CN118705072A - Method for generating injection quantity correction model and use thereof, controller and vehicle - Google Patents

Abstract

The invention relates to a method for generating an injection quantity correction model (16) for a gasoline engine using an algorithm for machine learning and the use of the injection quantity correction model (16). In addition, a corresponding controller (18) and a corresponding vehicle are provided. Based on at least one determined state variable of the air path of the gasoline engine, the generated injection quantity correction model (16) is used to estimate the deviation of the air mixture leaving the gasoline engine from the stoichiometric combustion air ratio. Based on this estimate, a correction (K _inj ) of the injection quantity is performed to reduce or prevent deviations from the stoichiometric combustion air ratio in order to reduce the emission of harmful gases and the consumption of fuel.

Description

Method for generating injection quantity correction model and use thereof, controller and vehicle

Technical Field

The present invention relates to a method for generating an injection quantity correction model for a gasoline engine by using an algorithm for machine learning and the use of the injection quantity correction model. In addition, a corresponding controller and a corresponding vehicle are also provided.

Background Art

In order to minimize the emission of harmful substances, such as nitrogen oxides, hydrocarbons and soot, vehicles driven by gasoline engines today have lambda sensors. The lambda sensor is the main sensor in the lambda control loop for catalytic exhaust gas purification. It compares the residual oxygen content in the exhaust gas with the oxygen content of a reference (usually the air in the current atmosphere) to determine the combustion air ratio lambda (the ratio of combustion air to fuel). The injection amount of fuel is adjusted according to the determined combustion air ratio combustion air ratio to achieve a balanced combustion air ratio as far as possible, that is, a stoichiometric combustion air ratio (λ=1), because deviations from the stoichiometric combustion air ratio usually increase fuel consumption and harmful gas emissions.

The implementation of a lambda controller, which corrects the fuel injection as a function of a determined combustion air ratio (measured value of the lambda sensor), is a typical component of the operating strategy of a gasoline engine. In principle, however, the controllers used (P, PI or PID) can only react to mixture deviations that have already occurred (combustion air ratio lambda ≠ 1). Therefore, in the known operating strategies, the main concern is not to prevent the occurrence of mixture deviations, but only to reduce the mixture deviations that have already occurred over time. In steady-state operation of a gasoline engine, for example at a constant speed and constant intake pressure, the lambda controller can compensate for the mixture deviations that occur very well. In dynamic operating conditions, i.e. when the speed and intake pressure are constantly changing, the mixture deviations change repeatedly, and the lambda regulators of the known type cannot reduce such deviations, or do not reduce them to an adequate degree.

Document DE 195 47 496 A1 discloses a method and a device for accurately determining the mass of air drawn into the cylinders of an internal combustion engine as a basis for measuring the fuel mass by means of a learnable observer.

Document DE 10 2020 116 488 B3 relates to a method for operating an internal combustion engine, in particular a method for controlling an air-fuel ratio using a neural network and a corresponding controller.

Document DE 197 06 750 A1 discloses a method for controlling the mixture of an internal combustion engine and a device for carrying out the method.

Document DE 10 2009 032 064 B3 relates to an internal combustion engine (gasoline engine) with controlled or regulated fuel injection.

Document DE 10 2016 205 241 A1 discloses a method and a device for operating an internal combustion engine.

Document DE 696 35 429 T2 relates to an internal combustion engine, in particular an internal combustion engine which can be operated according to a predefined injection pattern having one or more pilot injections and one or more main injections.

Summary of the invention

The object of the present invention is therefore to develop an operating strategy for a gasoline engine in order to prevent or at least further reduce the occurrence of mixture deviations from a stoichiometric (or chemically equivalent) combustion air ratio in a gasoline engine.

The above-mentioned problem is solved by a method for generating an injection quantity correction model for a gasoline engine by means of an algorithm for machine learning, a method for using an injection quantity correction model, a control device and a vehicle according to the independent claims. The content of the respective dependent claims is a preferred improvement.

A first aspect relates to a method for generating an injection quantity correction model for a gasoline engine, in particular a gasoline engine of a motor vehicle.

In one method step, an injection model is provided, which maps the influence of the injected quantity of fuel into the combustion chamber of the gasoline engine, or more precisely the injection correction quantity, over the time course on the combustion air ratio (determined, for example, by a lambda sensor), i.e., the exhaust gas mixture leaving the combustion chamber. In other words, the injection model maps a dependency, i.e., when and how an increased or decreased injection quantity is reflected in the exhaust gas mixture at the lambda sensor in the lambda signal starting from the injection time. The injection model can thus be used to determine how and at what point in time the injection quantity must or should be changed in order to prevent or at least reduce the (future) mixture from deviating from the stoichiometric combustion air ratio.

In a further method step, an algorithm for machine learning is generated which is trained using a training data set. The algorithm uses the injection model to establish an injection quantity correction model which estimates a correction value for the injection quantity of fuel into the combustion chamber of the gasoline engine as a function of at least one state variable from among the rotational speed, intake manifold pressure, intake camshaft phase and exhaust camshaft phase as input variables. The correction is preferably given with respect to the injection quantity calculated by the filling model from the steady-state operation of the gasoline engine. In this case, the estimated correction of the injection quantity correction model is a correction with respect to the injection quantity calculated in the steady-state operation.

The training data set includes a plurality of state parameters related to at least one input variable and the associated combustion air ratio (in time history). In other words, based on the prepared training data set, the algorithm learns the relationship between the change of the state variable in the air path of the gasoline engine and the resulting deviation from the stoichiometric combustion air ratio at the lambda sensor. Therefore, the created injection quantity correction model uses the trained or learned relationship between the change of the state variable in the air path of the gasoline engine and the injection quantity correction, which achieves the stoichiometric combustion air ratio (λ≈1) as much as possible. Therefore, the injection quantity can be adjusted by the injection quantity correction to achieve an almost stoichiometric combustion air ratio. In other words, the deviation that occurs is understood as the result of the influence of the interference variable on the system, and the deviation must be corrected, which is similar to the active noise cancellation algorithm. Therefore, the injection quantity correction model constructed according to the present invention uses the injection model to directly prevent or at least reduce the deviation from the stoichiometric combustion air ratio, that is, actively, rather than passively or delayed as in the known lambda controller. Therefore, the injection quantity can be well corrected not only in the steady-state operation of the gasoline engine, but also in the dynamic operation, so as to further reduce harmful gas emissions and excessive fuel consumption.

It is understood that any combination of the above four state variables, such as multiple or all, can also be included as input variables in the training data. Therefore, the algorithm used for machine learning takes into account the corresponding input variables when establishing the injection quantity correction model. The more state variables are considered when establishing the injection quantity correction model, the higher the prediction quality and the higher the quality of the injection quantity correction model. As the quality of the injection quantity correction model improves, the deviation from the stoichiometric combustion air ratio is further reduced.

In a preferred embodiment, it is provided that at least one state variable is an intake manifold pressure. Although the state variables speed, intake manifold pressure, intake camshaft phase and exhaust camshaft phase are related during the operation of a gasoline engine, it has been shown that changes in the state variable of the intake manifold pressure indicate the deviation from the stoichiometric combustion air ratio to a greater extent. Therefore, the use of the state variable of the intake manifold pressure can improve the quality of the injection quantity correction model.

In another preferred embodiment, it is provided that the training data set is extended by at least a plurality of relevant state parameters including further state variables of the gasoline engine air path that are different from the at least one state variable, and the injection quantity correction model also estimates the injection quantity correction as a function of the further state variables. In other words, the algorithm is trained using more than one state variable of the gasoline engine air path, thereby further improving the quality of the established injection quantity correction model.

The other state variable is preferably one of the rotational speed, intake pipe pressure, intake camshaft phase, exhaust camshaft phase, aging condition and temperature. The aging state is preferably mapped according to the coking state of one or more valves and/or injectors and/or the degree of contamination of one or more air filters. The temperature state variable is preferably the cylinder wall temperature, the exhaust gas temperature, the engine oil temperature and/or the intake temperature. The state variables rotational speed, intake pipe pressure, intake camshaft phase, exhaust camshaft phase, exhaust gas temperature, engine oil temperature and/or intake temperature are preferably detected by measurement technology, i.e. directly detected by corresponding sensors. The state variables aging state and/or cylinder wall temperature are preferably determined according to a model, i.e. indirectly determined using a known model. The use of other or additional state variables, in particular the state variables of the air path of a gasoline engine, can further improve the quality of the injection model.

In another preferred embodiment, it is provided that the injection model is provided using an FIR filter and/or another algorithm for machine learning. The FIR filter is a filter with a finite impulse response (FIR), which is very suitable for stable digital signal processing. Through the FIR filter, the injection model can be generated relatively simply using the data of the (changing) injection quantity and the related combustion air ratio. The FIR filter is preferably trained for different operating points in terms of the air path state variables of the gasoline engine in order to better map different engine characteristics. For example, the data used by the FIR filter preferably also includes the relevant speed and/or intake pipe pressure. Here, the FIR filter is preferably trained for different operating points in terms of speed and intake pipe pressure. It has been shown that when the speed and/or intake pipe pressure change, the behavior of the injection quantity affects the combustion air ratio, so taking into account the speed and/or intake pipe pressure state variables can further improve the quality of the injection model. It is relatively complex to establish an injection model using another algorithm for machine learning, but the quality of the injection model can be further improved.

In order to establish the injection model, the further algorithm is preferably trained using a further training data set. The prepared training set includes a large number of state parameters which are relevant to the corrective interventions on the injection quantity and the associated further combustion air ratio.

The further training data set is preferably also expanded to include relevant state variables with regard to the engine speed and/or the intake manifold pressure. By taking into account the speed and/or the intake manifold pressure, the quality of the created injection model can be further improved.

In another preferred embodiment, the algorithm and/or other algorithms include artificial neural networks. As the algorithm for machine learning, preferably a single-layer or multi-layer artificial neural network, but the present invention is not limited to this. On the contrary, other related machine learning algorithms can also be used. Examples for this can also be support vector machines and dynamic neural networks (LSTMs).

In another preferred embodiment, the training data set and/or the further training data set is created by a vehicle controller. Furthermore, the algorithm and/or the further algorithm is preferably trained by the vehicle controller. Training the algorithm or both algorithms on the vehicle controller enables the algorithm to be trained based on updated training data (e.g. training data generated by the controller), thereby optimizing the quality of the created injection and/or injection quantity correction model. In particular, recent disturbance influences on the combustion air ratio can thereby be taken into account in the injection quantity correction model, in order to further reduce the occurrence of deviations from the stoichiometric combustion air ratio.

In a further preferred embodiment, a plurality of state parameters relating to at least one input variable and the associated combustion air ratio and/or a plurality of state parameters relating to the injection quantity intervention and the associated further combustion air ratio are transmitted to a control unit external to the vehicle. In other words, the corresponding measurement data recorded by the controller are transmitted to the external computer unit for processing. By creating one or more external training data sets and training one or more algorithms for machine learning, the computing power and capacity of the vehicle controller can be saved or kept small, using the greater computing power and capacity of the external computer unit.

The injection model and/or the injection quantity correction model are preferably received from a control unit outside the vehicle. The storage requirements required for the injection quantity and/or the injection quantity correction model are relatively small, so that vehicles with controllers with low computing power and/or capacity can also use the injection quantity and/or the injection quantity correction model to correct the injection quantity. Therefore, the advantages described herein can be used for a large number of vehicles. Of course, additionally or alternatively preferably, the training data set and/or another training data set and/or the trained algorithm and/or another trained algorithm are received from a control unit outside the vehicle. The remaining steps for realizing the advantages introduced here can then be taken over by the controller of the vehicle.

A further aspect relates to a further method, namely a method using the above-described injection quantity correction model, more precisely a method using the injection quantity correction model constructed by the above-described method.

In a further method step, at least one state variable is determined as an input variable, namely the speed, intake manifold pressure, intake camshaft phase and exhaust camshaft phase of the injection quantity correction model. The state variable to be determined is selected depending on one or more input variables of the injection quantity correction model. If a plurality of input variables are used when generating the injection quantity correction model, the state variables associated with the input variables are determined in a similar manner.

In a further step of the further method, a correction of the injection quantity of fuel into a combustion chamber of the gasoline engine is determined based on an output of the injection quantity correction model using at least one determined state variable as an input variable.

Furthermore, the fuel injection quantity into the combustion chamber of the gasoline engine is adjusted according to the determined injection quantity correction. The optional features and advantages described by the above (first) method can be realized similarly by the further methods and can therefore be combined with each other as desired.

Another aspect of the invention relates to a control device for a gasoline engine, which is designed to carry out the above-described (first) method and/or the above-described further method. The gasoline engine is preferably a gasoline engine for a motor vehicle. The optional features and advantages described by the above-described method can be similarly realized by the control device and can therefore be combined with each other as desired.

Another aspect relates to a vehicle with a gasoline engine, the vehicle comprising the above-mentioned controller and at least one sensor for determining at least one of the state variables rotation speed, intake pipe pressure, intake camshaft phase and exhaust camshaft phase, the state variable being used as an input variable of the injection quantity correction model generated according to the above-mentioned (first) method. It is also understood that the vehicle comprises means for adjusting the injection quantity of fuel entering the combustion chamber of the gasoline engine and means for detecting the combustion air ratio. The design of these means is well known to those skilled in the art.

In a preferred embodiment, the vehicle further comprises a further sensor for determining a further state variable of the gasoline engine air path different from the at least one state variable, the further state variable being used as a further input variable of the injection quantity correction model. The vehicle preferably comprises any combination or all of the sensors required for determining the above-mentioned state variables. The specific design of the sensors required for detecting the state variables is generally known to those skilled in the art.

The controller is preferably implemented by electrical or electronic components or assemblies (hardware) or firmware (ASIC). Additionally or alternatively, the functions of the controller are implemented when executing an appropriate program (software). The controller is preferably also implemented by a combination of hardware, firmware and/or software. For example, individual components of the controller are designed as independent integrated circuits, or are arranged on a common integrated circuit to provide a single function.

According to the individual components of the controller, it is also preferably designed as one or more processes, which are run on one or more processors in one or more electronic computing devices and are generated when executing one or more computer programs. The computing device is preferably configured to cooperate with other components (for example, at least one sensor for determining state variables) to realize the functions described here. The instructions of the computer program are preferably stored in a memory, such as a RAM element. However, the computer program can also be stored in a non-volatile storage medium, such as a CD-ROM, a flash memory, etc.

It is also clear to a person skilled in the art that the functions of a plurality of computing units (data processing devices) can be combined or combined in a single device, or that the functions of a particular data processing device can be distributed among a plurality of devices in order to implement the functions of a controller.

On the other hand, a computer program comprising instructions is provided which, when executed by a computer, such as a controller of a gasoline engine of a motor vehicle (having at least one sensor for determining the rotational speed, the intake manifold pressure, the intake camshaft phase and the exhaust camshaft phase), causes the computer to execute one or both methods according to the invention, in particular a method for generating an injection quantity correction model for a gasoline engine or a method for using an injection quantity correction model.

Further advantageous embodiments of the invention result from the further features specified in the dependent claims.

Unless otherwise stated, the different embodiments of the invention described in this application can be combined with one another in an advantageous manner.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention is further described below in the examples according to the accompanying drawings.

FIG1 shows a schematic diagram for providing an injection model;

2a to 2d are schematic diagrams showing the input and output signals of FIG. 1 over time;

FIG3 shows a schematic diagram of training an algorithm for machine learning for establishing an injection quantity correction model;

4a to 4d are schematic diagrams showing the input signal of FIG. 3 over a time course;

FIG5 shows a schematic diagram of a controller having an injection quantity correction model derived from FIG3 according to an embodiment;

6 a to 6 c show schematic representations of determined corrections of the injection quantity and the associated combustion air ratio over the course of time.

DETAILED DESCRIPTION

The method for generating the injection quantity correction model 16 for a gasoline engine is described in detail below with reference to Figures 1 to 4. Figures 5 and 6 describe a method for using the injection quantity correction model 16 according to an embodiment, which is used by a motor vehicle with a controller 18 for a gasoline engine, which has stored the generated injection quantity correction model 16 in a memory. By the method according to the invention, the deviation of the gasoline engine combustion air ratio _λ2 is reduced by the correction _Kinj of the injection quantity.

Fig. 1 shows a schematic diagram for providing an injection model 10. The injection model 10 maps the influence of an injection quantity Δinj of an intervention fuel into a combustion chamber 12 of a gasoline engine on a combustion air ratio _λ1 over a time course.

Typically, a predetermined injection quantity Inj of fuel (e.g. calculated by a filling model based on the steady-state operation of the gasoline engine) is injected into a combustion chamber 12 of a gasoline engine. For the control of catalytic exhaust gas purification, the gasoline engine comprises a lambda sensor (not shown) which determines the current deviation of the mixture leaving the combustion chamber 12 from the stoichiometric combustion air ratio (λ=1). If the determined combustion air ratio λ deviates from the stoichiometric combustion air ratio (i.e. λ≠1), either too much fuel (rich mixture, λ<1) or too little fuel (lean mixture, λ>1) is injected into the combustion chamber 12. Since the deviation from the stoichiometric combustion air ratio λ increases fuel consumption and harmful exhaust gas emissions, the lambda controller 14 adjusts the combustion air ratio λ ₁ by corresponding interventions in the injection quantity based on the detected deviation. For the establishment of the injection model 10 and the injection quantity correction model 16 described below, the lambda controller 14 is deactivated. When the established injection quantity correction model 16 is used later, the lambda controller 14 is activated.

From the input and output signal diagrams of FIGS. 2a to 2d, it can be seen that there is a basic correlation between the injection quantity Δ _inj and the combustion air ratio λ _1. FIG. 2a shows a schematic diagram of an exemplary intervention in the injection quantity Δ _inj of the fuel entering the combustion chamber 12 of the gasoline engine. An injection quantity Δ _inj test signal in the form of a pseudorandom binary sequence (PRBS) is shown as an example, which can be used to obtain a data basis for parameterizing the injection model 10. FIG. 2b shows a related schematic curve of the combustion air ratio λ _1. Initially, the injection quantity Δ _inj is not intervened, and the combustion air ratio λ ₁ is stoichiometric (λ=1). For example, the injection quantity Inj is reduced (for example, by 5%) by intervening in the injection quantity Δ _inj , and a slow reaction to the increasingly lean mixture before the saturation point is reached can be seen in the combustion air ratio λ _1. Subsequently, the injection quantity Inj is increased by intervening in the injection quantity Δ _inj , and the value of the combustion air ratio λ ₁ falls back into the rich mixture range to a minimum value. After the injection quantity is reduced again, the behavior of the combustion air ratio λ ₁ is essentially repeated.

It can be seen from the schematic curves of the speed _neng and the intake pipe pressure _pin shown in Figures 2c and 2d that the curves shown in Figures 2a and 2b are recorded in the steady-state operating mode of the gasoline engine, because the speed _neng and the intake pipe pressure _pin are constant and remain constant. Combining Figures 2a and 2b, it can be seen that even in the steady-state operating mode, the control of the combustion air ratio _λ1 stoichiometric by corresponding intervention in the injection quantity _Δinj becomes difficult due to system inertia.

In the dynamic operating mode of the gasoline engine, the speed _neng and the intake pressure _pin change dynamically, so it is more difficult to determine the exact time point of intervening the injection quantity _Δinj . Therefore, in the method according to the present invention, an injection model 10 is first provided, which can map the injection quantity _Δinj of the fuel entering the combustion chamber 12 of the gasoline engine, more precisely, the effect of the intervention on the injection quantity _Δinj , and the influence on the combustion air ratio _λ1 in the time history. For this purpose, an algorithm for machine learning is used, that is, an artificial neural network, which is trained or has been trained through a prepared training set. The prepared training set includes state parameters of multiple state variables, that is, the intervention of the injection quantity _Δinj (Figure 2a), the related combustion air ratio _λ1 (Figure 2b) and the related speed _neng (Figure 2c) and the related intake pipe pressure _pin (Figure 2d). In the dynamic operating mode, that is, at different speeds _neng and intake pipe pressure _pin values, the above state parameters are selected as training data. As a result of the training, the artificial neural network learns the correlation between the injection time point, the amount of intervention injection quantity Δ _inj required and the influence on the combustion air ratio λ ₁ , and establishes a corresponding injection model 10 reflecting this correlation. By adding the speed n _eng and the intake pipe pressure _pin , the injection model 10 is also suitable for the dynamic operating mode of the gasoline engine. However, the present invention is not limited to the use of artificial neural networks. On the contrary, other mathematical methods can also be used to generate the injection model 10, such as FIR filters.

After the injection model 10 has been provided, the next method step takes place, namely the generation of an injection quantity correction model 16 , which is explained in more detail below in conjunction with FIGS. 3 and 4 a to 4 d .

FIG3 shows a training diagram of a machine learning algorithm for establishing the injection quantity correction model 16. FIG4a to FIG4d show a schematic diagram of the input signal of the training data set prepared for training the algorithm in the time history. The prepared training data set includes a plurality of input variables, namely, the speed _neng (FIG4a), the intake manifold pressure _pin (FIG4b), the intake camshaft phase (Figure 4c), exhaust camshaft phase (Fig. 4d) and the related state parameters of the combustion air ratio _λ2 (Fig. 6a). Of course, in the simplest case, only at least one of the above input variables and the related combustion air ratio _λ2 need to be used. For the sake of clarity, the method is described for the above four input variables, but is not limited thereto.

Based on the prepared training data set, the algorithm learns the relationship between the state variable change of the input signal related to the gasoline engine air path state variable and the resulting deviation from the stoichiometric combustion air ratio λ ₁ and establishes a corresponding injection quantity correction model 16, which can be based on the input variables, namely, the speed n _eng , the intake pipe pressure _pin , the intake camshaft phase and exhaust camshaft phase The correction K _inj of the injection quantity of fuel into the combustion chamber 12 of the gasoline engine is estimated. In this case, the algorithm uses the provided injection model 10 in order to be able to give the correct injection time point of the correction injection quantity K _inj . In other words, the injection quantity correction model 16 uses a learned or learned relationship between the curve of the input variables and the correction K _inj of the injection quantity, which leads to a stoichiometric combustion air ratio stoichiometric (λ≈1) as far as possible, so that the injection quantity Inj can be adjusted as a function of the measured (exemplary) four input variables in order to achieve an almost stoichiometric combustion air ratio λ ₁ with the correction K _inj of the injection quantity.

Therefore, the injection quantity correction model 16 constructed according to the present invention realizes the prevention or at least reduction of the occurrence of the deviation from the stoichiometric combustion air ratio λ ₁ directly (i.e., actively) rather than passively or slowly (i.e., only when the deviation occurs). Since the injection quantity correction model 16 is established using the injection model 10, when the injection quantity correction model 16 is used (for example, in the controller), the injection model 10 is no longer needed or used. Therefore, the injection quantity Inj is well corrected not only in the steady-state operation of the gasoline engine, but also in the dynamic operation situation, so as to further reduce harmful emissions and excessive fuel consumption.

FIG5 shows a schematic diagram of a controller 18 according to an embodiment, with the injection quantity correction model 16 of FIG3 , which is stored in a memory (not shown) of the controller 18. The controller 18 is a controller of a motor vehicle with a gasoline engine (not shown). The motor vehicle includes sensors for determining the state variables: speed n _eng , intake manifold pressure _pin , intake camshaft phase and exhaust camshaft phase The following uses a motor vehicle as an example to introduce the method of using the injection amount correction model 16.

In a first step, state parameters of four state variables serving as input variables for injection quantity correction model 16 are determined using four sensors of the motor vehicle.

In a further method step, a correction _Kinj of the fuel injection quantity into the combustion chamber 12 of the gasoline engine is determined based on the output of an injection quantity correction model 16, which is input to the injection quantity correction model 16 in response to the determined state variable.

Finally, the injection quantity Δ _inj of the fuel into the combustion chamber 12 of the gasoline engine is adjusted as a function of the determined correction _Kinj of the injection quantity. Figures 6a to 6c show the influence of the determined correction _Kinj of the injection quantity in more detail.

6 a to 6 c show diagrams of the determined correction of the injection quantity _Kinj and the associated combustion air ratio λ ₂ , λ _sum over the course of time.

FIG. 6a shows an example of a combustion air ratio _λ2 without correction of the injection quantity _Kinj . It can be seen that the combustion air ratio _λ2 controlled by the conventional lambda regulator 14 deviates greatly from the stoichiometric combustion air ratio _λ1 locally. The correction _Kinj of an exemplary injection quantity related to the combustion air ratio _λ2 output by the injection quantity correction model 16 is shown in FIG. 6b. By taking the injection model 10 into account when training the algorithm or constructing the injection quantity correction model 16, a substantial improvement in the air combustion ratio or the mixture characteristics can be achieved. This is clearly shown by the maximum values in FIG. 6b, which are shifted in time compared to the corresponding maximum values shown in FIG. 6a. FIG. 6c shows the combustion air ratio _λsum obtained by applying the correction _Kinj of the injection quantity of FIG. 6b to the injection quantity Inj. Compared with the curve of the combustion air ratio _λ2 , the deviation from the stoichiometric combustion air ratio _λ1 is significantly reduced.

Reference numerals list

10 jet model

12 Combustion Chamber

14λ controller

16 Injection quantity correction model

18 Controller

Inj fuel injection amount

Δ _inj intervention on injection quantity

n _eng speed

p _in suction pipe pressure

λ ₁ Combustion air ratio

Correction of _Kinj injection amount

λ ₂ Combustion air ratio of uncorrected injection quantity

λ _sum Combustion air ratio after injection amount correction

Intake camshaft phase

Exhaust camshaft phase

Claims (15)

1. A method for generating an injection quantity correction model (16) for a gasoline engine, wherein the method comprises the steps of:

Providing an injection model (10) mapping the effect of an injection correction amount (Delta _inj) of fuel entering a combustion chamber (12) of a gasoline engine on the combustion air ratio (lambda ₁) in the time course, wherein the injection model (10) comprises a correlation, i.e. when and how the varying injection correction amount (Delta _inj) is reflected in the exhaust gas mixture leaving the combustion chamber from the injection time,

-Generating an algorithm for machine learning trained by means of a training dataset, wherein the algorithm, using the injection model (10), builds an injection quantity correction model (16) which is dependent on at least one state variable used as input variable, namely rotational speed (n _eng), intake pipe pressure (p _in), intake camshaft phaseAnd exhaust camshaft phaseA correction (K _inj) to the injection quantity of fuel into the gasoline engine combustion chamber (12) is estimated, and the training data set includes a plurality of state parameters in terms of at least one input variable and an associated combustion air ratio (lambda ₂).

2. The method according to claim 1, wherein the at least one state variable is an inspiratory tube pressure (p _in).

3. The method according to claim 1 or 2, wherein the training dataset is expanded by at least a plurality of relevant state parameters including a further state variable of the gasoline engine air path different from the at least one state variable, and the injection quantity correction model (16) further estimates a correction (K _inj) of the injection quantity from the further state variable.

4. A method according to claim 3, wherein the further state variables are rotational speed (n _eng), suction line pressure (p _in), intake camshaft phaseExhaust camshaft phasingOne of aging state and temperature.

5. The method according to any of the preceding claims, wherein the injection model (10) is provided with the use of FIR filters and/or further algorithms for machine learning.

6. The method according to claim 5, wherein the further algorithm is trained by a further training data set to build the injection model (10), wherein the training set comprises a plurality of state parameters and a related further combustion air ratio (λ ₁) in terms of intervention on the injection correction quantity (Δ _inj).

7. The method according to claim 6, wherein the further training data set is expanded by relevant state parameters including engine speed (n _eng) and/or suction pipe pressure (p _in).

8. A method according to any preceding claim, wherein the algorithm and/or further algorithm comprises an artificial neural network.

9. The method according to any of the preceding claims, wherein the training data set and/or the further training data set is created by a controller (18) of the vehicle and/or the algorithm and/or the further algorithm is trained by the controller (18) of the vehicle.

10. The method according to any of the preceding claims, wherein a plurality of state parameters and associated combustion air ratios (λ ₂) in terms of at least one input variable and/or a plurality of state parameters and associated further combustion air ratios (λ ₁) in terms of injection correction (Δ _inj) interventions are transmitted to a control unit external to the vehicle.

11. The method according to claim 10, wherein the injection model (10) and/or the injection quantity correction model (16) are received from a control unit external to the vehicle.

12. A method of using the injection quantity correction model (16) according to any one of the preceding claims, wherein the method comprises the steps of:

-determining at least one state variable of the injection quantity correction model (16) used as input variable, namely rotational speed (n _eng), suction pipe pressure (p _in), intake camshaft phase And exhaust camshaft phase

-Determining a correction (K _inj) of the injection quantity of fuel into the combustion chamber (12) of the gasoline engine using at least one determined state variable as an input variable based on the output of the injection quantity correction model (16), and

-Adjusting the injection correction quantity (Δ _inj) of the fuel entering the gasoline engine combustion chamber (12) according to the determined correction (K _inj) of the injection quantity.

13. A controller (18) for a gasoline engine, the controller being arranged to perform the method according to any one of claims 1 to 11 and/or to perform the method according to claim 12, wherein an injection quantity correction model (16) according to the method of any one of claims 1 to 11 is stored on the controller (18).

14. A vehicle having a gasoline engine, comprising:

-a controller (18) according to claim 13 and

-At least one sensor for determining at least one state variable, i.e. speed (n _eng), suction line pressure (p _in), intake camshaft phase, used as input variable for an injection quantity correction model (16) generated according to the method according to one of claims 1 to 11And exhaust camshaft phase

15. The vehicle of claim 14, further comprising a further sensor for determining a further state variable of the gasoline engine air path, different from the at least one state variable, for use as a further input variable of the injection quantity correction model (16).