Machine learning model for prediction of target speed of motor vehicle and / or target accelerator pedal position

Abstract

Provided is a computer-implemented method (100) which comprises identifying (102) a motor vehicle (7) from a motor vehicle fleet (11) based on a fuel consumption achieved by the motor vehicle (7) during a predefined period of time, providing (104) training data comprising multiple training data sets, wherein each of the training data sets comprises an input variable data set for a machine learning model (13) and an output variable data set for the machine learning model (13), wherein the input variable data set and the output variable data set are based on predefined measurement values (15) measured by the identified motor vehicle (7) during the predefined period of time, the predefined measurement values (15) relate to a status of the identified motor vehicle (7), and the output variable data set comprises a speed and / or an accelerator pedal position of the motor vehicle (7), and determining (106) a machine learning model (13) using the training data set, wherein the determined machine learning model (13) is configured to predict the speed of the motor vehicle (7) and / or the accelerator pedal position using the input variable data set with a predefined accuracy. A device configured to carry out the method is also provided.

Description

[0001] Machine learning model for prediction of target speed of motor vehicle and / or target accelerator pedal position

[0002] The present disclosure is directed to a computer-implemented method for determining a machine learning model configured to predict a target speed of the motor vehicle and / or a target accelerator pedal position of an accelerator pedal of the motor vehicle.

[0003] Additionally or alternatively, the present disclosure is directed to a computer-implemented method comprising the use of the machine learning model in a motor vehicle to determine or predict a target speed of the motor vehicle and / or a target accelerator pedal position of an accelerator pedal of the motor vehicle.

[0004] Additionally or alternatively, the present disclosure is directed to a data processing device, e. g., a control device, configured to carry out at least one of the aforementioned methods at least partially.

[0005] Additionally or alternatively, the present disclosure is directed to a computer program comprising instructions that, when the program is executed by a computer, cause the computer to at least partially execute at least one of the aforementioned methods.

[0006] Additionally or alternatively, the present disclosure is directed to a computer-readable medium comprising instructions that, when the instructions are executed by a computer, cause the computer to at least partially execute at least one of the aforementioned methods.

[0007] Driver assistance systems are installed with an increasing number in modern motor vehicles, particularly in automobiles and trucks.

[0008] There are driver assistance systems that are designed to automatically control the lateral and / or longitudinal guidance of a motor vehicle. This may be done in a passive way, e. g., by displaying information with respect to the lateral and / or longitudinal guidance, and / or in an active way, e. g., by controlling the acceleration and / or a driving direction of the motor vehicle.

[0009] (Advanced) Driver Assistance Systems (A)DAS are electronic, in particular mechatronic, devices in motor vehicles to support the driver in certain driving situations. The focus is often on safety aspects, but also on increasing driving comfort, fuel economy and / or driving conditions.

[0010] WO 2019 / 241307 A2 is directed to an automated cruise control system to automatically decrease an overall energy consumption, the system comprising a plurality of sensors associated with a ground vehicle that maneuvers on a roadway that is configured to detect a plurality of driving parameters associated with the ground vehicle as the ground vehicle maneuvers along a segment of the roadway, wherein the driving parameters are indicative to a driving environment that the ground vehicle is operating in, a ground vehicle control detector associated with the ground vehicle that is configured to detect a plurality of ground vehicle control inputs associated with an operation of the ground vehicle as die ground vehicle maneuvers along the segment of the roadway, wherein the ground vehicle control inputs are generated from a longitudinal operation of the ground vehicle; and an energy consumption cruise controller configured to automatically adjust the operation of the ground vehicle as the ground vehicle maneuvers along the segment of the roadway to maintain the operation of the ground vehicle within an operation threshold based on the detected driving parameters and ground vehicle control inputs, wherein the operation threshold is the operation of the ground vehicle that decreases an amount of overall energy consumption by the ground vehicle and maintains a longitudinal speed of the ground vehicle within a longitudinal speed threshold associated with the segment of the roadway.

[0011] Against the background of this prior art, one object of the present disclosure is to disclose a device and / or a method, each of which is suitable for enriching the prior art.

[0012] The object is solved by the features of the independent claims. The dependent claims each comprise optional further developments of the disclosure as their content. The object is solved by a computer-implemented method comprising identifying a motor vehicle from a motor vehicle fleet based on a fuel consumption achieved by the motor vehicle during a predefined period of time.

[0013] The method comprises providing training data comprising multiple training data sets, wherein each of the training data sets comprises an input variable data set for a machine learning model and an output variable data set for the machine learning model.

[0014] The input variable data set and the output variable data set are based on predefined measurement values measured by the identified motor vehicle during the predefined period of time.

[0015] The predefined measurement values relate to a status of the identified motor vehicle.

[0016] The output variable data set comprises a speed and / or an accelerator pedal position of the motor vehicle.

[0017] The method comprises determining a machine learning model using the training data set, wherein the determined machine learning model is configured to predict the speed of the motor vehicle and / or the accelerator pedal position using the input variable data set with a predefined accuracy.

[0018] In other words, a training method for a machine learning model is suggested. All the steps above or parts of the method contribute to the technical character of the disclosure for the following reasons.

[0019] The method is a computer-implemented method, i. e., one, more or all steps of the method may be performed at least in part by a computer or a device for data processing, optionally a control device of the motor vehicle.

[0020] Moreover, it is suggested to use a machine learning algorithm to generate a machine learning model, e. g. a statistical model, that is configured to predict or determine a target speed and / or a target accelerator pedal position based on predefined input parameters. By selecting the training data, i. e., input parameters mapped to output parameters, used for generating or training of the machine learning model from a motor vehicle with a certain fuel consumption, training data may be used that describes a state of a motor vehicle having a low (optionally overall) fuel consumption in the given time period. By using such training data the resulting machine learning model will be automatically configured to determine a target speed and an accelerator pedal position for the same or another motor vehicle based on the current state of that motor vehicle that will also lead to a low fuel consumption of that motor vehicle.

[0021] The method moreover comprises a stop condition, i. e., the method automatically stops when the prediction of the machine learning model reaches a certain accuracy, e. g. when an uncertainty of the output is below a predefined threshold or when the output of the machine learning model is within a certain range in comparison to the output parameter included in test data used to test the accuracy of the machine learning model (e. g. 85% accuracy). This condition provides at least three technical effects. More specifically, the condition may be chosen or set such that the output of the machine learning model, i. e. the target speed of the motor vehicle and / or the target accelerator pedal position, leads to a target fuel consumption of the motor vehicle using the model with a certain accuracy. Additionally, the condition may be chosen or set such that the computing needed for determining or generating the machine learning model is suitable for the computing of the data processing device carrying out the method. Moreover, the condition may be chosen or set such that an overfitting of the machine learning model may be avoided.

[0022] The training data itself also contributes to the technical character of the disclosure. Besides the fact that the training data comprises measurement data, i. e., data measured in the real word that constitutes technical data per se, choosing the training data from a motor vehicle having a certain, e. g. low, fuel consumption leads to a machine learning model that is capable of reducing the fuel consumption of other motor vehicles of a motor vehicle fleet. Therefore, the training data as such contributes to the technical character of the disclosure because it has an intended technical use, i. e. training of a machine learning model configured to be used in or in connection with motor vehicles, and the training data causes a technical effect when used according to this intended technical use, i. e. the machine learning model trained with the training data is capable of reducing a fuel consumption of a motor vehicle. Therefore, the applicant reserves the right to file a separate application and / or a separate independent claim directed to the training data itself. The subject matter of such an application or claim may be as follows: Training data set configured to be used for training / deter- mining / generating a machine learning model, wherein the training data comprises:

[0023] - multiple training data sets, wherein each of the training data sets comprises an input variable data set for the machine learning model and an output variable data set for the machine learning model,

[0024] - wherein the input variable data set and the output variable data set are based on predefined measurement values measured by a motor vehicle during a predefined period of time,

[0025] - wherein the predefined measurement values relate to a status of the identified motor vehicle, and

[0026] - wherein the output variable data set comprises a speed and / or an accelerator pedal position of the motor vehicle.

[0027] The present disclosure can be combined partly or entirely with the above defined training data set, i. e. the description of this disclosure applies mutatis mutandis to the training data set.

[0028] A machine learning model according to the disclosure may be a (optionally: regression) model that is obtained using machine learning techniques / algorithm. More specifically, during training / determining, the machine learning algorithm is optimized to find certain outputs from a dataset, here the target speed of the motor vehicle and / or a target accelerator pedal position. The output of this training process - here a computer program with specific rules and data structures - may be called a machine learning model. The machine learning model may thus be a computer program that is configured to determine output parameters from a previously unseen dataset. The method described above may be interpreted as a method for manufacturing the machine learning model.

[0029] Identifying a motor vehicle from a motor vehicle fleet based on a fuel consumption achieved by the motor vehicle during a predefined period of time may be done by determining if the (optional overall) fuel consumption of the respective motor vehicle is below a predefined threshold. The method is not limited to identifying solely one motor vehicle, but more than one motor vehicle of the motor vehicle fleet may be identified based on its respective fuel consumption. It is for example possible that all of those motor vehicles from the motor vehicle fleet are identified / selected that have a fuel consumption that is below a certain / predefined threshold. Additionally or alternatively, it is possible that only the motor vehicle(s) with the lowest fuel consumption are identified, e. g. the ten motor vehicles from the motor vehicle fleet with the lowest fuel consumption over the past year. If more than one motor vehicle is identified, the training data may comprise measurement values of all the identified motor vehicles.

[0030] Optional further developments of the above-described method are explained in detail below.

[0031] The method may comprise extracting the predefined measurement values from a storage unit of the identified motor vehicle.

[0032] In other words, the identified motor vehicle may comprise a storage unit which may for example be part of a (optional electronic) control unit of the motor vehicle. In this storage unit the measurement values measured by the motor vehicle, e. g. by using a sensor system of the motor vehicle, are stored during the predefined time period. Even after identifying the motor vehicle, the measurement values are extracted from the storage unit. This provides - inter alia - the technical effect that data, i. e. the stored measurement values, are only transmitted or extracted from the motor vehicle when needed. Therefore, a bandwidth and computing needed for carrying out the method may be minimized.

[0033] The method may comprise determining the input variable data set such that the input variable data set comprises a gradient of each one of the predefined measurement values.

[0034] Using the gradient of the measurement values provides the technical effect that the change in the state of the motor vehicle is reflected by the respective input value of the machine learning model. This may lead to a more accurate prediction of the output with respect to a low fuel consumption of the motor vehicle. The method may also comprise defining a data spread of the input variable data set in order to carry out an outlier detection and to eliminate those outliers.

[0035] The predefined measurement values for the input variable data set may comprise at least one of the following: the speed of the motor vehicle, torque of the motor vehicle, fuel mass of the motor vehicle (optionally in mm3 / str or mg / Hub), tire pressure of the motor vehicle, wheel speed of the motor vehicle, air mass of the motor vehicle, and the accelerator pedal position of the motor vehicle.

[0036] As stated above, for all of these values the gradient may be used as input of the machine learning model.

[0037] The motor vehicle speed gradient provides information about how the motor vehicle's speed is changing over time. It may help to identify acceleration or deceleration patterns, which directly influence fuel consumption. Analyzing the speed gradient allows for optimizing speed profiles and minimizing fuel-intensive rapid changes in velocity.

[0038] The torque gradient indicates changes in engine load or power demand. By monitoring the torque gradient, the machine learning model can adjust engine performance and fuel delivery, thus ensuring efficient power generation based on the required driving conditions.

[0039] The fuel gradient reflects changes in fuel consumption over time. It helps identify patterns and trends of fuel consumption, enabling the model to adjust for optimal fuel efficiency. Analyzing the fuel gradient allows for detecting inefficient fuel usage and implementing strategies to reduce consumption.

[0040] The tire pressure gradient captures variations in tire pressure, which directly affects rolling resistance and fuel efficiency. By considering the tire pressure gradient, the model can recommend optimal tire pressure levels for minimizing resistance and maximizing fuel economy.

[0041] The wheel speed gradient represents changes in the rotational speed of the vehicle's wheels. It provides insights into braking or coasting patterns, which impact fuel consumption. Analyzing the wheel speed gradient helps optimize braking strategies and reduce energy loss during deceleration.

[0042] The airmass gradient reflects changes in the amount of air entering the engine, which affects combustion efficiency and fuel consumption. By monitoring the airmass gradient, the model can adjust fuel-air mixture ratios for optimal combustion, leading to improved fuel efficiency.

[0043] Gaussian Process Regression may be used for determining the machine learning model.

[0044] More specifically, Gaussian Process Regression (GPR) allows to predict the output variables mentioned in the machine learning model especially using gradients as inputs. A typical Gaussian process regression formula used for predictions is:

[0045] Where xi, X2 .... XD are the input variables, y is an output variable; D = rate of change of output for the respective input, N is number of trained data points, Qi is combination of signal noise for each data point; lj = length scale; Xij position of training data in the input space.

[0046] After mapping of input variables and output variables as input and output for the pre- diction / machine learning model, a machine learning GPR algorithm may be used to determine the machine learning model. It is for example possible to divide the training data into training sets, validation sets and test sets. Against each predicted output value, the coefficient of determination for identifying the model accuracy may be calculated. This coefficient may be used in the manner described above to automatically stop the method when the predefined accuracy is reached. The machine learning model may be trained by inducting the larger dataset with reduced noise (outlier detection in the data, see above).

[0047] The present disclosure is also directed to a computer-implemented method that comprises measuring predefined measurement values by a motor vehicle, wherein the predefined measurement values relate to a status of the motor vehicle.

[0048] The method comprises providing an input variable data set for a machine learning model, wherein the input variable data set is based on the measured predefined measurement values.

[0049] The method comprises to input the provided input variable data set into the machine learning model, wherein the machine learning model is configured to predict an output data set comprising a target speed of the motor vehicle and / or a target accelerator pedal position using the input variable data set.

[0050] The method comprises determining the target speed of the motor vehicle and / or the target accelerator pedal position using the machine learning model based on the input variable data set input into the machine learning model.

[0051] In other words, a method comprising the use of the machine learning model trained as described above is provided.

[0052] The description given above with respect to the method for determining the machine learning model applies mutatis mutandis to the method for using the machine learning model and vice versa. The features of the two methods may be combined partly or entirely.

[0053] Additionally, the use of the machine learning model provides the technical effect that the prediction of the target speed of the motor vehicle and / or the target accelerator pedal position may be carried out using a relatively low computing power. More specifically, in comparison to traditional algorithm the machine learning model in general requires less computing power to generate the desired output. Instead, training of the machine learning model, e. g., in the case of using the Gaussian Process Regression mainly creating the kernel, is the task that is intensive with respect to a required computing power. Once the model is trained, using the model is not computation power intensive anymore. In the present context, training of the model may be done on a server, i. e. outside of the vehicle, where a more or less unlimited computing power may be available. The use or application of the already trained model may take place in the motor vehicle itself where the resources are much more limited. Therefore, the present disclosure is narrowly tailored to the needs of the respective technical field in which the disclosure may be used, i. e. motor vehicles.

[0054] The method may comprise determining the input variable data set such that the input variable data set comprises a gradient of each one of the predefined measurement values.

[0055] Optional further developments of the method described above are explained in detail below.

[0056] The predefined measurement values for the input variable data set comprise at least one of the following: a measured speed of the motor vehicle, a measured torque of the motor vehicle, a measured fuel mass of the motor vehicle, a measured tire pressure of the motor vehicle, a measured wheel speed of the motor vehicle, a measured air mass of the motor vehicle, and a measured accelerator pedal position of the motor vehicle.

[0057] The method may comprise determining a control signal based on the determined target speed of the motor vehicle and / or the determined target accelerator pedal position. The control signal may be configured to cause a human machine interface to output an information corresponding to the determined target speed of the motor vehicle and / or the determined target accelerator pedal position.

[0058] The method may comprise the output of technical data, i. e. the control signal output to the human machine interface.

[0059] The method may comprise to output the determined control signal to the human machine interface, and to output the information via the human machine interface. The human machine interface may comprise a display unit and the method may comprise to output the information via the display unit.

[0060] In other words, the output which may consist of or comprise the predicted speed and acceleration pedal position (e. g. as percentage), can be displayed to a driver, optionally in real time, on a dashboard of the motor vehicle. Providing this information, optionally at a millisecond level granularity (see also advantage of reduced computing power necessary for generating the output using the machine learning model as described above), allows the driver to be aware of the maximum and minimum demanded vehicle speed and acceleration pedal values for precise control and optimization thereof. Therefore, the model may provide real-time recommendations for the desired motor vehicle speed. The target speed can be shown to the driver on the display unit, indicating the optimum speed with respect to fuel consumption at a given moment. This helps the driver in maintaining a speed that is fuel-efficient and aligned with the model's predictions. Additionally or alternatively, the model may suggest the acceleration pedal position (percentage) for efficient fuel consumption. The driver can be provided with a recommended range and / or a specific target value for the acceleration pedal position. This allows the driver to modulate the pedal input to achieve the desired acceleration while minimizing fuel usage.

[0061] In conclusion, to output, e. g. the feature of displaying, the information produces a technical effect since it credibly assists a driver of the motor vehicle in performing a technical task, i. e. driving / controlling the technical system “motor vehicle”, by means of a continued and / or guided human-machine interaction process. More specifically, the method described above may be carried out continuously. The output information gives the driver of the motor vehicle continuous guidance on the target speed and / or target acceleration pedal position. Any change of the target speed and / or target acceleration pedal position caused by the user may change the current state of the motor vehicle as represented by the measurement values that are comprised in the input data of the machine learning model. The same is true for changes in the environment of the motor vehicle, such as changed driving conditions (e. g. due to weather and / or route planning), and / or a state of the motor vehicle itself (e. g. change in tire pressure). Since the machine learning model is trained to output the target speed and the target accelerator pedal position based on the current input data reflecting the state of the motor vehicle, the information output via the human machine interface, optionally the display (unit) thereof, continuously assists the driver of the motor vehicle to control the longitudinal guidance of the vehicle in a manner that leads to a low fuel consumption.

[0062] The above disclosure can be summarized in other words and with respect to one (out of multiple) possible and more concrete implementation of the disclosure as described below, with the following description to be interpreted as not being limiting to the disclosure:

[0063] In one implementation of the disclosure, there may be a fleet of 50 motor vehicles. Among these motor vehicles, there are 20 motor vehicles that have consistently achieved the best fuel economy. Leveraging the historical data from these 20 motor vehicles, valuable insights regarding their operating parameters such as speed gradient, torque gradient, fuel gradient, airmass gradient, wheel speed gradient, and tire pressure gradient may be extracted.

[0064] By utilizing Gaussian Process Regression, a predictive model may be developed that takes these gradients as inputs and accurately predicts the corresponding output values of speed and acceleration pedal. This model can effectively capture the relationships between the input parameters and the desired output variables, which have proven to be successful in achieving optimal fuel economy in the 20 vehicles with the best performance.

[0065] With this model in place, it can also be used in the remaining 30 vehicles in the fleet. By displaying the predicted output values of the speed and acceleration pedal on the dashboard of the motor vehicles, the drivers are enabled to align their driving behavior with the patterns observed in the successful vehicles. This approach allows the drivers, in particular the drivers of the 21st to 50th motor vehicle, to replicate the fuelefficient strategies that were previously employed by the top-performing motor vehicles. Implementing this approach is technically feasible as it takes existing data from the 20 vehicles and training the model on a larger dataset that has reduced noise and outliers such that the accuracy and reliability of the predictions can be ensured. The model can be integrated into the dashboard systems of the fleet vehicles, providing real-time guidance and feedback to the drivers.

[0066] By adopting this approach, the fleet owner can optimize the fuel consumption across the entire fleet, thus not only to the already successful 20 vehicles. This enables the owner to achieve consistent fuel efficiency improvements, reduce operating costs, and minimize the fleet's overall environmental impact. Furthermore, it promotes standardized fuel-efficient driving practices within the fleet and ensures that the best strategies are shared and implemented across all vehicles.

[0067] Further provided is a data processing device configured to carry out at least one of the methods described above.

[0068] The data processing device may comprise a processor and a storage unit wherein the method may be stored in the form of a computer program in the storage unit and may be carried out by the processor.

[0069] The data processing device may comprise a control device or control unit. The control device or control unit may be part of or constitute a driving assistance system. The control device may, for example, be an electronic control unit (ECU). The electronic control unit may form part of the vehicle’s electrical system, e. g. together with telematics control units and / or an environment sensor system. The electronic control unit can be an intelligent processor-controlled unit which can communicate with other modules of the vehicle, e. g. via a central gateway (CGW) and / or via field buses such as the CAN bus, LIN bus, MOST bus, FlexRay and / or via the automotive Ethernet.

[0070] It is conceivable that the data processing device is configured to control functions relevant to the driving behavior of the motor vehicle, such as the steering, the engine control, the power transmission, and / or the braking system. In addition, driver assistance systems, such as a parking assistant, adaptive cruise control (ACC), a lane departure warning system, a lane change assistant, a traffic sign recognition system, a light signal recognition system, a start-up assistant, a night vision assistant, and / or an intersection assistant, can be controlled by the data processing device.

[0071] What has been described above with reference to the methods applies analogously to the data processing device and vice versa.

[0072] Further provided is a motor vehicle comprising the data processing device described above.

[0073] The motor vehicle may comprise a human machine interface configured to output an information corresponding to a target speed of the motor vehicle and / or a target accelerator pedal position based on a control signal received from the data processing device.

[0074] The motor vehicle may comprise a display unit being part of the human machine interface and being configured to output the information.

[0075] The motor vehicle may be a passenger car, in particular an automobile, or a commercial vehicle, such as a truck.

[0076] The motor vehicle may be automated. The motor vehicle may be configured to provide, by means of the data processing device, longitudinal guidance and / or lateral guidance during automated driving of the motor vehicle, at least partially and / or at least temporarily.

[0077] The automated driving may be such that the driving of the motor vehicle is (largely) autonomous. The automated driving can be controlled at least partially and / or at least temporarily by the control device.

[0078] It is conceivable that the motor vehicle intervenes in the transverse guidance of the motor vehicle actively, e. g. by adjusting an actual steering wheel position, and / or passively, e. g. by displaying a turn indication, by means of a driving assistance system. Additionally or alternatively, it is conceivable that the motor vehicle intervenes in the longitudinal guidance of the motor vehicle actively, e. g. by adjusting an actual acceleration of the vehicle, and / or passively, e. g. by displaying a break indication by means of a driving assistance system

[0079] The motor vehicle may be a motor vehicle of automation level 0, i. e. the driver assumes the dynamic driving task even if support systems (e. g. ABS or ESP) are present.

[0080] The motor vehicle may be a motor vehicle of automation level 1 , i. e. have certain driver assistance systems that support the driver in vehicle operation, such as adaptive cruise control (ACC).

[0081] The motor vehicle can be a motor vehicle of automation level 2, i. e. be partially automated in such a way that functions such as automatic parking, lane keeping or lateral guidance, general longitudinal guidance, acceleration and / or braking are performed by driver assistance systems.

[0082] The motor vehicle may be a motor vehicle of automation level 3, i. e. automated on such a conditional basis that the driver does not have to monitor the system “vehicle” throughout. The motor vehicle autonomously performs functions such as triggering the turn signal, changing lanes, and / or lane keeping. The driver can attend to other pursuits, but is prompted by the system to take the lead within a warning time if needed.

[0083] The motor vehicle can be a motor vehicle of automation level 4, i. e. be so highly automated that the driving task of the vehicle is permanently taken over by the vehicle. If the driving task is no longer handled by the system, the driver may be asked to take over control.

[0084] The motor vehicle may be a motor vehicle of automation level 5, i. e. fully automated such that the driver is not required to perform the driving task. No human intervention is required except to set the destination and start the system. The motor vehicle can operate without a steering wheel or pedals. The above described with reference to the methods and the data processing device applies analogously to the motor vehicle and vice versa.

[0085] A computer program and / or a computer readable medium comprising instructions which, when the program is or the instructions are executed by a computer, cause the computer to carry out the method described above is / are provided.

[0086] More specifically, a computer program comprising instructions which, when the program is executed by a computer, cause the computer to at least partially execute the method described above is provided.

[0087] A program code of the computer program may be in any code, in particular in a code suitable for a control device of a motor vehicle.

[0088] The above described with reference to the method, the data processing device and the motor vehicle applies analogously to the computer program and vice versa.

[0089] As stated above, a computer-readable medium, in particular a computer-readable storage medium, is provided. The computer-readable medium comprises instructions which, when the instructions are executed by a computer, cause the computer to at least partially execute or perform the method described above.

[0090] That is, a computer-readable medium comprising a computer program as defined above may be provided.

[0091] The computer-readable medium may be any digital data storage device, such as a USB flash drive, a hard drive, a CD-ROM, an SD card, or an SSD card (or SSD drive / SSD hard drive).

[0092] The computer program does not necessarily have to be stored on such a computer- readable storage medium in order to be made available to the motor vehicle or the data processing device, but may also be obtained externally via the Internet or otherwise. That is, the computer-readable medium may comprise or may be a data signal comprising instructions which, when the instructions are executed by a computer, cause the computer to at least partially execute or perform the method described above.

[0093] What has been described above with reference to the method, the data processing device, the motor vehicle and the computer program also applies analogously to the computer-readable medium and vice versa.

[0094] An optional embodiment of the disclosure is described below with reference to figures 1 and 2.

[0095] Fig. 1 shows schematically a data processing system with a data processing device and a motor vehicle with a control device, and

[0096] Fig. 2 shows schematically a flow chart of a method according to the disclosure carried out by the system shown in figure 1 .

[0097] The data processing system 1 shown only schematically in figure 1 comprises a server unit 2, a storage unit 3 and an electronic control unit 4. The server unit 2 is connected to the storage unit 3 via a wireless data connection 5, e. g. the internet and / or a telecommunication connection, via or by means of which a bidirectional data exchange between these two units 2, 3 can take place. The storage unit 3 and the electronic control unit 4 are connected via a communication bus 6 via which a bidirectional data exchange between these two units 3, 4 can take place. The storage unit 3 and the electronic control unit 4 form part of a motor vehicle 7, here a truck. The motor vehicle 7 further comprises a human machine interface 8 controllable by the electronic control unit 4. The human machine interface 8 comprises a display unit 9 which may display, i. e. output, an information to a driver of the motor vehicle 7 based on a control signal received by the electronic control unit 4. The motor vehicle 7 further comprises a sensor system 10 configured to measure predefined measurement values relating to a status of the motor vehicle 7. The motor vehicle 7 forms part of a motor vehicle fleet 1 1 comprising a plurality of further motor vehicles 12 each of which comprises the same components as the motor vehicle 7 described above and each of which is connected to the server unit 2 via a wireless data connection 5 as described above.

[0098] Given the above set-up, the computer-implemented method 1000 described below with respect to figure 2 displaying a flow chart of the method 1000 may be executed by the data processing system 1 .

[0099] The method comprises two sub-methods, i. e. a (first) method 100 for determin- ing / training of a machine learning model 13 (see figure 1 ) and a (second) method 200 for using the trained machine learning model 13 in the motor vehicles 7, 12 of the motor vehicle fleet 11 , respectively.

[0100] In an initial step 101 of the first method 100 data is collected by the sensor system 10 of each of the motor vehicles 7, 12 of the motor vehicle fleet 11 during a predefined period of time (e. g. one year), respectively. The data comprises a fuel consumption 14 of the respective motor vehicle 7, 12. The data further comprises predefined measurement values 15 relating to a status of the respective motor vehicle 7, 12. The predefined measurement values include at least on of the following: a measured speed of the motor vehicle, a measured torque of the motor vehicle, a measured fuel mass of the motor vehicle, a measured tire pressure of the motor vehicle, a measured wheel speed of the motor vehicle, a measured air mass of the motor vehicle, and a measured accelerator pedal position of the motor vehicle. The sensor system 10 is controlled by the electronic control unit 4 and the data collected by the sensor system 10 is stored via the electronic control unit 4 in the storage unit 3 of the respective motor vehicle 7, 12. At the end of the predefined period of time the respective fuel consumption 14 stored in the respective storage unit 3 of the respective motor vehicle 7, 12 is given out as output to the server unit 2 via the data connection 5, i. e. the respective fuel consumption 14 is extracted from the vehicles 7, 12.

[0101] In a first step 102 of the first method 100 the motor vehicle 7 from the motor vehicle fleet 11 is identified by the server 2 based on its fuel consumption 14 that was achieved by the motor vehicle 7 during the predefined period of time because the motor vehicle 7 had the lowest fuel consumption 14 of all motor vehicles 7, 12 of the fleet 11 during the predefined period of time. In a second step 103 of the first method 100 the server unit 2 extracts the predefined measurement values 15 from the storage unit 3 of the identified motor vehicle 7 via the data connection 5.

[0102] In a third step 104 of the first method 100, the server unit 2 determines an input variable data set for the machine learning model 13 such that the input variable data set comprises a gradient of each one of the predefined measurement values 15, i. e. a gradient of the speed of the identified motor vehicle 7, a gradient of the torque of the identified motor vehicle 7, a gradient of the fuel mass of the identified motor vehicle 7, a gradient of the tire pressure of the identified motor vehicle 7, a gradient of the wheel speed of the identified motor vehicle 7, a gradient of the air mass of the identified motor vehicle 7, and a gradient of the accelerator pedal position of the identified motor vehicle 7.

[0103] In a fourth step 105 of the first method 100, the server unit 2 provides, i. e. determines, training data for the machine learning model 13. The training data comprises multiple training data sets, wherein each of the training data sets comprises an input variable data set for the machine learning model 13 (determined as explained above) and an output variable data set for the machine learning model 13. The output variable data set comprises the extracted speed and / or the extracted accelerator pedal position of the identified motor vehicle 7.

[0104] In a fifth step 106 of the first method 100, the server unit 2 determines, i. e. trains, the machine learning model 13 using the training data set and using Gaussian Process Regression. At the end of the fifth step 106 of the first method 100, the determined machine learning model 13 is configured to predict the speed of the identified motor vehicle 7 and / or the accelerator pedal position of the identified motor vehicle 7 using the input variable data set with a predefined accuracy, e. g. 85%. The determined machine learning model 13 is thus configured to predict a target speed of anyone of the motor vehicles 7, 12 and / or a target accelerator pedal position of anyone of the motor vehicles 7, 12 using the predefined measurement values in an input variable data set. Those measurement values in the input variable data set lead to a model 13 that it in turn predicts the target speed and / or target accelerator pedal position in a way that leads to a fuel consumption that is substantially equally as low as the fuel consumption of the identified motor vehicle 7.

[0105] In a sixth step 107 of the first method 100, the server unit 2 outputs the trained model 13 via the data connection 5 to the respective storage units 3 of each one of the motor vehicles 7, 12 of the fleet 11 . Thus, the trained model 13 is accessible by the respective control unit 4 of each one of the motor vehicles 7, 12 of the fleet 11 .

[0106] After the sixth step 107 of the first method 100 has ended, the second method 200 starts. The method 200 is continuously carried out by each of the motor vehicles 7, 12. The method is described below with respect to only one of the vehicles 7 but the description applies mutatis mutandis to all the other vehicles 12 of the fleet 11 .

[0107] In a first step 201 of the second method 200, the sensor system 10 measures the predefined measurement values 15 (see above) and outputs the measured values 15 to the electronic control unit 4.

[0108] In a second step 202 of the second method 200, the electronic control unit 4 provides, i. e. determines, an input variable data set for the machine learning model 13 such that the input variable data set comprises a gradient of each one of the predefined measurement values 15 (analogously to the third step 104 of the first method 100).

[0109] In a third step 203 of the second method 200, the electronic control unit 4 loads the machine learning model 13 from the storage unit 3 and inputs the determined input variable data set into the machine learning model 13.

[0110] As explained above, the machine learning model 13 is configured to predict an output data set comprising a target speed of the motor vehicle 7, 12 and / or a target accelerator pedal position using the input variable data set.

[0111] Consequently, in a fourth step 204 of the second method 200, the electronic control unit 4 determines the target speed of the motor vehicle 7, 12 and / or the target accelerator pedal position using the machine learning model 13 based on the input variable data set input into the machine learning model 13 in the third step 203 of the second method 200.

[0112] In a fifth step 205 of the second method 200, the electronic control unit 4 determines a control signal 16 based on the determined target speed of the motor vehicle 7, 12 and / or the determined target accelerator pedal position and outputs the determined control signal 16 to the display unit 9 of the human machine interface 8.

[0113] The control signal 16 is configured to cause the display unit 9 of the human machine interface 8 to output, i. e. display, an information corresponding to the determined target speed of the motor vehicle 7, 12 and / or the determined target accelerator pedal position.

[0114] Consequently, in a sixth step 206 of the second method 200, the display unit 9 displays the information corresponding to the determined target speed of the motor vehicle 7, 12 and / or the determined target accelerator pedal position. Since the second method 200 is carried out continuously, the displayed information is continuously, optionally in real time, updated.

[0115] Reference signs data processing system server unit storage unit electronic control unit wireless data connection wired data connection motor vehicle human machine interface control unit sensor system motor vehicle fleet further motor vehicles of motor vehicle fleet machine learning model fuel consumption measurement values control signal method for determining / training of machine learning model measure predefined measurement values and fuel consumption, and receive fuel consumption at server unit identify motor vehicle based on fuel consumption receive predefined measurement values from identified motor vehicle at server unit determine input variable data set for machine learning model determine training data set for machine learning model train machine learning model output trained machine learning model to all motor vehicles of fleet method for using the trained machine learning model in motor vehicles measure predefined measurement values determine input variable data set for machine learning model input the input variable data set into machine learning model 13 204 determine target speed of motor vehicle and / or target accelerator pedal position using machine learning model

[0116] 205 determine control signal for human machine interface

[0117] 206 output information relating to target speed of motor vehicle and / or target accelerator pedal position using human machine interface

[0118] 1000 computer-implemented method

Claims

Claims1 . Computer-implemented method (100), characterized in that the method (100) comprises:- identifying (102) a motor vehicle (7) from a motor vehicle fleet (1 1 ) based on a fuel consumption achieved by the motor vehicle (7) during a predefined period of time,- providing (104) training data comprising multiple training data sets, wherein each of the training data sets comprises an input variable data set for a machine learning model (13) and an output variable data set for the machine learning model (13), wherein:- the input variable data set and the output variable data set are based on predefined measurement values (15) measured by the identified motor vehicle (7) during the predefined period of time,- the predefined measurement values (15) relate to a status of the identified motor vehicle (7), and- the output variable data set comprises a speed and / or an accelerator pedal position of the motor vehicle (7), and- determining (106) a machine learning model (13) using the training data set, wherein the determined machine learning model (13) is configured to predict the speed of the motor vehicle (7) and / or the accelerator pedal position using the input variable data set with a predefined accuracy.

2. Computer-implemented method (100) according to claim 1 , characterized in that the method (100) comprises extracting the predefined measurement values (15) from a storage unit (3) of the identified motor vehicle (7).

3. Computer-implemented method (100) according to claim 1 or 2, characterized in that the method comprises determining (104) the input variable data set such that the input variable data set comprises a gradient of each one of the predefined measurement values (15).

4. Computer-implemented method (100) according to any of claims 1 to 3, characterized in that the predefined measurement values (15) for the input variable data set comprise at least one of the following:- the speed of the motor vehicle (7),- torque of the motor vehicle (7),- fuel mass of the motor vehicle (7),- tire pressure of the motor vehicle (7),- wheel speed of the motor vehicle (7),- air mass of the motor vehicle (7), and- the accelerator pedal position of the motor vehicle (7).

5. Computer-implemented method (100) according to any of claims 1 to 4, characterized in that Gaussian Process Regression is used for determining (106) the machine learning model (13).

6. Computer-implemented method (200), characterized in that the method (200) comprises:- measuring (201 ) predefined measurement values (15) by a motor vehicle (7, 12), wherein the predefined measurement values (15) relate to a status of the motor vehicle (7, 12),- providing (202) an input variable data set for a machine learning model (13), wherein the input variable data set is based on the measured predefined measurement values (15),- input (203) the provided input variable data set into the machine learning model (13), wherein the machine learning model (13) is configured to predict an output data set comprising a target speed of the motor vehicle (7, 12) and / or a target accelerator pedal position using the input variable data set, and- determine (204) the target speed of the motor vehicle (7, 12) and / or the target accelerator pedal position using the machine learning model (13) based on the input variable data set input into the machine learning model (13).

7. Computer-implemented method (200) according to claim 6, characterized in that the method (200) comprises determining (202) the input variable data set such that the input variable data set comprises a gradient of each one of the predefined measurement values (15).

8. Computer-implemented method (200) according to claim 6 or 7, characterized in that the predefined measurement values (15) for the input variable data set comprise at least one of the following:- a measured speed of the motor vehicle (7, 12),- a measured torque of the motor vehicle (7, 12),- a measured fuel mass of the motor vehicle (7, 12),- a measured tire pressure of the motor vehicle (7, 12),- a measured wheel speed of the motor vehicle (7, 12),- a measured air mass of the motor vehicle (7, 12), and- a measured accelerator pedal position of the motor vehicle (7, 12).

9. Computer-implemented method (200) according to any of claims 6 to 8, characterized in that the method (100) comprises:- determining (205) a control signal (16) based on the determined target speed of the motor vehicle (7, 12) and / or the determined target accelerator pedal position,- wherein the control signal (16) is configured to cause a human machine interface (8) to output an information corresponding to the determined target speed of the motor vehicle (7,12) and / or the determined target accelerator pedal position.

10. Computer-implemented method (200) according to claim 9, characterized in that the method (200) comprises:- output (205) the determined control signal (16) to the human machine interface (8), and- output (206) the information via the human machine interface (8).11 . Computer-implemented method (200) according to claim 10, characterized in that the human machine interface (8) comprises a display unit (9), wherein the method comprises to output (206) the information via the display unit (9).

12. A data processing device (1 ) comprising means (2 - 4) for carrying out the method (100, 200) according to any of claims 1 to 5 and / or according to any of claims 6 to 11 .

13. A motor vehicle (7, 12), characterized in that the motor vehicle (7, 12) comprises:- a data processing device (1 ) according to claim 12,- as far as being dependent on claim 10, a human machine interface (8) configured to output an information corresponding to a target speed of the motor vehicle (7, 12) and / or a target accelerator pedal position based on a control signal (16) received from the data processing device (1 ), and- as far as being dependent on claim 11 , a display unit (9) being part of the human machine interface (8) and being configured to output the information.

14. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method (100, 200) according to any of claims 1 to 5 and / or according to any of claims 6 to 11 .

15. A computer readable medium comprising instructions which, when the instructions are executed by a computer, cause the computer to carry out the method (100, 200) according to any of claims 1 to 5 and / or according to any of claims 6 to 11 .

Images