Method and device for carrying out at least one vehicle function of a vehicle, and method and device for creating a trained machine learning exhaust gas determination model

A computer-implemented method for creating a trained machine learning exhaust gas determination model to ascertain at least one combustion variable of a predefined fuel, using fuel data that are characteristic of a composition of the predefined fuel, wherein the fuel data concerning the predefined fuel include at least one analysis variable and a fractional portion associated with the analysis variable, wherein the analysis variable is characteristic of a group of compounds from the predefined fuel that have at least one predefined functionality, and wherein the fractional portion is characteristic of a quantity fraction of the compounds that are present in the particular fuel and grouped in the group characterized by the analysis variable, the method comprising:

• providing a trainable machine learning exhaust gas determination model that includes a set of trainable parameters and that generates at least one combustion variable as an output variable, based on fuel data collected for a predefined fuel or data derived therefrom as an input variable, the combustion variable being characteristic of at least one combustion property of the fuel;
• generating a training data set that includes a plurality of collected fuel data concerning training fuels as well as a plurality of measured combustion variables, with the combustion variables being characteristic of at least one combustion property of the particular training fuel, and a plurality of vehicle variables, with the vehicle variables being characteristic of at least one adjustment variable of an engine that is used for combustion of the training fuels;
• and
• training the machine learning exhaust gas determination model based on the training data set.

A device for creating a trained machine learning exhaust gas determination model to ascertain at least one combustion variable of a predefined fuel, using fuel data that are characteristic of a composition of the predefined fuel,

• wherein the fuel data concerning the predefined fuel include at least one analysis variable and a fractional portion associated with the analysis variable, the analysis variable being characteristic of a group of compounds from the predefined fuel and having at least one predefined functionality,
• wherein the fractional portion is characteristic of a quantity fraction of the compounds that are present in the particular fuel and grouped in the group characterized by the analysis variable,
• wherein the device provides a trainable machine learning exhaust gas determination model that includes a set of trainable parameters and generates at least one combustion variable as an output variable, based on fuel data collected for a predefined fuel or data derived therefrom as an input variable,
• wherein the combustion variable is characteristic of at least one combustion property of the fuel,
• wherein the device being generates a training data set that includes a plurality of collected fuel data concerning predefined training fuels as well as a plurality of combustion variables, the combustion variables being characteristic of at least one combustion property of the particular training fuel, and a plurality of vehicle variables,
• wherein the vehicle variables are characteristic of at least one adjustment variable of an engine that is used for combustion of the training fuels,
• wherein the device trains the machine learning exhaust gas determination model based on the training data set.