oder Beurteilung der Emissionsmenge;
• 2 Beispielhafter Verlauf der realen Emissionsmenge 1b und der mit dem Verfahren 100 vorhergesagten Emissionsmenge 1a von NO_x aus einer Brennkraftmaschine 1;
• 3 Ausführungsbeispiel des Verfahrens 200 zum Training eines Klassifikators, und/oder eines Regressors;
• 4 Beispielhaftes Ergebnis 4 des Vorklassifizierens 210 für einen Antriebsstrang 2 mit einer Brennkraftmaschine 1 (4a); schematische Darstellung des Antriebsstrangs 2 (4b).

It shows:

• 1 Embodiment of the method 100 to predict and / or assess the amount of emissions;
• 2 Exemplary course of the real emission quantity 1b and the one with the procedure 100 predicted amount of emissions 1a of NO _x from an internal combustion engine 1 ;
• 3 Embodiment of the method 200 for training a classifier and / or a regressor;
• 4th Exemplary result 4th of pre-classifying 210 for a drive train 2 with an internal combustion engine 1 ( 4a) ; schematic representation of the drive train 2 ( 4b) .

1 Figure 3 is a flow diagram of one embodiment of the method 100 for prediction and / or assessment 1a the amount of emissions of at least one substance in the exhaust gas of an internal combustion engine 1 . The internal combustion engine 1 and their embedding in a drive train 2 are in 4b shown in more detail.

In step 110 becomes a plurality of values, the state variables 2a-2f of the drive train 2 with the internal combustion engine 1 at least two different previous times t-τ ₁ , t-τ ₂ , ..., t-τ _D in a time horizon τ _D assume a classifier 3a , and / or a regressor 3b , as input variables 31 fed. Inside the box 110 some exemplary possibilities are given, such as the state variables 2a-2f can be determined.

According to block 111 the time horizon τ _D can be based on the length of the path from a fuel supply and / or an air supply to the internal combustion engine 1 until the exhaust gas is expelled from the internal combustion engine 1 , and / or on the basis of the time required for the fuel and / or the air to cover this distance.

• • eine Zufuhrmenge 2a des Brennstoffs, und/oder
• • ein Zufuhrdruck 2b des Brennstoffs, und/oder
• • eine Zufuhrmenge 2c von Luft, und/oder
• • eine Katalysatortemperatur 2d in einem Abgasnachbehandlungssystem 5, und/oder
• • eine vor dem Eintritt in das Abgasnachbehandlungssystem 5 gemessene Menge 2e der mindestens einen Substanz im Abgas, und/oder
• • eine Emissionsmenge 2f der mindestens einen Substanz zu mindestens einem zurückliegenden Zeitpunkt, die als Vorhersage 1a oder auch als Messung 1b vorliegen kann.

According to block 112 can be certain specific state variables 2a-2f be chosen that are specific to the exhaust gas behavior of internal combustion engines 1 have proven relevant in vehicles. In detail these are:

• • a feed amount 2a of the fuel, and / or
• • a feed pressure 2 B of the fuel, and / or
• • a feed amount 2c of air, and / or
• • a catalyst temperature 2d in an exhaust aftertreatment system 5 , and or
• • one before entering the exhaust aftertreatment system 5 measured amount 2e the at least one substance in the exhaust gas, and / or
• • a quantity of emissions 2f the at least one substance at at least one previous point in time, which is used as a forecast 1a or as a measurement 1b may exist.

According to block 113 can specifically such state variables 2a-2f can be selected, which can be determined from measured values that are used for normal operation of the drive train 2 necessary sensors are supplied.

In step 120 become the input variables 31 from the classifier 3a , and / or from the regressor 3b , to at least one output variable 32 pictured. In the in 1 The example shown includes the classifier 3a , and / or the regressor 3b , a modeling 33 of the combustion process in the internal combustion engine 1 as a Gaussian process. The behavior of this modeling 33 is through parameters 34 characterized.

In step 130 becomes from the at least one output variable 32 the forecast or assessment you are looking for 1a evaluated. This does not rule out the initial size 32 also the prediction or assessment directly 1a can deliver. The evaluation can in particular, for example, be an evaluation of the conformity to the standard and / or the environmental impact of an output variable 32 include the delivered emission quantity of the substance.

• • als Maßnahme 141 die Motorkontrollleuchte eines Fahrzeugs aktiviert, und/oder
• • als Maßnahme 142 in progressiv kürzer werdenden Zeitabständen ein Warnton ausgegeben, und/oder
• • als Maßnahme 143 mindestens ein fahrdynamisches System des Fahrzeugs angesteuert, um die Fahrdynamik des Fahrzeugs unkomfortabler zu machen, und/oder
• • als Maßnahme 144 mindestens eine Einrichtung des Fahrzeugs, die den Aufenthalt im Fahrzeug angenehmer macht, deaktiviert, und/oder
• • als Maßnahme 145 nach dem Abschalten der Brennkraftmaschine 1 ein erneuter Start unterbunden.

In the in 1 The example shown is checked whether the prediction and / or assessment 1a a given criterion 140 Fulfills. If so, countermeasures are taken 141-145 seized. In detail, for example:

• • as a measure 141 the engine control lamp of a vehicle is activated, and / or
• • as a measure 142 a warning tone is issued at progressively shorter intervals, and / or
• • as a measure 143 at least one driving dynamics system of the vehicle is activated in order to make the driving dynamics of the vehicle less comfortable, and / or
• • as a measure 144 at least one device in the vehicle that makes staying in the vehicle more comfortable, deactivates, and / or
• • as a measure 145 after switching off the internal combustion engine 1 a restart is prevented.

2 shows an exemplary comparison of the curves of an actual emission quantity 1b of nitrogen oxide, NO _x , from an internal combustion engine 1 in a vehicle, as well as one with the procedure 100 determined prediction 1a this amount of emissions plotted over time t. There are a few places where the prediction 1a short-term about the actual amount of emissions 1b overshoots, and even fewer places where the actual amount of emissions 1b about the prediction 1a overshoots. The mean, which is important for the environmental impact and thus also the function monitoring of the exhaust system, is that of the load changes of the internal combustion engine 1 cause the actual emission amount to change over time 1b predicted with sufficient accuracy.

3 Figure 3 is a flow diagram of one embodiment of the method 200 to train the classifier 3a , and / or the regressor 3b , for use in the in 1 procedure shown 100 . For this procedure 200 become values of state variables 2a-2j of the drive train 2 as well as values of actual emission quantities 1b the at least one substance used. These values each relate to a plurality of previous times t-τ ₁ , t-τ ₂ , ..., t-τ _D within a time horizon τ _D.

In step 210 is for each state variable 2a-2j and each of the previous points in time t-τ ₁ , t-τ ₂ , ..., t-τ _D , that is, for pairs of state variables and points in time, a binary pre-classification 4th determined to that effect whether the value of the state variable 2a-2j at the respective point in time t-τ ₁ , t-τ ₂ , ..., t-τ _D for the prediction and / or assessment 1a the amount of emissions is relevant. This pre-classification 4th can include, for example, that of certain state variables 2a-2f only the values at a few previous times (e.g. t-τ ₂ and t-τ ₄ ) are relevant. The pre-classification 4th can also include, for example, that other state variables (in the in 3 The example shown is the state variables 2g-2y ) for determining the prediction and / or assessment 1a are not needed. As explained above, this does not mean that these state variables do not physically affect the emission quantity sought. Rather, for example, the need to take into account a state variable 2g-2y do not apply if this is due to other state variables that have already been taken into account 2a-2f expresses.

According to the pre-classification 4th state variables that are considered relevant (here 2a-2f) are shown in step 250 used to parameter 34 to optimize the behavior of the classifier 3a , and / or the regressor 3b , characterize. This optimization is aimed at ensuring that the corresponding values of the state variables 2a-2f at the respective previous times t-τ ₁ , t-τ ₂ , ..., t-τ _D from the classifier 3a , and / or from the regressor 3b , to at least one output variable 32 are mapped according to a model cost function 35 as closely as possible in line with the values of the actual emission quantities 1b stands.

Inside the box 210 an exemplary way is given in which the pre-classification 4th can be determined. In step 220 will be at least one candidate 4 * for the pre-classification 4th , ie for the mapping of pairs of state variables 2a-2j and previous times t-τ ₁ , t-τ ₂ , ..., t-τ _D to the classes “relevant” and “not relevant”. This can, for example, according to block 221 contain the available pairs of state variables 2a-2j and previous times t-τ ₁ , t-τ ₂ , ..., t-τ _D successively preclassified in binary. For this purpose, the pairs can be drawn in random order from the “pool” that is still available.

In step 230 the extent to which the pre-classification according to the currently examined candidate 4 * is applicable is assessed in accordance with a pre-cost function 35 *. The result 230a this evaluation is used as feedback for the optimization of the candidate 4 * in step 240 used. The aim of the optimization is to change the candidate 4 * in such a way that it is likely when the step is run through again 230 a better rating 230a receives.

Inside the box 230 is given an exemplary way in which the assessment 230a can be determined. According to block 231 the values of the state variables classified as relevant according to the candidate 4 * 2a-2f in connection with the previous times t-τ ₁ , t-τ ₂ , ..., t-τ _D , to which they relate, from a training classifier and / or training regressor to at least one output variable 231a pictured. It can be particularly advantageous according to the sub-block 231b a training classifier or training regressor can be selected, its evaluation in comparison to the evaluation of the classifier actually to be trained 3a or regressors 3b by at least one factor 10 is faster. As explained above, a significantly less complex classifier or regressor is sufficient for assessing the relevance than that for determining the ultimate prediction and / or evaluation 1a is used. With a fast training classifier or training regressor, a large number of irrelevant pairs of state variables and points in time can be “screened out” to find the more complex classifier 3a or regressor 3b then only apply to the relevant information.

According to block 232 an assessment is made of the extent to which the output variable supplied by the training classifier or the training regressor 231a in accordance with the values of the actual emission quantities 1b stands in the evaluation 230a of the candidate 4 * by the pre-cost function 35 *.

• • Drehmoment 2g,
• • Motordrehzahl 2h,
• • Ladedruck 2i und
• • durch die Abgasrückführung zurückgeführte Abgasmenge 2j

wird für den aktuellen Zeitpunkt t sowie für zehn zurückliegende Zeitpunkte t-τ₁, t-τ₂, ..., t-t_D angegeben, ob das entsprechende Paar aus Zustandsgröße und Zeitpunkt für die Vorhersage und/oder Bewertung der Emissionsmenge der Substanz zum Zeitpunkt t relevant ist (Symbol ✓ ) oder nicht (Symbol x). Die Zeitintervalle τ₁, τ₂, ..., τ_D sind in 1-Sekunden-Abständen gestaffelt. Das heißt, τ₁=1 Sekunde, τ₂=2 Sekunden, ..., τ_D=10 Sekunden. 4a shows an exemplary result 4th of pre-classifying 210 for a drive train 2 who uses a diesel engine as an internal combustion engine 1 as well as an exhaust aftertreatment system 5 with a storage catalytic converter, NSC, includes. For the state variables already mentioned 2a-2f as well as for four other state variables

• • torque 2g ,
• • engine speed 2h ,
• • boost pressure 2i and
• • Amount of exhaust gas recirculated by exhaust gas recirculation 2j

it is indicated for the current time t and for ten previous times t-τ ₁ , t-τ ₂ , ..., tt _D whether the corresponding pair of state variable and time for the prediction and / or evaluation of the emission quantity of the substance at the time t is relevant (symbol ✓) or not (symbol x). The time intervals τ ₁ , τ ₂ , ..., τ _D are staggered in 1-second intervals. That is, τ ₁ = 1 second, τ ₂ = 2 seconds, ..., τ _D = 10 seconds.

The table shows that of the state variables 2a-2f only the values from a few previous times are required and then no values of the state variables either 2g-2y more are needed.

In 4b is the powertrain 2 outlined, for which the in 4a shown result 4th was determined. The supply amount 2a and the feed pressure (rail pressure) 2 B of the fuel, as well as the amount supplied 2c of air, determine the stoichiometry during the combustion of the fuel in the internal combustion engine 1 . This stoichiometry has an impact on the quantity 2e of the nitrogen oxide NO _x in the exhaust gas before entering the exhaust gas aftertreatment system 5 of the drive train 2 . In the exhaust aftertreatment system 5 decides among other things the catalyst temperature 2d on the reaction rate for the conversion of the nitrogen oxide into harmless products and thus on the final amount of emissions 2f , their prediction and / or evaluation 1a with the procedure 100 is determined and its measurement 1b for training with the procedure 200 can be used.

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.

Patent literature cited

• DE 102017204300 A1 [0003]

Claims (16)

Method (100) for predicting and / or assessing (1a) the amount of emissions of at least one substance in the exhaust gas of an internal combustion engine (1) with the following steps: a plurality of values, the state variables (2a-2f) of a drive train (2) with the internal combustion engine (1) Assume at least two different previous times (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) in a time horizon (τ _D ), a classifier (3a), and / or a regressor (3b), supplied (110) as input variables (31); • the input variables (31) are mapped (120) by the classifier (3a) and / or the regressor (3b) onto at least one output variable (32); • The prediction or assessment (1a) sought is evaluated (130) from the at least one output variable (32). Method (100) according to Claim 1 , the time horizon (τ _D ) based on the length of the path from a fuel supply and / or an air supply to the internal combustion engine (1) to the exhaust of the exhaust gas from the internal combustion engine (1), and / or based on the duration of the fuel and / or the air required to cover this distance is determined (111). Method according to one of the Claims 1 to 2 , where • a supply amount (2a) of the fuel, and / or • a supply pressure (2b) of the fuel, and / or • a supply amount (2c) of air, and / or • a catalyst temperature (2d) in an exhaust gas aftertreatment system (5) , and / or • an amount (2e) of the at least one substance in the exhaust gas measured before entering the exhaust gas aftertreatment system (5), and / or • a measured or predicted emission amount (2f, 1a, 1b) of the at least one substance for at least one previous point in time can be selected as state variables (2a-2f) (112). Method (100) according to one of the Claims 1 to 3 , the state variables (2a-2f) being determined (113) from measured values which were supplied by sensors necessary for normal operation of the drive train (2). Method (100) according to one of the Claims 1 to 4th , the classifier (3a) and / or the regressor (3b) including a modeling (33) of the combustion process in the internal combustion engine (1) as a Gaussian process. Method (100) according to one of the Claims 1 to 5 , in response to the fact that the prediction and / or evaluation (1a) of the emission quantity fulfills a predetermined criterion (140), • the engine control lamp of a vehicle is activated (141), and / or • a warning tone is output in progressively shorter time intervals (142), and / or • at least one driving dynamics system of the vehicle is activated (143) in order to make the driving dynamics of the vehicle less comfortable, and / or • at least one device of the vehicle that makes staying in the vehicle more comfortable is deactivated ( 144), and / or • after the internal combustion engine (1) has been switched off, restarting is prevented (145). Method (200) for training a classifier (3a) and / or regressor (3b) for use in the method (100) according to one of the Claims 1 to 6th using values of state variables (2a-2j) of the drive train (2) as well as values of actual emission quantities (1b) of the at least one substance, each of which relates to a plurality of previous points in time (t-τ ₁ , t-τ ₂ , ... , t-τ _D ) within a time horizon (τ _D ), with the steps: • for each state variable (2a-2j) and each of the previous times (t-τ ₁ , t-τ ₂ , ..., t- τ _D ) is pre-classified in binary (210) whether the value of the state variable (2a-2j) at the respective point in time (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) for the prediction and / or Assessment (1a) of the emission volume is relevant (4); • Parameters (34) that characterize the behavior of the classifier (3a) and / or regressor (3b) are optimized (250) so that the relevant pre-classified (4) values of the state variables (2a-2f) correspond to the respective past Points in time (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) by the classifier (3a), and / or by the regressor (3b), are mapped to at least one output variable (32), which according to According to a model cost function (35), it is as consistent as possible with the values of the actual emission quantities (1b). Method (200) according to Claim 7 , the pre-classification (210) comprising the following steps: • at least one candidate (4 *) for mapping the state variables (2a-2j) and previous times (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) the classes “relevant” and “not relevant” are determined (220); • In accordance with a pre-cost function (35 *) it is assessed (230) to what extent the pre-classification according to this candidate (4 *) is correct (230a); • The candidate (4 *) is optimized (240) with the aim of improving the assessment (230a) by means of the pre-cost function (35 *). Method (200) according to Claim 8 , where the values of the state variables (2a-2f) classified as relevant according to the candidate (4 *) in connection with the respective previous times (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) which they relate, mapped onto at least one output variable (231a) by a training classifier and / or training regressor (231) and an assessment of the extent to which this output variable (231a) is in line with the values of the actual emission quantities (1b) is included in the assessment (230a) by the pre-cost function (35 *) (232). Procedure according to Claim 9 , whereby a training classifier or training regressor is selected (231b), the evaluation of which is at least a factor of 10 faster than the evaluation of the classifier (3a) or regressor (3b) to be trained. Method (200) according to one of the Claims 7 to 10 , whereby the available pairs of state variables (2a-2j) and previous times (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) are binary preclassified one after the other (221). Parameter set, comprising • a binary pre-classification (4) of values of state variables (2a-2j) of a drive train (2) with an internal combustion engine (1), which refer to several previous points in time (t-τ ₁ , t-τ ₂ , .. ., t-τ _D ), as relevant or not relevant for the prediction and / or assessment (1a) of the emission amount of at least one substance in the exhaust gas of the internal combustion engine (1), and / or • parameters (34) that determine the behavior of a Classifier (3a) and / or regressor (3b) characterize the values of state variables (2a-2f) at the respective previous times (t-τ ₁ , t-τ ₂ , ..., t-τ _D ) to at least one with regard to the prediction and / or assessment (1a) depicts evaluable output variable (32), obtained with the method (200) according to one of the Claims 7 to 11 . Computer program containing machine-readable instructions which, when executed on one or more computers, cause the computer or computers to implement a method (100, 200) according to one of the Claims 1 to 11 to execute. Machine-readable data carrier and / or download product with the parameter set according to Claim 12 , and / or with the computer program Claim 13 . Computer with the parameter set Claim 12 , with the computer program Claim 13 , and / or with the machine-readable data carrier and / or download product Claim 14 . Field-programmable logic gate arrangement, FPGA, configured to carry out a method (100, 200) according to one of the Claims 1 to 11 .

Images